JP7523342B2

JP7523342B2 - Buried pipe determination device, exploration device, and buried pipe determination method

Info

Publication number: JP7523342B2
Application number: JP2020212823A
Authority: JP
Inventors: 勝綱崎; 貴志染田; 直樹木虎; 快彦岩尾; 圭一郎高橋
Original assignee: Osaka Gas Co Ltd
Current assignee: Osaka Gas Co Ltd
Priority date: 2020-12-22
Filing date: 2020-12-22
Publication date: 2024-07-26
Anticipated expiration: 2040-12-22
Also published as: JP2022099068A

Description

本発明は、地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、地中に向けて放射した探査用電磁波の反射波を処理し、走査ラインを含む垂直断面視での埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて地中における埋設管の有無を判定可能な埋設管判定装置、探査装置、及び埋設管判定方法に関する。 The present invention relates to a buried pipe determination device, an exploration device, and a buried pipe determination method that, when scanning along a scanning line set in a search range including buried pipes buried underground, processes the reflected waves of exploration electromagnetic waves emitted toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of a buried pipe underground based on the cross-sectional data image.

従来、埋設管判定装置としては、地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、地中に向けて放射した探査用電磁波の反射波を処理し、走査ラインを含む垂直断面視での埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に含まれる二次曲線画像を抽出し、当該二次曲線画像の広がりを１点に収斂し埋設管を抽出するマイグレーション処理を実行するものが知られている（特許文献１を参照）。
一方、他の埋設管判定装置として、地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、地中に向けて放射した探査用電磁波の反射波を処理し、走査ラインを含む垂直断面視での埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像を機械学習させ、ＡＩにより埋設管の特徴を抽出し、断面データ画像から埋設管を抽出する研究もなされている。 Conventionally, a buried pipe determination device is known that, when scanned along a scanning line set in a search range including a buried pipe buried underground, processes reflected waves of exploration electromagnetic waves emitted toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, extracts a quadratic curve image included in the cross-sectional data image, and performs a migration process to converge the spread of the quadratic curve image to a single point and extract the buried pipe (see Patent Document 1).
Meanwhile, research is also being conducted on another buried pipe determination device that processes the reflected waves of exploration electromagnetic waves emitted toward the ground when scanned along a scanning line set in a search range that includes buried pipes buried underground, obtains a cross-sectional data image that shows the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, uses machine learning on the cross-sectional data image, extracts the characteristics of the buried pipe using AI, and extracts the buried pipe from the cross-sectional data image.

特開平１０－１４１９０７号公報Japanese Patent Application Publication No. 10-141907

上記マイグレーション処理を実行して断面データ画像から埋設管を抽出する埋設管判定装置では、二次曲線画像がノイズが多い二次曲線を含んでいる場合、即ち、横縞等の強い信号が含まれている場合、当該横縞を誤って埋設管として抽出してしまうという問題があった。
また、二次曲線の収斂度を信号強度で判定するため、地層境界などからの強い信号の影響を大きく受けるという課題もあった。
一方、断面データ画像を機械学習し埋設管を抽出する技術では、学習のための教師データとしての断面データ画像が非常に多く必要となると共に、未学習の埋設管は検出が困難であるという問題があった。
また、断面データ画像では、土質、含水比、石の混入具合、舗装の種類、舗装における鉄筋の有無、近接管の影響、掘削跡の影響を考慮する必要があると共に、埋設管の口径・材質・埋設深さの影響も考慮する必要があり、断面データ画像を機械学習し埋設管を抽出する技術では、判定したときに的中率が低い傾向、即ち、埋設管の見逃し率が高い傾向があり、改善の余地があった。 In a buried pipe determination device that performs the above-mentioned migration process to extract buried pipes from cross-sectional data images, there was a problem that if the quadratic curve image contained a noisy quadratic curve, i.e., if it contained strong signals such as horizontal stripes, the horizontal stripes would be erroneously extracted as buried pipes.
In addition, because the degree of convergence of the quadratic curve is determined by the signal strength, there is also the problem that it is significantly affected by strong signals from geological boundaries, etc.
On the other hand, technology that uses machine learning to learn cross-sectional data images to extract buried pipes requires a large number of cross-sectional data images as training data for learning, and has the problem that it is difficult to detect buried pipes that have not been learned about.
In addition, when extracting cross-sectional data images, it is necessary to take into account the soil type, moisture content, amount of stone mixed in, type of pavement, presence or absence of reinforcing bars in the pavement, the influence of nearby pipes, and the influence of excavation marks, as well as the influence of the diameter, material, and burial depth of the buried pipe. Therefore, technology that uses machine learning to extract buried pipes from cross-sectional data images tends to have a low hit rate when making a judgment, i.e., a high rate of buried pipes being overlooked, leaving room for improvement.

本発明は、上述の課題に鑑みてなされたものであり、その目的は、教師データの増加を抑制しながらも、埋設管の候補を自動で抽出する際に、埋設管の見逃し率を低減し得る技術を提供する点にある。 The present invention was made in consideration of the above-mentioned problems, and its purpose is to provide a technology that can reduce the rate of overlooking buried pipes when automatically extracting candidates for buried pipes, while suppressing the increase in training data.

上記目的を達成するための埋設管判定装置は、
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定装置であって、その特徴構成は、
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出部と、
前記候補領域導出部にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の予め定められた特徴を数値化した特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出部と、
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出部とを備え、
前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域の外接矩形の高さ方向で所定の位置において幅方向に沿って輝度値を導出した、幅方向での位置のそれぞれにおいて輝度値をプロットしたときの波形における振幅と波長とピーク数との少なくとも一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出する点にある。 In order to achieve the above object, a buried pipe determination device is provided.
A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and its characteristic configuration is as follows:
a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe ;
The candidate region vector derivation unit derives brightness values along the width direction at predetermined positions in the height direction of the circumscribing rectangle of the candidate region derived by the candidate region derivation unit, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector .

上記目的を達成するための埋設管判定方法は、
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定方法であって、その特徴構成は、
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出工程と、
前記候補領域導出工程にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出工程と、
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出工程とを実行し、
前記候補領域ベクトル導出工程は、前記候補領域導出工程にて導出された前記候補領域の外接矩形の高さ方向で所定の位置において幅方向に沿って輝度値を導出した、幅方向での位置のそれぞれにおいて輝度値をプロットしたときの波形における振幅と波長とピーク数との少なくとも一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出する点にある。 The buried pipe determination method for achieving the above object is as follows:
A buried pipe determination method that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and can determine the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and the method is characterized by the following configuration:
a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe ;
The candidate region vector derivation process derives brightness values along the width direction at predetermined positions in the height direction of the circumscribing rectangle of the candidate region derived in the candidate region derivation process, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector .

上記特徴構成によれば、まずもって、候補領域導出部が、断面データ画像から埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出するから、例えば、一の二次曲線候補を含む候補領域を導出し、その後の埋設管判定処理を実行することで、断面データ画像全体を用いて埋設管判定処理を実行する場合に比べて、埋設管判定処理における判定ロジックを簡素化して結果が得られるまでの演算処理量及び演算処理時間を十分に低減することができる。 According to the above characteristic configuration, first, the candidate area derivation unit derives a candidate area including one quadratic curve candidate as a possible signal of a buried pipe from the cross-sectional data image. For example, by deriving a candidate area including one quadratic curve candidate and then executing the buried pipe determination process, it is possible to simplify the determination logic in the buried pipe determination process and sufficiently reduce the amount of calculation processing and the calculation processing time until the result is obtained, compared to when the buried pipe determination process is executed using the entire cross-sectional data image.

さて、従来の機械学習を用いた埋設管の抽出技術では、断面データ画像を直接機械学習させていたが、上記特徴構成によれば、候補領域における二次曲線候補が埋設管であることを判定するための特徴を予め定め、当該特徴に係る特徴量が候補領域においてどの程度の値であるかを算出するため、断面データ画像を直接機械学習する場合に比べて、必要とする教師データ量を低減できる。
ここで、例えば、図７の上に示すような断面データ画像の候補領域において、探査方向（矢印Ｘ方向）で略中央を通る直線Ｐに沿う深さ方向（矢印Ｚ方向）に沿って、輝度値（強度）をプロットしたグラフ（図７の下）を考える。
候補領域から得られたグラフでは、一般的に、図７の下に示すように、振幅の最大値（図７の下でＬ１）と、当該最大の振幅が得得られる波の周波数（図７の下でＬ２から算出される値）とを得ることができる。
埋設管候補に対応する候補領域の複数と、埋設管の可能性があるものに対応する候補領域の複数と、非埋設管候補に対応する候補領域の複数との夫々に関し、振幅の最大値の分布をプロットしたものを図８に示し、周波数の分布をプロットしたものを図９に示す。
図８、図９に示されるように、埋設管候補及び非埋設管候補の何れにおいても、振幅の最大値及び周波数の分布は重畳している部分が多く、これらの特徴量の単体では、埋設管候補と非埋設管候補とを分離する、即ち、埋設管候補を抽出し難いことがわかる。 In conventional buried pipe extraction techniques using machine learning, cross-sectional data images are directly machine-learned. However, with the above feature configuration, features for determining whether a quadratic curve candidate in a candidate region is a buried pipe are determined in advance, and the value of the feature quantity related to the feature in the candidate region is calculated. This makes it possible to reduce the amount of training data required compared to when cross-sectional data images are directly machine-learned.
Here, for example, in a candidate region of a cross-sectional data image as shown in the top of FIG. 7, consider a graph (bottom of FIG. 7) in which brightness values (intensity) are plotted along the depth direction (arrow Z direction) along a straight line P that passes through approximately the center in the exploration direction (arrow X direction).
In the graph obtained from the candidate area, generally, as shown at the bottom of Figure 7, it is possible to obtain the maximum amplitude (L1 at the bottom of Figure 7) and the wave frequency at which this maximum amplitude is obtained (the value calculated from L2 at the bottom of Figure 7).
FIG. 8 shows a plot of the distribution of maximum amplitudes for multiple candidate areas corresponding to potential buried pipes, multiple candidate areas corresponding to possible buried pipes, and multiple candidate areas corresponding to potential non-buried pipes, and FIG. 9 shows a plot of the distribution of frequencies for each.
As shown in Figures 8 and 9, there are many overlapping areas in the maximum amplitude and frequency distributions for both buried pipe candidates and non-buried pipe candidates, and it is clear that it is difficult to separate buried pipe candidates and non-buried pipe candidates, i.e., to extract buried pipe candidates, using these features alone.

そこで、上記特徴構成によれば、候補領域ベクトル導出部が、候補領域導出部にて導出された複数の候補領域の夫々に関し、当該候補領域における二次曲線候補が埋設管であることを判定するための複数の予め定められた特徴に関する特徴量を導出し、複数の特徴量の夫々を構成要素として有する候補領域ベクトルを導出し、埋設管候補抽出部が、導出した複数の候補領域ベクトルを、埋設管の候補となる埋設管候補ベクトルと埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、埋設管の候補となる候補領域を抽出する。
つまり、候補領域ベクトルには、複数の特徴に係る特徴量が構成要素として含まれており、それらの値の組み合わせにより、候補領域ベクトルが、埋設管候補に相当するか非埋設管候補に相当するかを分類するから、単一の特徴にて分類し難かった両者を、効果的に分類し得る。
また、埋設管候補と非埋設管候補とに分類する分類ルール群、即ち、候補領域ベクトルの構成要素の夫々に関する閾値を有する閾値ベクトルの複数の導出についてのみ、機械学習を用いて導出するから、断面データ画像を機械学習する場合に比べて、教師データの数を大幅に低減できる。
また、前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域の外接矩形の高さ方向で所定の位置において幅方向に沿って輝度値を導出した、幅方向での位置のそれぞれにおいて輝度値をプロットしたときの波形における振幅と波長とピーク数との少なくとも一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出することが好ましい。
以上より、教師データの増加を抑制しながらも、埋設管の候補を自動で抽出する際に、埋設管の見逃し率を低減し得る埋設管判定装置を実現できる。 Therefore, according to the above characteristic configuration, the candidate area vector derivation unit derives feature quantities related to multiple predetermined features for determining that a quadratic curve candidate in a candidate area is a buried pipe, for each of the multiple candidate areas derived by the candidate area derivation unit, and derives a candidate area vector having each of the multiple feature quantities as components, and the buried pipe candidate extraction unit classifies the multiple derived candidate area vectors based on a group of classification rules that classify them into buried pipe candidate vectors that are candidates for buried pipes and non-buried pipe candidate vectors that are not candidates for buried pipes, thereby extracting candidate areas that are candidates for buried pipes.
In other words, the candidate area vector contains feature quantities related to multiple features as components, and the combination of these values is used to classify the candidate area vector as either a buried pipe candidate or a non-buried pipe candidate, making it possible to effectively classify both types of vectors that would be difficult to classify using a single feature.
In addition, machine learning is used only to derive the classification rules for classifying buried pipe candidates and non-buried pipe candidates, i.e., multiple threshold vectors having thresholds for each of the components of the candidate area vector, so the amount of training data can be significantly reduced compared to when machine learning is used on cross-sectional data images.
In addition, it is preferable that the candidate region vector derivation unit derives brightness values along the width direction at predetermined positions in the height direction of the circumscribing rectangle of the candidate region derived by the candidate region derivation unit, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.
From the above, it is possible to realize a buried pipe determination device that can reduce the rate of overlooking buried pipes when automatically extracting candidates for buried pipes while suppressing an increase in training data.

埋設管判定装置の更なる特徴構成は、
前記候補領域導出部は、
前記断面データ画像上での所定の直線に沿った輝度の変化を計算し、輝度変化が大きい領域のみを抽出する形態で前記二次曲線候補に外接する矩形領域を切り出す切出工程と、当該切出工程にて切り出された前記二次曲線候補の形状に基づく値又は前記二次曲線候補の輝度値又は前記矩形領域の周囲の輝度値に基づく値を用いて、前記二次曲線候補に含まれないものを除外し、残った前記二次曲線候補に対応する前記矩形領域を含む候補領域として導出する除外工程とを実行する第１候補領域導出処理と、
前記切出工程と、前記矩形領域に含まれる前記二次曲線候補と既知の前記埋設管に対応する反射波波形との類似度が一定以上である場合に前記矩形領域を候補領域として導出するマッチング工程とを実行する第２候補領域導出処理との少なくとも何れか一方を実行するものである点にある。 Further characteristic configurations of the buried pipe determination device include:
The candidate region derivation unit
a first candidate region derivation process that executes a cutting step of calculating a change in luminance along a predetermined straight line on the cross-sectional data image and cutting out a rectangular region circumscribing the quadratic curve candidate in a form of extracting only a region with a large change in luminance, and an exclusion step of excluding a region not included in the quadratic curve candidate using a value based on the shape of the quadratic curve candidate cut out in the cutting step, or a luminance value of the quadratic curve candidate, or a value based on the luminance value of a periphery of the rectangular region, and deriving a candidate region including the rectangular region corresponding to the remaining quadratic curve candidate;
The feature of the present invention is that the method performs at least one of the cutting out process and a second candidate area derivation process, which performs a matching process of deriving the rectangular area as a candidate area if the similarity between the quadratic curve candidate included in the rectangular area and the reflected wave waveform corresponding to the known buried pipe is equal to or greater than a certain level.

上記特徴構成の如く、候補領域導出部は、断面データ画像上での所定の直線に沿った輝度の変化を計算し、輝度変化が大きい領域のみを抽出する形態で二次曲線候補に外接する矩形領域を切り出す切出工程と、当該切出工程にて切り出された二次曲線候補の形状に基づく値又は二次曲線候補の輝度値に基づく値又は前記矩形領域の周囲の輝度値を用いて、二次曲線候補に含まれないものを除外し、残った二次曲線候補に対応する矩形領域を含む候補領域として導出する除外工程とを実行する第１候補領域導出処理と、切出工程と、矩形領域に含まれる二次曲線候補と既知の埋設管に対応する反射波波形との類似度が一定以上である場合に矩形領域を候補領域として導出するマッチング工程とを実行する第２候補領域導出処理との少なくとも何れか一方を実行することで、候補領域として、明らかに埋設管に起因しない二次曲線候補を除外することができ、その後の埋設管候補導出部による埋設管候補の導出処理の効率化を図ることができる。 As described above, the candidate area derivation unit performs at least one of the following steps: a first candidate area derivation process that calculates the change in luminance along a predetermined straight line on the cross-sectional data image and cuts out a rectangular area circumscribing the quadratic curve candidate in a form that extracts only areas with large luminance changes; and an exclusion process that uses a value based on the shape of the quadratic curve candidate cut out in the cut-out process, or a value based on the luminance value of the quadratic curve candidate, or the luminance value around the rectangular area to exclude those not included in the quadratic curve candidate and derive a candidate area including a rectangular area corresponding to the remaining quadratic curve candidate; and a second candidate area derivation process that performs a cut-out process and a matching process that derives a rectangular area as a candidate area if the similarity between the quadratic curve candidate included in the rectangular area and the reflected wave waveform corresponding to a known buried pipe is equal to or greater than a certain level. This makes it possible to exclude quadratic curve candidates that are clearly not due to buried pipes as candidate areas, and to improve the efficiency of the subsequent process of deriving buried pipe candidates by the buried pipe candidate derivation unit.

さて、発明者らは、候補領域ベクトルの構成要素の一つとして採用することで、埋設管候補と非埋設管候補とを分類するのに有効な特徴として、以下の特徴を見出した。 Now, the inventors have discovered that the following features are effective for classifying buried pipe candidates and non-buried pipe candidates when used as components of a candidate region vector.

例えば、前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域の前記断面データ画像内における位置に基づいて導出される位置情報を前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出することが好ましい。 For example, it is preferable that the candidate region vector derivation unit defines position information derived based on the position of the candidate region in the cross-sectional data image derived by the candidate region derivation unit as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.

また、例えば、前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域の外接矩形の幅及び高さ及びそれらに関連する値、又は前記候補領域に含まれる前記二次曲線候補の二次係数及び一次係数及び二次係数と一次係数との比の少なくとも一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出することが好ましい。 Furthermore, for example, it is preferable that the candidate region vector derivation unit determines, as one of the features, the width and height of the circumscribing rectangle of the candidate region derived by the candidate region derivation unit and values related thereto, or the quadratic coefficient, the linear coefficient, and the ratio between the quadratic coefficient and the linear coefficient of the quadratic curve candidate included in the candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.

また、例えば、前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域を、所定の大きさに規格化した規格化領域における複数のピクセルの輝度値の夫々を前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出することが好ましい。 Furthermore, for example, it is preferable that the candidate region vector derivation unit defines each of the luminance values of a plurality of pixels in a normalized region that is normalized to a predetermined size of the candidate region derived by the candidate region derivation unit as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.

上記目的を達成するための埋設管判定装置は、
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定装置であって、その特徴構成は、
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出部と、
前記候補領域導出部にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の予め定められた特徴を数値化した特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出部と、
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出部とを備え、
前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域を対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域の数、又は前記候補領域導出部にて導出された前記候補領域を前記対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域と前記対象候補領域との位置関係に基づく値の何れか一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出する点にある。
上記目的を達成するための埋設管判定方法は、
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定方法であって、その特徴構成は、
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出工程と、
前記候補領域導出工程にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出工程と、
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出工程とを実行し、
前記候補領域ベクトル導出工程は、前記候補領域導出工程にて導出された前記候補領域を対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域の数、又は前記候補領域導出工程にて導出された前記候補領域を前記対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域と前記対象候補領域との位置関係に基づく値の何れか一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出する点にある。
上述の如く、前記候補領域ベクトル導出部は、前記候補領域導出部にて導出された前記候補領域を対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域の数、又は前記候補領域導出部にて導出された前記候補領域を前記対象候補領域としたときに、前記断面データ画像のうちで前記対象候補領域の周囲の所定範囲内に存在する前記候補領域と前記対象候補領域との位置関係に基づく値の何れか一つを前記特徴の一つとし、当該特徴を数値化した特徴量を前記候補領域ベクトルの前記構成要素の一つとして導出することが好ましい。 In order to achieve the above object, a buried pipe determination device is provided.
A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and its characteristic configuration is as follows:
a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe;
The candidate region vector derivation unit determines, as one of the features, the number of candidate regions that exist within a specified range around the target candidate region in the cross-sectional data image when the candidate region derived by the candidate region derivation unit is set as the target candidate region, or a value based on the positional relationship between the candidate region that exists within a specified range around the target candidate region in the cross-sectional data image and the target candidate region when the candidate region derived by the candidate region derivation unit is set as the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.
The buried pipe determination method for achieving the above object is as follows:
A buried pipe determination method that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and can determine the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and the method is characterized by the following configuration:
a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe;
The candidate region vector derivation process determines, as one of the features, the number of candidate regions that exist within a predetermined range around the target candidate region in the cross-sectional data image when the candidate region derived in the candidate region derivation process is set as the target candidate region, or a value based on the positional relationship between the candidate region that exists within a predetermined range around the target candidate region in the cross-sectional data image and the target candidate region when the candidate region derived in the candidate region derivation process is set as the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.
As described above, it is preferable that the candidate region vector derivation unit, when the candidate region derived by the candidate region derivation unit is set as a target candidate region, sets as one of the features the number of candidate regions that exist within a predetermined range around the target candidate region in the cross-sectional data image, or a value based on the positional relationship between the candidate region that exists within a predetermined range around the target candidate region in the cross-sectional data image and the target candidate region when the candidate region derived by the candidate region derivation unit is set as the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.

上記目的を達成するための埋設管判定装置は、In order to achieve the above object, a buried pipe determination device is provided.
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定装置であって、その特徴構成は、A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and its characteristic configuration is as follows:
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出部と、a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
前記候補領域導出部にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の予め定められた特徴を数値化した特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出部と、a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出部とを備え、a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe;
前記候補領域ベクトルに関し、ランダムな構成要素を有する閾値ベクトルを生成する閾値ベクトル生成処理と、前記埋設管であることが既知の前記候補領域ベクトルである教師データとしての埋設管ベクトルの複数及び前記埋設管でないことが既知の前記候補領域ベクトルである前記教師データとしての非埋設管ベクトルの複数を含む親集団を前記閾値ベクトルにて２つに分類したときに、一方の分類に含まれる前記候補領域ベクトルのうちの前記埋設管ベクトルの割合が、前記親集団における前記埋設管ベクトルの割合よりも多い判定割合以上である場合、前記閾値ベクトルを前記分類ルール群に加える分類ルール導出処理とを実行する分類ルール導出部を備え、a classification rule derivation unit that executes a threshold vector generation process for generating a threshold vector having random components for the candidate region vector; and a classification rule derivation process for adding the threshold vector to the classification rule group when a parent population including a plurality of buried pipe vectors as teacher data which are the candidate region vectors known to be the buried pipes and a plurality of non-buried pipe vectors as teacher data which are the candidate region vectors known not to be the buried pipes are classified into two groups using the threshold vector and a proportion of the buried pipe vectors among the candidate region vectors included in one group is equal to or greater than a determination proportion that is greater than the proportion of the buried pipe vectors in the parent population,
前記埋設管候補抽出部は、前記分類ルール導出部にて導出された複数の前記分類ルール群に基づいて、前記埋設管候補ベクトルと前記非埋設管候補ベクトルとを分類し、前記埋設管の候補となる前記候補領域を抽出する点にある。The buried pipe candidate extraction unit classifies the buried pipe candidate vectors and the non-buried pipe candidate vectors based on the plurality of classification rule groups derived by the classification rule derivation unit, and extracts the candidate areas that are candidates for the buried pipes.
上記目的を達成するための埋設管判定方法は、The buried pipe determination method for achieving the above object is as follows:
地中に埋設された埋設管を含む探索範囲において設定された走査ラインに沿って走査されたときに、前記地中に向けて放射した探査用電磁波の反射波を処理し、前記走査ラインを含む垂直断面視での前記埋設管の埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて前記地中における前記埋設管の有無を判定可能な埋設管判定方法であって、その特徴構成は、A buried pipe determination method that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves radiated toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and can determine the presence or absence of the buried pipe in the ground based on the cross-sectional data image, and the method is characterized by the following configuration:
前記断面データ画像から前記埋設管の可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出工程と、a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
前記候補領域導出工程にて導出された複数の前記候補領域の夫々に関し、当該候補領域における前記二次曲線候補が前記埋設管であることを判定するための複数の特徴量を導出し、複数の前記特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出工程と、a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
導出した複数の前記候補領域ベクトルを、前記埋設管の候補となる埋設管候補ベクトルと前記埋設管の候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、前記埋設管の候補となる前記候補領域を抽出する埋設管候補抽出工程とを実行し、a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe;
前記候補領域ベクトルに関し、ランダムな構成要素を有する閾値ベクトルを生成する閾値ベクトル生成処理と、前記埋設管であることが既知の前記候補領域ベクトルである教師データとしての埋設管ベクトルの複数及び前記埋設管でないことが既知の前記候補領域ベクトルである前記教師データとしての非埋設管ベクトルの複数を含む親集団を前記閾値ベクトルにて２つに分類したときに、一方の分類に含まれる前記候補領域ベクトルのうちの前記埋設管ベクトルの割合が、前記親集団における前記埋設管ベクトルの割合よりも多い判定割合以上である場合、前記閾値ベクトルを前記分類ルール群に加える分類ルール導出処理とを実行する分類ルール導出工程を備え、a threshold vector generation process for generating a threshold vector having random components with respect to the candidate region vector; and a classification rule derivation process for adding the threshold vector to the classification rule group when a parent population including a plurality of buried pipe vectors as teacher data which are the candidate region vectors known to be the buried pipes and a plurality of non-buried pipe vectors as teacher data which are the candidate region vectors known not to be the buried pipes are classified into two groups using the threshold vector and the proportion of buried pipe vectors among the candidate region vectors included in one group is equal to or greater than a determination proportion that is greater than the proportion of buried pipe vectors in the parent population,
前記埋設管候補抽出工程は、前記分類ルール導出工程にて導出された複数の前記分類ルール群に基づいて、前記埋設管候補ベクトルと前記非埋設管候補ベクトルとを分類し、前記埋設管の候補となる前記候補領域を抽出する点にある。The buried pipe candidate extraction process classifies the buried pipe candidate vectors and the non-buried pipe candidate vectors based on the plurality of classification rule groups derived in the classification rule derivation process, and extracts the candidate areas that are candidates for the buried pipes.

上記特徴構成によれば、分類ルール導出部は、ランダムに閾値ベクトルを生成するベクトル生成処理を実行すると共に、埋設管であることが既知の候補領域ベクトルである埋設管候補ベクトルの複数及び埋設管でないことが既知の候補領域ベクトルである非埋設管候補ベクトルの複数を含む親集団を閾値ベクトルにて２つに分類したときに、一方の分類に含まれる埋設管候補ベクトルのうちの埋設管ベクトルの割合が、親集団における埋設管ベクトルの割合よりも多い判定割合以上である場合、閾値ベクトルを分類ルール群に加える分類ルール導出処理とを実行するので、機械学習においては、ランダムに閾値ベクトルを生成すると共に、当該生成した閾値ベクトルが、埋設管ベクトルと非埋設管ベクトルとを優位に分類するものか否かを判断することに機械学習を採用するから、教師データとしての断面データ画像で機械学習を実行して埋設管候補を抽出する場合に比べ、必要とする教師データ（埋設管ベクトル及び非埋設管ベクトル）の数を大幅に低減できる。 According to the above characteristic configuration, the classification rule derivation unit executes a vector generation process to randomly generate a threshold vector, and when a parent population including a plurality of buried pipe candidate vectors that are candidate area vectors known to be buried pipes and a plurality of non-buried pipe candidate vectors that are candidate area vectors known not to be buried pipes is classified into two by the threshold vector, if the proportion of buried pipe vectors among the buried pipe candidate vectors included in one classification is equal to or greater than the determination proportion that is greater than the proportion of buried pipe vectors in the parent population, a classification rule derivation process is executed to add the threshold vector to the classification rule group. Therefore, in machine learning, a threshold vector is randomly generated, and machine learning is adopted to determine whether the generated threshold vector preferentially classifies buried pipe vectors and non-buried pipe vectors. Therefore, the amount of required training data (buried pipe vectors and non-buried pipe vectors) can be significantly reduced compared to the case where machine learning is performed on a cross-sectional data image as training data to extract buried pipe candidates.

また、これまで説明してきた埋設管判定装置を備えた探査装置も、これまで説明してきた作用効果を奏する探査装置として有効に働くものである。 In addition, an exploration device equipped with the buried pipe determination device described above also functions effectively as an exploration device that provides the effects described above.

本発明の実施形態に係る探査装置の概略構成図である。1 is a schematic configuration diagram of an exploration device according to an embodiment of the present invention. 本発明の実施形態に係る制御装置のブロック図である。FIG. 2 is a block diagram of a control device according to the embodiment of the present invention. 候補領域導出部による処理を説明するための図である。11 is a diagram for explaining a process performed by a candidate region derivation unit; FIG. 候補領域導出部による処理を説明するための図である。11 is a diagram for explaining a process performed by a candidate region derivation unit; FIG. 候補領域導出部にて導出された領域で埋設管候補と非埋設管候補が含まれる図の例である。13 is an example of a diagram in which buried pipe candidates and non-buried pipe candidates are included in a region derived by a candidate region derivation unit. 埋設管候補抽出部が分類ルールに基づいて埋設管候補を抽出する流れを説明するための図である。11 is a diagram for explaining a flow in which a buried pipe candidate extraction unit extracts buried pipe candidates based on classification rules. FIG. 探査データの候補領域のベクトルを構成する特徴量である反射波の振幅及び周波数を示す図である。1 is a diagram showing the amplitude and frequency of a reflected wave, which are feature quantities constituting a vector of a candidate area of the exploration data. FIG. 埋設管候補と、埋設管の可能性のあるものと、非埋設管候補とで、振幅毎の存在確率（密度）を示すグラフ図である。1 is a graph showing the existence probability (density) for each amplitude of buried pipe candidates, possible buried pipes, and non-buried pipe candidates. FIG. 埋設管候補と、埋設管の可能性のあるものと、非埋設管候補とで、周波数毎の存在確率（密度）を示すグラフ図である。1 is a graph showing the probability (density) of existence for each frequency of buried pipe candidates, possible buried pipes, and non-buried pipe candidates. FIG.

当該実施形態に係る埋設管判定装置Ｓ、それを備えた探査装置３、及び埋設管判定方法は、教師データの増加を抑制しながらも、埋設管Ｋの候補を自動で抽出する際に、特に、埋設管Ｋの見逃し率を低減し得る技術に関するものである。 The buried pipe determination device S, the exploration device 3 equipped with the buried pipe determination device, and the buried pipe determination method according to the present embodiment are technologies that can reduce the rate of overlooking buried pipes K, particularly when automatically extracting candidates for buried pipes K, while suppressing the increase in training data.

〔探査装置〕
図１及び図２に示すように、本発明の実施形態に係る探査装置３は、作業者により手押しされ、地中に埋設された埋設管Ｋを含む探索範囲ＳＡにおいて設定された走査ラインＳＬに沿って走査される。探査装置３は、走査ラインＳＬに沿って走査されたときに、地中に向けて探査用電磁波を放射し、当該探査用電磁波の反射波を処理し、走査ラインＳＬを含む垂直断面視での埋設管Ｋの埋設状況を示す断面データ画像を取得し、当該断面データ画像に基づいて地中における埋設管の有無を判定する埋設管判定装置Ｓを備える。さらに、埋設管判定装置Ｓにより、当該断面データ画像が探査装置３の上部に設置された表示部Ｓ６に表示されるように構成される。 [Exploration Device]
1 and 2, an exploration device 3 according to an embodiment of the present invention is manually pushed by an operator and scans along a scanning line SL set in a search range SA including a buried pipe K buried in the ground. The exploration device 3 includes a buried pipe determination device S that, when scanned along the scanning line SL, emits exploration electromagnetic waves into the ground, processes reflected waves of the exploration electromagnetic waves, obtains a cross-sectional data image showing the buried state of the buried pipe K in a vertical cross-sectional view including the scanning line SL, and determines whether or not the buried pipe is buried in the ground based on the cross-sectional data image. Furthermore, the buried pipe determination device S is configured to display the cross-sectional data image on a display unit S6 installed on the upper part of the exploration device 3.

より詳しくは、探査装置３は、手押し式または自走式のレーダ探査装置であり、埋設管Ｋの探査作業の対象となる所定の探索範囲ＳＡにおいて設定された走査ラインＳＬに沿って走行する。探査装置３は走行しながら、アンテナＡから探査用電磁波を地中に放射する。放射された電磁波の伝播経路に埋設管Ｋなどの埋設物が存在するとそこで反射される。この反射されて戻ってくる反射波が埋設管判定装置Ｓで処理され、目的となる埋設物（埋設管Ｋ）の存在を評価する断面データ画像が出力される。 More specifically, the exploration device 3 is a hand-pushed or self-propelled radar exploration device that travels along a scanning line SL set in a predetermined search range SA that is the target of the buried pipe K exploration work. As the exploration device 3 travels, it radiates exploration electromagnetic waves from antenna A into the ground. If a buried object such as a buried pipe K is present in the propagation path of the radiated electromagnetic waves, they are reflected by the buried pipe. The reflected waves that return are processed by the buried pipe determination device S, and a cross-sectional data image is output that evaluates the presence of the target buried object (buried pipe K).

断面データ画像は、埋設管判定装置Ｓにより適切な画像処理が施され、図１（ｂ）に示すような断面データ画像として表示部Ｓ６上に可視化される。作業者は、表示部Ｓ６に表示された断面データ画像から作業対象となる埋設管Ｋの地中位置を把握できる。 The cross-sectional data image is subjected to appropriate image processing by the buried pipe determination device S, and is visualized on the display unit S6 as a cross-sectional data image as shown in FIG. 1(b). The worker can grasp the underground position of the buried pipe K that is the subject of the work from the cross-sectional data image displayed on the display unit S6.

探索範囲ＳＡは、一般的には管轄地区の歩道や道路などの特定区画であり、この探索範囲ＳＡ内で予め決められたパターンで走査ラインＳＬが設定される。その際、この走査ラインＳＬを規定する基準マーカＭが指標として探索範囲ＳＡの地表に付与される。 The search range SA is generally a specific area such as a sidewalk or road in a jurisdiction, and within this search range SA, a scanning line SL is set in a predetermined pattern. At that time, a reference marker M that defines this scanning line SL is added to the ground surface of the search range SA as an index.

ここでいう、走査ラインＳＬを規定する基準マーカＭとは、例えば走査ラインＳＬの起点、中間点、終点などを示す文字や記号であり、チョークやペンキなどで直接地表に描画してもよいし、三角コーンなどの標識体を地面に載置してもよい。あるいは、走査ラインＳＬを示す線を描画する方法やロープを載置するような方法でも走査ラインＳＬを規定する基準マーカＭを作り出すことができる。つまり、この基準マーカＭの地表の位置により、実際の走査位置と、その走査位置での探査データとが関係付けられる。 The reference marker M that defines the scanning line SL here is, for example, a letter or symbol that indicates the start, midpoint, end point, etc. of the scanning line SL, and may be drawn directly on the ground with chalk or paint, or a sign such as a triangular cone may be placed on the ground. Alternatively, the reference marker M that defines the scanning line SL can be created by drawing a line indicating the scanning line SL or by placing a rope. In other words, the position of this reference marker M on the ground surface associates the actual scanning position with the exploration data at that scanning position.

図１の例では、基準マーカＭ（Ｍ１、Ｍ２、Ｍ３）は、３本の走査ラインＳＬ（ＳＬ１、ＳＬ２、ＳＬ３）の各起点に黒丸と操作方向を示す矢印とからなる、地面にチョークで描かれた指標である。 In the example of Figure 1, the reference markers M (M1, M2, M3) are indices drawn on the ground with chalk, consisting of black circles and arrows indicating the operation direction at the starting points of each of the three scanning lines SL (SL1, SL2, SL3).

本実施形態においては、探査装置３は、探索範囲ＳＡにおいて互いに平行で、かつ、同じ長さである複数の走査ラインＳＬを走査する。より具体的には、少なくとも３本の走査ラインＳＬ（ＳＬ１～ＳＬ３）を走査し、当該走査ラインＳＬ１～ＳＬ３それぞれについて断面データ画像を取得する。 In this embodiment, the exploration device 3 scans multiple scanning lines SL that are parallel to each other and have the same length in the search range SA. More specifically, it scans at least three scanning lines SL (SL1 to SL3) and obtains cross-sectional data images for each of the scanning lines SL1 to SL3.

走査ラインＳＬの間隔Ｌは、探索範囲ＳＡ下に存在する埋設管Ｋの状態に基づいて決定される。具体的には、例えば、５０ｃｍ間隔で設けられる。なお、埋設管Ｋの埋設状態があらかじめ分かっている場合（例えば、長距離にわたって直線に存在することが判明している場合）には、より広い間隔で設けても構わない。 The interval L of the scanning lines SL is determined based on the state of the buried pipes K that exist under the search range SA. Specifically, for example, they are set at intervals of 50 cm. Note that if the buried state of the buried pipes K is known in advance (for example, if it is known that they exist in a straight line over a long distance), the intervals may be wider.

図２に示すように、探査装置３は、大きく分けて、アンテナＡ、埋設管判定装置Ｓ、表示部Ｓ６を備える。埋設管判定装置Ｓは、アンテナＡで受信した電磁波を信号処理し、表示部Ｓ６は、埋設管判定装置Ｓで信号処理された断面データ画像を可視化した画像等を、作業者に表示する。 As shown in FIG. 2, the exploration device 3 is broadly composed of an antenna A, a buried pipe determination device S, and a display unit S6. The buried pipe determination device S performs signal processing on the electromagnetic waves received by the antenna A, and the display unit S6 displays to the operator a visualized image of the cross-sectional data image signal-processed by the buried pipe determination device S.

探査装置３のアンテナＡは、好ましくは複数のアンテナ素子から構成されると良い。探査装置３は、アンテナＡを通じてマイクロ波領域のパルス状の電磁波を地中に向けて所定の繰り返し周波数で放射するための高周波電源と送信部（いずれも図示せず）、及びアンテナＡを通じて地中から反射してきた反射波を受信する受信部（図示せず）を備える。 Antenna A of the exploration device 3 is preferably composed of multiple antenna elements. The exploration device 3 is equipped with a high-frequency power source and a transmitter (neither shown) for emitting pulsed electromagnetic waves in the microwave range through antenna A toward the ground at a predetermined repetition frequency, and a receiver (not shown) for receiving the waves reflected from the ground through antenna A.

ところで、探査装置３は、チョークで地面に描画された指標である基準マーカＭを起点として、又は基準マーカＭから所定の位置を起点として設定される走査ラインＳＬに沿って移動させられる。探査装置３には、その移動距離ないしは移動点を検出する位置検出センサユニット（図示せず）が装備されている。当該位置検出センサユニットにて検出された位置データは埋設管判定装置Ｓに送られ、地中から反射してきた反射波と合わされ、断面データ画像を作成するために利用される。位置検出センサユニットは、簡単には走行車輪に連結したロータリエンコーダによって構築することができるが、ＧＰＳやジャイロによって構築しても良い。 The exploration device 3 is moved along a scanning line SL that is set with a reference marker M, an index drawn on the ground with chalk, as the starting point, or with a specified position from the reference marker M as the starting point. The exploration device 3 is equipped with a position detection sensor unit (not shown) that detects the moving distance or moving point. The position data detected by the position detection sensor unit is sent to the buried pipe determination device S, where it is combined with the reflected waves reflected from the ground and used to create a cross-sectional data image. The position detection sensor unit can be constructed simply by a rotary encoder connected to the running wheels, but it may also be constructed by a GPS or gyro.

表示部Ｓ６は、断面データ画像を作業者が視覚的に確認できる形態で表示するフラットパネルディスプレイからなる。図１（ｂ）に示すように、フラットパネルディスプレイは、探査装置３の上面に、探査装置３を手押しする作業者から良く見えるように傾斜姿勢で設けられる。 The display unit S6 is a flat panel display that displays the cross-sectional data image in a form that can be visually confirmed by the operator. As shown in FIG. 1(b), the flat panel display is provided on the top surface of the exploration device 3 in an inclined position so that it can be clearly seen by the operator who is pushing the exploration device 3 by hand.

探査装置３は、受信部で受け取った受信信号を適宜増幅し、増幅された受信信号を、表示部Ｓ６に表示できる形態とするように埋設管判定装置Ｓによって信号処理する。
さて、当該埋設管判定装置Ｓは、教師データの増加を抑制しながらも、埋設管Ｋの候補を自動で抽出する際に、埋設管Ｋの見逃し率を低減するべく、以下の埋設管判定方法を実行する構成を有する。
即ち、当該埋設管判定装置Ｓは、断面データ画像から埋設管Ｋの可能性のあるシグナルとしての二次曲線候補を一つ含む候補領域を導出する候補領域導出処理を実行する候補領域導出部Ｓ３と、候補領域導出部Ｓ３にて導出された複数の候補領域の夫々に関し、当該候補領域における二次曲線候補が埋設管Ｋであることを判定するための複数の予め定められた特徴を数値化した特徴量を導出し、複数の特徴量の夫々を構成要素として有する候補領域ベクトルを導出する候補領域ベクトル導出工程を実行する候補領域ベクトル導出部Ｓ２と、導出した複数の候補領域ベクトルを、埋設管Ｋの候補となる埋設管候補ベクトルと埋設管Ｋの候補ではない非埋設管候補ベクトルとに分類する分類ルール群に基づいて分類して、埋設管Ｋの候補となる候補領域を抽出する埋設管候補抽出工程を実行する埋設管候補抽出部Ｓ１とを備えて構成されている。 The exploration device 3 appropriately amplifies the signal received by the receiving section, and the amplified signal is processed by the buried pipe determination device S so as to be in a form that can be displayed on the display section S6.
Now, the buried pipe determination device S has a configuration for executing the following buried pipe determination method in order to reduce the rate of overlooking buried pipes K when automatically extracting candidates for buried pipes K while suppressing an increase in training data.
That is, the buried pipe determination device S is configured to include a candidate area derivation unit S3 that executes a candidate area derivation process to derive a candidate area including one quadratic curve candidate as a possible signal of a buried pipe K from the cross-sectional data image; a candidate area vector derivation unit S2 that executes a candidate area vector derivation process to derive feature amounts that quantify multiple predetermined features for determining that the quadratic curve candidate in the candidate area is a buried pipe K for each of the multiple candidate areas derived by the candidate area derivation unit S3, and derive a candidate area vector having each of the multiple feature amounts as components; and a buried pipe candidate extraction unit S1 that executes a buried pipe candidate extraction process to classify the multiple derived candidate area vectors based on a group of classification rules into buried pipe candidate vectors that are candidates for buried pipe K and non-buried pipe candidate vectors that are not candidates for buried pipe K, and extract candidate areas that are candidates for buried pipe K.

当該候補領域導出部Ｓ３は、第１候補領域導出処理として、断面データ画像上での所定の直線に沿ったの輝度の変化を計算し、輝度変化が大きい領域のみを抽出する形態で二次曲線候補に外接する矩形領域を切り出す切出工程と、当該切出工程にて切り出された二次曲線候補の形状に基づく値又は二次曲線候補の輝度値又は矩形領域の周囲の輝度値に基づく値を用いて、二次曲線候補に含まれないものを除外し、残った二次曲線候補に対応する矩形領域を含む候補領域として導出する除外工程とを実行する。
尚、ここで、矩形領域及び候補領域には、一の二次曲線候補だけでなく、他の二次曲線が含まれていても構わない。
当該切出工程では、図３（ａ）に示す反射波画像から、図３（ｂ）に示す高輝度（シグナル振幅の大きな）の領域を検出し、当該検出した領域から図３（ｃ）に示す二次曲線に近い形状を切り出す処理を実行する。説明を追加すると、図３（ｂ）では、微分フィルタや平滑フィルタの差を利用する。その際、所定のサイズ範囲（行・列ともに５～２５ピクセル）の畳み込み行列を用いる。図３（ｃ）では、図３（ｂ）で得られる線構造上の点の集まりの中で、２次曲線候補として確率的に尤もらしい点集合を抽出する。具体的には、二次曲線でフィッティングした際に、平均の位置誤差が６ピクセル以内（より好ましくは、３ピクセル以内）となる点集合を抽出する。
更に、上記除外工程では、二次曲線候補の形状に基づく値として、二次曲線候補の外接矩形の幅や高さ（図４において破線で示す矩形の幅や高さ）、二次曲線候補の係数の少なくとも１つを用いて、これらの値と閾値とを比較的して、対象外のものを二次曲線候補（候補領域）から除外する。
一例をあげると、二次曲線の幅（二次曲線候補の外接矩形の幅）は、２００ピクセルを超えるものを除外し、二次係数の範囲が０．００２未満及び０．０２を超えるものについては除外する。 As a first candidate region derivation process, the candidate region derivation unit S3 executes an extraction step of calculating the change in luminance along a predetermined straight line on the cross-sectional data image and extracting a rectangular region circumscribing the quadratic curve candidate in a manner that extracts only regions with large luminance changes, and an exclusion step of excluding those not included in the quadratic curve candidate using a value based on the shape of the quadratic curve candidate extracted in the extraction step, or a luminance value of the quadratic curve candidate, or a value based on the luminance values around the rectangular region, and deriving a candidate region including a rectangular region corresponding to the remaining quadratic curve candidate.
It should be noted that the rectangular region and the candidate region may include not only one quadratic curve candidate but also other quadratic curves.
In the extraction process, a high-intensity (large signal amplitude) region shown in FIG. 3(b) is detected from the reflected wave image shown in FIG. 3(a), and a process of extracting a shape similar to a quadratic curve shown in FIG. 3(c) from the detected region is performed. To add to the explanation, in FIG. 3(b), the difference between a differential filter and a smoothing filter is used. In this case, a convolution matrix of a predetermined size range (5 to 25 pixels for both rows and columns) is used. In FIG. 3(c), a set of points that are probabilistically likely to be candidates for a quadratic curve is extracted from the set of points on the line structure obtained in FIG. 3(b). Specifically, a set of points that has an average position error of 6 pixels or less (more preferably, 3 pixels or less) when fitted with a quadratic curve is extracted.
Furthermore, in the above-mentioned exclusion process, at least one of the width and height of the circumscribing rectangle of the quadratic curve candidate (the width and height of the rectangle indicated by the dashed line in Figure 4) and the coefficients of the quadratic curve candidate is used as a value based on the shape of the quadratic curve candidate, and these values are compared with a threshold value to exclude non-target quadratic curve candidates (candidate area).
As an example, quadratic curves having a width (the width of a circumscribing rectangle of a quadratic curve candidate) exceeding 200 pixels are excluded, and quadratic coefficients in the range of less than 0.002 and greater than 0.02 are excluded.

図４を用いて説明すると、（ａ）は二次係数が０．０２より大きいため除外、（ｂ）は二次係数が０．００２未満且つ横幅が２００ピクセルを超えるため除外、（ｃ）と（ｄ）のみを二次曲線候補（候補領域）として残す。 To explain using Figure 4, (a) is excluded because its quadratic coefficient is greater than 0.02, (b) is excluded because its quadratic coefficient is less than 0.002 and its width exceeds 200 pixels, and only (c) and (d) are left as quadratic curve candidates (candidate regions).

次に、候補領域ベクトル導出部Ｓ２は、候補領域導出部Ｓ３にて導出された複数の二次曲線候補（候補領域）の夫々に関し、当該候補領域における二次曲線候補が埋設管であることを判定するための複数の予め定められた特徴を数値化した特徴量を導出し、複数の特徴量の夫々を構成要素として有する候補領域ベクトルを導出する。
候補領域ベクトル導出部Ｓ２が導出する候補領域ベクトルの構成要素としての特徴量としては、以下の少なくとも２つ以上を含むことが好ましい。 Next, the candidate area vector derivation unit S2 derives feature quantities that quantify multiple predetermined features for determining that the quadratic curve candidate in the candidate area is a buried pipe, for each of the multiple quadratic curve candidates (candidate areas) derived by the candidate area derivation unit S3, and derives a candidate area vector having each of the multiple feature quantities as components.
It is preferable that the feature amounts as components of the candidate region vector derived by the candidate region vector derivation unit S2 include at least two of the following:

例えば、候補領域ベクトル導出部Ｓ２は、候補領域導出部Ｓ３にて導出された候補領域の断面データ画像内における位置に基づいて導出される位置情報を特徴の一つとし、当該特徴を数値化した特徴量を候補領域ベクトルの構成要素の一つとして導出する。
具体的には、断面データ画像内における候補領域の中央位置のＸ座標Ｚ座標、当該候補領域の中央位置に関し、複数の断面データ画像の並び方向（図１（ａ）で基準マーカＭの並び方向）での座標等が挙げられる。
尚、当該特徴に関しては、候補領域のみからは抽出できないので、当該特徴を候補領域ベクトルの構成要素として用いる場合、候補領域を抽出した断面データ画像についても候補領域と併せて記憶部Ｓ５に保持しておくことになる。
また、位置情報としては、二次曲線候補の外接矩形内の領域のみならず、その周囲の領域を含む候補領域内にある位置を含むものである。
また、位置情報としては、二次曲線候補をマイグレーションした位置を含むものである。 For example, the candidate region vector derivation unit S2 regards the position information derived based on the position of the candidate region in the cross-sectional data image derived by the candidate region derivation unit S3 as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.
Specifically, examples of such coordinates include the X and Z coordinates of the center position of the candidate region in the cross-sectional data image, and the coordinates of the center position of the candidate region in the arrangement direction of the multiple cross-sectional data images (the arrangement direction of the reference marker M in Figure 1 (a)).
Furthermore, since this feature cannot be extracted from the candidate region alone, when this feature is used as a component of the candidate region vector, the cross-sectional data image from which the candidate region is extracted will also be stored in memory unit S5 together with the candidate region.
The position information includes not only the area within the circumscribing rectangle of the quadratic curve candidate, but also the position within the candidate area including the surrounding area.
The position information includes the position to which the quadratic curve candidate is migrated.

例えば、候補領域ベクトル導出部Ｓ２は、候補領域導出部Ｓ３にて導出された候補領域の外接矩形の幅及び高さ及びそれらに関連する値、又は候補領域に含まれる二次曲線候補の二次係数及び一次係数及び二次係数と一次係数との比の少なくとも一つを特徴の一つとし、当該特徴を数値化した特徴量を候補領域ベクトルの構成要素の一つとして導出する。 For example, the candidate region vector derivation unit S2 determines, as one of the features, the width and height of the circumscribing rectangle of the candidate region derived by the candidate region derivation unit S3 and their associated values, or the quadratic coefficient, the linear coefficient, and the ratio between the quadratic coefficient and the linear coefficient of the quadratic curve candidate included in the candidate region, and derives a feature quantity that quantifies the feature as one of the components of the candidate region vector.

また、例えば、候補領域ベクトル導出部Ｓ２は、候補領域導出部Ｓ３にて導出された候補領域の外接矩形の高さ方向で所定の位置において幅方向に沿って輝度値を導出した、幅方向での位置のそれぞれにおいて輝度値をプロットしたときの波形における振幅と波長とピーク数との少なくとも一つを特徴の一つとし、当該特徴を数値化した特徴量を候補領域ベクトルの構成要素の一つとして導出する。
具体的には、図７の上に示すような断面データ画像の候補領域において、深さ方向（矢印Ｚ方向）で所定の位置（例えば、中央位置）を通る直線に沿う幅方向（矢印Ｘ方向）に沿って、輝度値（強度）をプロットしたグラフにおいて、輝度値としてのシグナルの振幅・周波数・ピーク値の少なくとも一つを、特徴の一つとしても構わない。 Furthermore, for example, the candidate area vector derivation unit S2 derives brightness values along the width direction at predetermined positions in the height direction of the circumscribing rectangle of the candidate area derived by the candidate area derivation unit S3, and considers at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate area vector.
Specifically, in a candidate region of a cross-sectional data image as shown in the upper part of FIG. 7 , in a graph in which brightness values (intensity) are plotted along a width direction (arrow X direction) along a straight line passing through a predetermined position (e.g., a central position) in the depth direction (arrow Z direction), at least one of the amplitude, frequency, and peak value of the signal as the brightness value may be one of the features.

更に、例えば、候補領域ベクトル導出部Ｓ２は、候補領域導出部Ｓ３にて導出された候補領域を、所定の大きさに規格化した規格化領域における複数のピクセルの輝度値の夫々を特徴の一つとし、当該特徴を数値化した特徴量を候補領域ベクトルの構成要素の一つとして導出する。
ここで、規格化領域とは、２×２の窓枠を用いて、矩形上を走査しながら、窓枠中の最大輝度の１ピクセルで２×２と置き換えることで、画像を１／２×１／２のサイズに変換する処理を、最終的に、４×４のサイズになるまで繰り返して得られる領域である。 Furthermore, for example, the candidate area vector derivation unit S2 treats each of the luminance values of multiple pixels in a normalized area that has been normalized to a predetermined size in the candidate area derived by the candidate area derivation unit S3 as one of its features, and derives a feature amount that has been quantified as one of the components of a candidate area vector.
Here, the normalized area is an area obtained by repeatedly converting an image to 1/2 x 1/2 size by scanning a rectangle using a 2 x 2 window frame and replacing the 2 x 2 with one pixel of maximum brightness in the window frame, until the final size is 4 x 4.

また、例えば、候補領域ベクトル導出部Ｓ２は、候補領域の周囲の所定範囲に存在する二次曲線候補（候補領域）の数や、候補領域を対象候補領域としたときに断面データ画像のうちで対象候補領域の周囲の所定範囲内に存在する二次曲線候補と対象候補領域との位置関係を、特徴の一つとしても構わない。 In addition, for example, the candidate region vector derivation unit S2 may consider as one of the features the number of quadratic curve candidates (candidate regions) present within a predetermined range around the candidate region, or the positional relationship between the target candidate region and the quadratic curve candidates present within a predetermined range around the target candidate region in the cross-sectional data image when the candidate region is set as the target candidate region.

さて、このように得られた候補領域ベクトルを、埋設管Ｋに対応するか非埋設管に対応するものかを分類するべく、分類ルール導出部Ｓ４が設けられる。
当該分類ルール導出部Ｓ４は、候補領域ベクトルに関し、ランダムな構成要素を有する閾値ベクトルＺ_１～Ｚ_ｎを生成する閾値ベクトル生成処理と、埋設管Ｋであることが既知の候補領域ベクトルである教師データとしての埋設管ベクトルＸ_１～Ｘ_ｎ（ｎ＝２以上の整数）の複数及び埋設管Ｋでないことが既知の候補領域ベクトルである教師データとしての非埋設管ベクトルＹ_１～Ｙ_ｎ（ｎ＝２以上の整数）の複数を含む親集団を閾値ベクトルＺ_１～Ｚ_ｎにて２つに分類したときに、一方の分類に含まれる候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎのうちの埋設管ベクトルＸ_１～Ｘ_ｎの割合が、親集団における埋設管ベクトルＸ_１～Ｘ_ｎの割合よりも多い判定割合以上である場合、閾値ベクトルＺ_１～Ｚ_ｎを分類ルール群に加える分類ルール導出処理とを実行する。
判定割合は、土壌条件等に応じて分類ルールを導出する際に任意に決定される割合である。 A classification rule derivation unit S4 is provided to classify the candidate area vectors thus obtained as corresponding to a buried pipe K or a non-buried pipe.
The classification rule derivation unit S4 executes a threshold vector generation process for generating threshold vectors Z ₁ to Z _n having random components for the candidate region vectors, and a classification rule derivation process for adding the threshold vectors Z 1 to Z n to the classification rule group when a parent population including a plurality of buried pipe vectors X ₁ to X _n (n = an integer of 2 or more) as teacher data which are candidate region vectors known to be buried pipes K and a plurality of non-buried pipe vectors Y ₁ to Y _n (n = an integer of 2 or more) as teacher data which are candidate region vectors known not to be buried pipes K is classified into two _using the threshold vectors Z ₁ to Z _n , and when the proportion of buried pipe vectors X ₁ to X _n among the candidate region vectors X ₁ to X _n , Y ₁ to Y _{n included in one classification is equal to or greater than a determination proportion which is greater than the proportion of buried pipe vectors X 1 to X n} _in _the _parent population.
The judgment ratio is a ratio that is arbitrarily determined when deriving the classification rule depending on the soil conditions, etc.

ここで、まず、埋設管ベクトルと非埋設管ベクトルについて、図５に基づいて説明を加える。図５に示すように、埋設管Ｋであることが既知の候補領域と、埋設管Ｋでないことが既知の候補領域との夫々に対し、候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎを導出する。
候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎは、上述した複数の特徴に対応する特徴量を数値化したものを構成要素として有するものであり、例として、候補領域ベクトルＸ_１について説明すると、二次曲線候補の二次係数（Ｘ_１１）、二次曲線候補の一次係数（Ｘ_１２）、規格化領域における複数のピクセルの輝度値の一つ（Ｘ_１３）を特徴とする場合、それらを数値化した値を構要素として、候補領域ベクトルは以下の〔式１〕のようになる。
Ｘ_１＝（Ｘ_１１、Ｘ_１２、Ｘ_１３）・・・・〔式１〕 First, buried pipe vectors and non-buried pipe vectors will be explained with reference to Fig. 5. As shown in Fig. 5, candidate area vectors _X1 to _Xn and _Y1 to _Yn are derived for candidate areas known to be buried pipes K and candidate areas known not to be buried pipes K, respectively.
Candidate region vectors _X1 to _Xn , _Y1 to _Yn have, as components, quantified feature amounts corresponding to the multiple features described above. For example, in the case of candidate region vector _X1 , if the features are the quadratic coefficient of the quadratic curve candidate ( _X11 ), the linear coefficient of the quadratic curve candidate ( _X12 ), and one of the luminance values of multiple pixels in the normalized region ( _X13 ), the candidate region vector will be expressed as shown in the following [Formula 1], with these quantified values as components.
X ₁ = (X ₁₁ , X ₁₂ , X ₁₃ )... [Formula 1]

当該候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎを、埋設管Ｋであることが既知の候補領域に対して導出したものを埋設管ベクトルＸ_１～Ｘ_ｎとし、埋設管Ｋでないことが既知の候補領域に対して導出したものを非埋設管ベクトルＹ_１～Ｙ_ｎとする。 Of the candidate area vectors X ₁ to X _n and Y ₁ to Y _n , those derived for candidate areas known to be buried pipes K are called buried pipe vectors X ₁ to X _n , and those derived for candidate areas known not to be buried pipes K are called non-buried pipe vectors Y ₁ to Y _n .

分類ルール導出部Ｓ４は、さらに、上述の候補領域ベクトルに関し、ランダムな構成要素を有する閾値ベクトルＺ_１～Ｚ_ｎ（ｎ＝１以上の整数）を生成する。当該閾値ベクトルＺ_１～Ｚ_ｎは、上述した候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎに対応する構成要素数を有するものであり、当該構成要素に対応する数値がランダムに自動で生成されたものである。 The classification rule derivation unit S4 further generates threshold vectors Z ₁ to Z _n (n=an integer equal to or greater than 1) having random components for the above-mentioned candidate region vectors. The threshold vectors Z ₁ to Z _n have numbers of components corresponding to the above-mentioned candidate region vectors X ₁ to X _n and Y ₁ to Y _n , and the numerical values corresponding to the components are automatically generated randomly.

分類ルール導出部Ｓ４は、分類ルール群を自動生成するべく、図６に示すように、埋設管ベクトルＸ_１～Ｘ_ｎと非埋設管ベクトルＹ_１～Ｙ_ｎとを含む親集団を、ランダムに生成された閾値ベクトルＺ_１～Ｚ_ｎのうちの一つ（図６でルール１）にて２つに分類する。ここで、分類の仕方は、分類ルール導出部Ｓ４により、任意に設定され、例えば、埋設管ベクトルＸ_１を閾値ベクトルＺ_１にて分類する場合、（Ｘ_１＜Ｚ_１、Ｘ_２＜Ｚ_２、Ｘ_３＞Ｚ_３）の条件をすべて満たすものと、少なくとも１つ以上を満たさないものに分類する。
その後、分類ルール導出部Ｓ４は、親集団を、閾値ベクトルＺ_１～Ｚ_ｎにて２つに分類したときに、一方の分類に含まれる候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎうちの埋設管ベクトルＸ_１～Ｘ_ｎの割合（図６でα）が、親集団における埋設管ベクトルＸ_１～Ｘ_ｎの割合（図６でδ）よりも多い判定割合θ以上である場合、閾値ベクトルＺ_１～Ｚ_ｎのうちの一つ（図６でルール１）を分類ルール群に加える分類ルール導出処理を実行する。 In order to automatically generate a group of classification rules, the classification rule derivation unit S4 classifies a parent population including buried pipe vectors _X1 to _Xn and non-buried pipe vectors _Y1 to _Yn into two groups using one of the randomly generated threshold vectors _Z1 to _Zn (rule 1 in FIG. 6) as shown in FIG. 6. Here, the classification method is arbitrarily set by the classification rule derivation unit S4, and for example, when classifying a buried pipe vector _X1 using the threshold vector _Z1 , it is classified into a group that satisfies all of the conditions ( _X1 < _Z1 , _X2 < _Z2 , _X3 > _Z3 ) and a group that does not satisfy at least one of them.
Thereafter, when the parent population is classified into two groups based on the threshold vectors Z ₁ to Z _n , if the proportion of buried pipe vectors X ₁ to X _n among the candidate area vectors X 1 to X n, Y ₁ to Y _n included in one group (α in FIG. 6) is equal to or greater than a determination proportion θ that is greater than the proportion of buried pipe vectors X ₁ to X _n in the parent population (δ in FIG. 6), the classification rule derivation _unit _S4 executes a classification rule derivation process that adds one of the threshold vectors Z ₁ to Z _n (rule 1 in FIG. 6) to the classification rule group.

ここで、図６に示すように、閾値ベクトルＺ_１～Ｚ_ｎを振り分ける際の親集団は、すべての教師データを含むものであっても構わないし、他の閾値ベクトルＺ_１～Ｚ_ｎにて、分類された後の集団であっても構わない。したがって、図６で言えば、ルール２に相当する閾値ベクトルＺ_１～Ｚ_ｎについても、一方の分類に含まれる候補領域ベクトルＸ_１～Ｘ_ｎ、Ｙ_１～Ｙ_ｎうちの埋設管ベクトルＸ_１～Ｘ_ｎの割合（図６でβ又はγ）が、判定割合θよりも多い場合、当該ルール２が分類ルール群に加えられることになり、判定割合θよりも少ない場合、分類ルール群に加えられないことになる。 Here, as shown in Fig. 6, the parent group when sorting the threshold vectors Z ₁ to Z _n may include all the teacher data, or may be a group after classification by other threshold vectors Z ₁ to Z _n . Therefore, in the case of Fig. 6, for the threshold vectors Z ₁ to Z _n corresponding to rule 2, if the ratio (β or γ in Fig. 6) of buried pipe vectors X ₁ to X _n among the candidate area vectors X ₁ to X _n , Y ₁ to Y _n included in one classification is greater than the judgment ratio θ, then the rule 2 is added to the classification rule group, and if it is less than the judgment ratio θ, then the rule 2 is not added to the classification rule group.

埋設管候補抽出部Ｓ１は、分類ルール導出部Ｓ４にて導出された複数の分類ルール群に基づいて、即ち、分類ルール群に含まれる複数のルールとしての閾値ベクトルＺ_１～Ｚ_ｎにより、埋設管候補ベクトルと非埋設管候補ベクトルとを分類し、埋設管の候補となる候補領域を抽出する。 The buried pipe candidate extraction unit S1 classifies buried pipe candidate vectors and non-buried pipe candidate vectors based on the plurality of classification rule groups derived by the classification rule derivation unit S4, i.e., based on the threshold vectors Z ₁ to Z _n as the plurality of rules included in the classification rule group, and extracts candidate areas that are candidates for buried pipes.

〔別実施形態〕
（１）上記実施形態において、埋設管判定装置Ｓは、探査装置３と一体で設けられている構成例を示したが、当該埋設管判定装置Ｓの単体が、外部の探査装置３から取得された断面データ画像を、通信手段等を介して取得し、上述した候補領域導出工程と、候補領域ベクトル度導出工程と、埋設管候補抽出工程とを実行する構成を採用しても構わない。
即ち、埋設管判定装置Ｓの単体も、本願の権利範囲に含まれるものである。 [Another embodiment]
(1) In the above embodiment, a configuration example was shown in which the buried pipe determination device S is provided integrally with the exploration device 3, but a configuration may also be adopted in which the buried pipe determination device S alone acquires cross-sectional data images obtained from the external exploration device 3 via a communication means, etc., and executes the above-mentioned candidate area derivation process, candidate area vector degree derivation process, and buried pipe candidate extraction process.
In other words, the buried pipe determination device S alone is also included in the scope of the rights of this application.

（２）上記実施形態において、候補領域導出部Ｓ３は、第１候補領域導出処理を実行する例を示したが、他の処理を実行しても構わない。
例えば、候補領域導出部Ｓ３は、上記実施形態にて実行した切出工程と、矩形領域に含まれる二次曲線候補と既知の埋設管Ｋに対応する反射波波形との類似度が一定以上である場合に矩形領域を候補領域として導出するマッチング工程とを実行する第２候補領域導出処理と実行する構成を採用しても構わない。
また、候補領域導出部Ｓ３は、第１候補領域導出処理と第２候補領域導出処理との双方を実行する形態で、他候補領域を導出しても構わない。 (2) In the above embodiment, the candidate region derivation unit S3 executes the first candidate region derivation process. However, the candidate region derivation unit S3 may execute other processes.
For example, the candidate area derivation unit S3 may be configured to execute a second candidate area derivation process that performs the cutting out process executed in the above embodiment and a matching process that derives a rectangular area as a candidate area when the similarity between the quadratic curve candidate included in the rectangular area and the reflected wave waveform corresponding to a known buried pipe K is equal to or higher than a certain level.
Furthermore, the candidate region derivation unit S3 may derive other candidate regions by executing both the first candidate region derivation process and the second candidate region derivation process.

（３）上記候補領域導出部Ｓ３の第１候補領域導出処理における除外工程では、二次曲線候補の輝度値に基づく値として、切り出した二次曲線の外接矩形内あるいはその周囲も含んだ領域の輝度値、より具体的には、例えば、図７の上に示すような断面データ画像の候補領域において、探査方向（矢印Ｘ方向）で略中央を通る直線Ｐに沿う深さ方向（矢印Ｚ方向）に沿って、輝度値（強度）をプロットしたグラフ（図７下）において、輝度値がなすシグナルの振幅・周波数・シグナルの平均値などと、閾値とを比較して、所定のものを、二次曲線候補（候補領域）から除外する構成を採用しても構わない。 (3) In the exclusion step in the first candidate region derivation process of the candidate region derivation unit S3, the brightness values of the region including the circumscribing rectangle of the cut-out quadratic curve or its surroundings may be used as values based on the brightness values of the quadratic curve candidates. More specifically, in a graph (bottom of Figure 7) in which brightness values (intensity) are plotted along the depth direction (arrow Z direction) along a straight line P that passes through approximately the center in the exploration direction (arrow X direction) in the candidate region of a cross-sectional data image as shown in the top of Figure 7, the amplitude, frequency, average value of the signal formed by the brightness values, etc. may be compared with a threshold value to exclude certain ones from the quadratic curve candidates (candidate regions).

（４）上記実施形態では、分類ルール導出部Ｓ４が設けられる構成例を示したが、埋設管判定装置Ｓとは別の他の演算装置に分類ルール導出部Ｓ４を設け、当該分類ルール導出部Ｓ４にて分類ルール群を導出し、導出された分類ルール群のみを、埋設管判定装置Ｓの記憶部Ｓ５に記憶して、必要に応じて利用する構成を採用しても構わない。 (4) In the above embodiment, a configuration example in which a classification rule derivation unit S4 is provided is shown, but a configuration in which the classification rule derivation unit S4 is provided in another calculation device other than the buried pipe determination device S, the classification rule group is derived by the classification rule derivation unit S4, and only the derived classification rule group is stored in the memory unit S5 of the buried pipe determination device S and used as needed may be adopted.

尚、上記実施形態（別実施形態を含む、以下同じ）で開示される構成は、矛盾が生じない限り、他の実施形態で開示される構成と組み合わせて適用することが可能であり、また、本明細書において開示された実施形態は例示であって、本発明の実施形態はこれに限定されず、本発明の目的を逸脱しない範囲内で適宜改変することが可能である。 The configurations disclosed in the above embodiment (including other embodiments, the same applies below) can be applied in combination with configurations disclosed in other embodiments, provided no contradictions arise. Furthermore, the embodiments disclosed in this specification are merely examples, and the present invention is not limited to these embodiments. Appropriate modifications can be made without departing from the scope of the present invention.

本発明の埋設管判定装置、探査装置、及び埋設管判定方法は、教師データの増加を抑制しながらも、埋設管の候補を自動で抽出する際に、埋設管の見逃し率を低減し得る技術を提供する埋設管判定装置、探査装置、及び埋設管判定方法として、有効に利用可能である。 The buried pipe determination device, exploration device, and buried pipe determination method of the present invention can be effectively used as a buried pipe determination device, exploration device, and buried pipe determination method that provides technology that can reduce the rate of overlooking buried pipes when automatically extracting buried pipe candidates while suppressing the increase in training data.

３：探査装置
Ｋ：埋設管
Ｓ：埋設管判定装置
Ｓ１：埋設管候補抽出部
Ｓ２：候補領域ベクトル導出部
Ｓ３：候補領域導出部
Ｓ４：分類ルール導出部
Ｘ：候補領域ベクトル
Ｙ：非埋設管ベクトル
Ｚ：閾値ベクトル
θ ：判定割合 3 : exploration device K : buried pipe S : buried pipe determination device S1 : buried pipe candidate extraction unit S2 : candidate area vector derivation unit S3 : candidate area derivation unit S4 : classification rule derivation unit X : candidate area vector Y : non-buried pipe vector Z : threshold vector θ : determination ratio

Claims

A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves emitted toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image,
a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe ;
The candidate area vector derivation unit derives brightness values along the width direction at a predetermined position in the height direction of the circumscribing rectangle of the candidate area derived by the candidate area derivation unit, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate area vector .

A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves emitted toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image,
a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe;
The candidate region vector derivation unit, when the candidate region derived by the candidate region derivation unit is set as a target candidate region, determines as one of the features the number of candidate regions that exist within a specified range around the target candidate region in the cross-sectional data image, or a value based on the positional relationship between the candidate region that exists within a specified range around the target candidate region in the cross-sectional data image and the target candidate region when the candidate region derived by the candidate region derivation unit is set as the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector .

A buried pipe determination device that, when scanning along a scanning line set in a search range including a buried pipe buried in the ground, processes reflected waves of exploration electromagnetic waves emitted toward the ground, obtains a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and is capable of determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image,
a candidate region deriving unit that derives a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation unit that derives, for each of the plurality of candidate areas derived by the candidate area derivation unit, feature quantities that are numerical values of a plurality of predetermined features for determining that the quadratic curve candidate in the candidate area is the buried pipe, and derives a candidate area vector having each of the plurality of feature quantities as a component;
a buried pipe candidate extraction unit that classifies the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracts the candidate area that is a candidate for the buried pipe;
a classification rule derivation unit that executes a threshold vector generation process for generating a threshold vector having random components for the candidate region vector; and a classification rule derivation process for adding the threshold vector to the classification rule group when a parent population including a plurality of buried pipe vectors as teacher data which are the candidate region vectors known to be the buried pipes and a plurality of non-buried pipe vectors as teacher data which are the candidate region vectors known not to be the buried pipes are classified into two groups using the threshold vector and a proportion of the buried pipe vectors among the candidate region vectors included in one group is equal to or greater than a determination proportion that is greater than the proportion of the buried pipe vectors in the parent population,
The buried pipe candidate extraction unit classifies the buried pipe candidate vectors and the non-buried pipe candidate vectors based on the multiple classification rule groups derived by the classification rule derivation unit, and extracts the candidate areas that are candidates for buried pipes .

The candidate region derivation unit
a first candidate region derivation process that executes a cutting step of calculating a change in luminance along a predetermined straight line on the cross-sectional data image and cutting out a rectangular region circumscribing the quadratic curve candidate in a form of extracting only a region with a large change in luminance, and an exclusion step of excluding a region not included in the quadratic curve candidate using a value based on the shape of the quadratic curve candidate cut out in the cutting step, or a luminance value of the quadratic curve candidate, or a value based on the luminance value of a periphery of the rectangular region, and deriving a candidate region including the rectangular region corresponding to the remaining quadratic curve candidate;
A buried pipe determination device as described in any one of claims 1 to 3, which executes at least one of the cutting out process and a second candidate area derivation process which executes a matching process of deriving the rectangular area as a candidate area if the similarity between the quadratic curve candidate contained in the rectangular area and the reflected wave waveform corresponding to the known buried pipe is equal to or greater than a certain level .

The buried pipe determination device according to any one of claims 1 to 4, wherein the candidate area vector derivation unit regards position information derived based on the position of the candidate area derived by the candidate area vector derivation unit as one of the features, and derives a feature quantity that quantifies the feature as one of the components of the candidate area vector.

The buried pipe determination device of any one of claims 1 to 5, wherein the candidate area vector derivation unit determines, as one of the features, the width and height of the circumscribing rectangle of the candidate area derived by the candidate area derivation unit and values related to them, or the quadratic coefficient, the linear coefficient, and the ratio between the quadratic coefficient and the linear coefficient of the quadratic curve candidate included in the candidate area, and derives a feature amount that quantifies the feature as one of the components of the candidate area vector.

The buried pipe determination device of claim 2 or 3, wherein the candidate area vector derivation unit derives brightness values along the width direction at a predetermined position in the height direction of the circumscribing rectangle of the candidate area derived by the candidate area vector derivation unit, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate area vector.

The buried pipe determination device according to any one of claims 1 to 7, wherein the candidate area vector derivation unit defines each of the brightness values of a plurality of pixels in a normalized area, which is normalized to a predetermined size of the candidate area derived by the candidate area derivation unit, as one of the features, and derives a feature quantity, which is a numerical value of the feature, as one of the components of the candidate area vector.

4. The buried pipe determination device according to claim 1 or 3, wherein the candidate region vector derivation unit, when the candidate region derived by the candidate region derivation unit is set as a target candidate region, determines as one of the features the number of candidate regions that exist within a predetermined range around the target candidate region in the cross-sectional data image, or, when the candidate region derived by the candidate region derivation unit is set as the target candidate region, a value based on a positional relationship between the candidate region that exists within a predetermined range around the target candidate region in the cross-sectional data image and the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector.

An exploration device comprising the buried pipe determination device according to any one of claims 1 to 9 .

A buried pipe determination method capable of processing reflected waves of exploration electromagnetic waves radiated toward the ground when scanning along a scanning line set in a search range including a buried pipe buried in the ground, obtaining a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, comprising:
a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe ;
The candidate area vector derivation process derives brightness values along the width direction at predetermined positions in the height direction of the circumscribing rectangle of the candidate area derived in the candidate area derivation process, and determines at least one of the amplitude, wavelength, and number of peaks in the waveform when the brightness values are plotted at each position in the width direction as one of the features, and derives a feature amount that quantifies the feature as one of the components of the candidate area vector .

A buried pipe determination method capable of processing reflected waves of exploration electromagnetic waves radiated toward the ground when scanning along a scanning line set in a search range including a buried pipe buried in the ground, obtaining a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, comprising:
a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe;
The candidate region vector derivation process determines, as one of the features, the number of candidate regions that exist within a predetermined range around the target candidate region in the cross-sectional data image when the candidate region derived in the candidate region derivation process is set as a target candidate region, or a value based on the positional relationship between the candidate region that exists within a predetermined range around the target candidate region in the cross-sectional data image and the target candidate region when the candidate region derived in the candidate region derivation process is set as the target candidate region, and derives a feature amount that quantifies the feature as one of the components of the candidate region vector .

A buried pipe determination method capable of processing reflected waves of exploration electromagnetic waves radiated toward the ground when scanning along a scanning line set in a search range including a buried pipe buried in the ground, obtaining a cross-sectional data image showing the buried state of the buried pipe in a vertical cross-sectional view including the scanning line, and determining the presence or absence of the buried pipe in the ground based on the cross-sectional data image, comprising:
a candidate region derivation step of deriving a candidate region including one quadratic curve candidate as a possible signal of the buried pipe from the cross-sectional data image;
a candidate area vector derivation step of deriving a plurality of feature amounts for determining that the quadratic curve candidate in each of the plurality of candidate areas derived in the candidate area derivation step is the buried pipe, and deriving a candidate area vector having each of the plurality of feature amounts as a component;
a buried pipe candidate extraction step of classifying the derived plurality of candidate area vectors into buried pipe candidate vectors that are candidates for the buried pipe and non-buried pipe candidate vectors that are not candidates for the buried pipe based on a group of classification rules, and extracting the candidate area that is a candidate for the buried pipe;
a threshold vector generation process for generating a threshold vector having random components with respect to the candidate region vector; and a classification rule derivation process for adding the threshold vector to the classification rule group when a parent population including a plurality of buried pipe vectors as teacher data which are the candidate region vectors known to be the buried pipes and a plurality of non-buried pipe vectors as teacher data which are the candidate region vectors known not to be the buried pipes are classified into two groups using the threshold vector and the proportion of buried pipe vectors among the candidate region vectors included in one group is equal to or greater than a determination proportion that is greater than the proportion of buried pipe vectors in the parent population,
The buried pipe candidate extraction process is a buried pipe determination method in which the buried pipe candidate vectors and the non-buried pipe candidate vectors are classified based on a group of multiple classification rules derived in the classification rule derivation process, and the candidate areas that are candidates for buried pipes are extracted .