JP2788431B2 - Underground buried object type identification device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an underground object type discriminating apparatus capable of stably and accurately specifying the type of material of an underground object by utilizing an elastic wave inherent to a substance. .

[0002]

2. Description of the Related Art Conventionally, methods for detecting underground objects include:
Methods using electromagnetic waves, ultrasonic waves, and the like have been adopted. These methods transmit electromagnetic waves and the like from the ground surface to the ground and detect reflected waves or refracted waves at the underground objects, and determine the position and size of the underground objects. Information can be obtained. However, these methods cannot obtain information such as a material for specifying the type of the underground object.

[0003] It is also possible to detect underground objects using a metal detector using magnetism or the like,
In this case, the underground objects that can be detected are only those made of metal, and those that are made of, for example, plastic or glass cannot be detected.

Further, as a conventional technique for specifying the type of an object to be buried, a probe penetrates into the ground and the operator himself / herself determines the feeling when the penetrated probe is struck and the degree of slippage of the probe. There is a way to identify. However, in this case,
Since the identification accuracy is determined by the operator's subjectivity, it is difficult to stably identify the type of the buried object. In addition, since the judgment is performed by a person, it takes time and hinders labor saving and labor saving of the identification work. In order to solve such inconvenience, objective recognition using a recognition device as described below can be considered.

FIG. 10 is a block diagram showing the structure of a typical conventional pattern recognition apparatus 71. As shown in FIG. An input pattern such as an image or a sound is provided to the feature extracting unit 72. The feature extracting means 72 extracts a feature amount characterizing the pattern from the input pattern. The feature amount is, for example, an image, which is an essential information used to express an image pattern more simply and to be used for identification described later. The feature vector having the extracted feature amount as an element is provided to the identification unit 73.

[0006] The identification means 73 calculates and outputs a category vector representing which of a plurality of predetermined types of categories the input pattern belongs to, based on the input feature vector. The identification means 73 is realized by a neural network in which the relationship between the feature vector and the category vector has been learned in advance. The calculated category vector is provided to the identification result determination means 74.
The identification result determination means 74 determines a category to which the input pattern belongs based on the given category vector,
Output as the identification result. The identification result is provided to the output unit 75. The output unit 75 is realized by a display device such as a CRT (cathode ray tube), and outputs (notifies) the identification result to the outside (operator).

The neural network used for the identification means 73 is, for example, a back propagation neural network (hereinafter, referred to as a known technique).
"BPNN") is used. This BPNN
About `` NATURE vol.323 9 p536-553 (Oct 1986)
Learning Representations by backpropagationerror
s ".

[0008] BPNN comprises an input layer, an output layer, and one or more intermediate layers, each of which is composed of a plurality of node groups. In the identification by the neural network, for example, in the case of a general configuration in which each node of the output layer of the neural network is associated with a category to be identified on a one-to-one basis, a maximum value among a plurality of nodes constituting the output layer is output. The category corresponding to the node being identified is used as the identification result.

[0009]

In the prior art for detecting a reflected wave or a refracted wave by using an electromagnetic wave or the like, there is a problem that the type of the underground object cannot be specified. In addition, in the case of identification based on the feel when the probe penetrates into the ground and the degree of slippage of the probe, the type of the buried object 14 is specified by the operator as described above. If it is different, the identification accuracy will also be different, and it will be difficult to perform stable identification. In addition, since the judgment is performed by a person, it takes time and hinders labor saving and labor saving of the identification work.

Further, in the case of the identification using the pattern recognition device 71, although the inconvenience such as the case where the operator identifies as described above is solved, what kind of data should be inputted as the input pattern is optimal. It is not clear how concrete identification can be realized.

An object of the present invention is to provide an underground buried object type discriminating apparatus capable of discriminating the type of buried object material at high speed with stability and high accuracy.

[0012]

According to the present invention, there is provided an oscillating means having a vibrating probe inserted into the ground until it comes into contact with an underground object, and oscillating means for applying vibration to the underground object. A vibration receiving probe that is inserted into the ground until it comes into contact with the buried object, receives an elastic wave generated by the oscillation means applying vibration to the buried object, and outputs an elastic wave signal. Means, signal processing means for creating an identification signal for identifying the type of the underground object from the elastic wave signal output by the vibration receiving means, and an underground object represented by the identification signal based on the identification signal Identification means for identifying a type of the input layer, an input layer having a plurality of input units, an output layer having a number of output units corresponding to the type of the underground object, the input layer and the output And one or more intermediate layers interposed between the layers Includes a network, said signal processing means, as the identification signal, the acoustic wave signal analog /
A frequency response signal is created by performing a Fourier transform after digital conversion, and a divided response signal obtained by dividing the frequency response signal into a plurality of frequency bands is provided to a plurality of input units of an input layer of the neural network. It is a device for identifying the type of underground buried object. According to the present invention,
The vibration probe of the oscillating means and the vibration probe of the vibration receiving means,
It is inserted into the ground until it abuts on the underground object, gives vibration to the underground object through the vibrating probe of the oscillating means, and transmits the elastic wave generated by the underground object to the vibration receiving means. Vibration is received through the probe. An identification signal for identifying the type of material of the underground object is created from the elastic wave signal based on the received elastic wave, and the type of the underground object represented by the identification signal is identified based on the created identification signal. Is done. The elastic wave generated in the underground object is unique to the underground object, and therefore, the type of the material of the underground object can be identified as described above. The underground object identification method as described above is performed using an underground object type identification device including an oscillating unit, a vibration receiving unit, a signal processing unit, and an identification unit. The signal processing means creates the identification signal, and the identification means identifies the type of material of the underground object represented by the identification signal based on the identification signal created by the signal processing means. The identification means includes: an input layer having a plurality of input sections; an output layer having a number of output sections corresponding to the type of the underground object; and one or a plurality of output layers interposed between the input layer and the output layer. The signal processing means applies the following signals to the plurality of inputs of the input layer of the neural network as the identification signals. That is, the signal processing unit performs analog-to-digital conversion on the elastic wave signal, and then performs Fourier transform to create a frequency response signal. Give each to the department. The division response signal is an identification signal. Since the type of the material of the underground object is identified by the above-described method using the underground object type identification device, the type of the underground object can be manually identified as in the related art. Compared with this, stable and highly accurate identification becomes possible. In addition, the time required for identification is reduced, and labor and labor can be saved. Furthermore, if a plurality of oscillation and vibration receiving means are used, it is possible to identify many types of underground objects at once, and the identification efficiency can be further improved. When a neural network is used, it is relatively easy to change the number of input data to the input layer and the number of output data from the output layer.

Further, the identification means of the present invention determines the type of the underground object based on the output signal level output from the output section of the output layer of the neural network for each type of the underground object. It is characterized by including. According to the present invention, the identification means includes a determination means, and determines the type of the material of the underground object based on the output signal level output for each type of the underground object from the output unit of the output layer of the neural network. judge. Therefore, the identification accuracy can be further improved.

Further, the present invention is characterized in that it comprises output means for outputting the result of identification by the identification means. According to the present invention, the underground object identification device includes an output unit, and outputs an identification result of the identification unit. Therefore, the operator:
The identification result of the identification means can be known.

Further, the output means of the present invention is a display means for displaying an identification result of the identification means. According to the present invention, the underground object identification device includes an output unit, and outputs an identification result of the identification unit. The output means is realized by, for example, a display means. Therefore,
The operator can know the identification result of the identification means.

Further, the vibration probe and the vibration probe of the present invention are the same probe. According to the present invention, the excitation probe and the vibration receiving probe are realized by the same probe. Therefore, only one probe needs to be inserted into the underground, and the operation can be simplified.

Further, the present invention includes a storage unit for sequentially storing the elastic wave signals output from the vibration receiving unit, and the identification unit stores the elastic wave signal when the identification result by the identification unit is incorrect. Read the elastic wave signal, and train the neural network of the identification means using the read elastic wave signal. According to the present invention, the underground object identification device includes a storage unit that sequentially stores the elastic wave signals output by the vibration receiving unit, and the identification unit is configured to perform an identification when the identification result by the identification unit is incorrect. Reading the elastic wave signal stored in the storage means, and learning the neural network of the identification means using the read elastic wave signal. Therefore, even if erroneous identification is performed, correct identification can be performed from the next time, and the identification performance is further improved.
In addition, the identification performance can be easily improved.

[0018]

FIG. 1 is a block diagram showing an underground object identification device 27 according to an embodiment of the present invention. The identification device 27 includes the oscillation device 21, the vibration receiving device 22, the signal processing circuit 23, the identification circuit 24, the display processing circuit 25, the display device 26, and the database circuit 43.

The oscillating device 21 includes a vibrating probe 28, a hammer 29, a motor 30, and a spring 31. On one end side of the vibrating probe 28, the vibrating probe 2 is
A hammer 29 that is biased by a spring is disposed in parallel with the longitudinal direction (vertical direction) of the hammer. After the hammer 29 is pulled up by driving the motor 30 by the hammer drive circuit 30a, when the hammer 29 is dropped downward, an impact is applied to the vibrating probe 28 by the hammer 29.

The other end of the vibrating probe 28 is
The impact applied as described above by the hammer 29 until it comes into contact with the ground is transmitted to the vibrating probe 28 as attenuated vibration and transmitted to the buried object 35.
In the embedded object 35 to which the vibration has been transmitted, an elastic wave unique to the substance is generated. The vibrating probe 28 is inserted, for example, at an angle of 30 ° to 45 ° with respect to the ground surface,
Vibration is applied at a weight of ~ 0.3 kg.

The vibration receiving device 22 includes a vibration receiving probe 32 and a sensor 33. A sensor 33 is attached to one end of the vibration receiving probe 32. The other end of the vibration receiving probe 32 is inserted into the underground 34 until it comes into contact with the buried object 35. The elastic wave generated in the buried object 35 as described above is
The vibration is received by the vibration receiving probe 32, propagated through the vibration receiving probe 32, and converted into an elastic wave signal by the sensor 33. The elastic wave signal is input to the signal amplifier 36 of the signal processing circuit 23, the database circuit 43, and the display processing circuit 25.

The signal processing circuit 23 includes a signal amplifier 36, A
/ D converter 37, waveform storage circuit 38, FFT operation circuit 3
9 and an input vector generation circuit 40. In the signal processing circuit 23, an identification signal for identifying an underground object is created from the received elastic waves. That is, the elastic wave signal converted by the sensor 33 is input to the signal amplifier 36 and amplified. Unnecessary signals are removed using a filter included in the signal amplifier 36 as necessary. The signal processed by the signal amplifier 36 is A /
The signal is input to a D (analog / digital) converter 37, converted into a digital signal, further input to a waveform storage circuit 38, and stored as a time axis response signal.

The time axis response signal stored in the waveform storage circuit 38 is Fourier transformed by an FFT (Fast Fourier Transform) operation circuit 39 to generate a frequency response signal. The generated frequency response signal is divided by the input vector generation circuit 40 into a plurality of frequency bands to generate a divided response signal. For example, a signal in a frequency band of 20 Hz to 5 kHz is divided into a range of 10 to 100. For example, it is divided into 50. These divided signals are learned in advance, divided by the maximum value of each component in the data stored in the database circuit 43, and normalized to a value of 0 to 1. The divided response signal thus created is the identification signal.

The identification circuit 24 includes a neural network 41 and an output vector determination circuit 42. As the neural network 41, for example, B
PNN is used. The neural network 41 receives the divided response signal from the input vector generation circuit 40, creates an output signal for each type of underground object based on the divided response signal, and provides the output signal to the output vector determination circuit 42. The output vector determination circuit 42 determines the type of the underground object based on the output signal level from the neural network 41.

The waveform storage circuit 38, the FFT operation circuit 39,
Input vector generation circuit 40, neural network 4
1 and the output signal from the output vector determination circuit 42 are
The image is processed by the display processing circuit 25 and is displayed on a display device 26 realized by, for example, a liquid crystal display device.

FIG. 2 is a diagram showing a schematic configuration of the neural network 41 of the identification circuit 24. The neural network 41 is the BPNN, and the input layer 4
4, an intermediate layer 45 and an output layer 46. The input layer 44 includes a plurality of nodes NI1 to NIm. The intermediate layer 45 has one or a plurality of layer structures, and each layer includes a plurality of nodes NM. In this embodiment, the intermediate layer 45 is a single layer, and the intermediate layer 45 includes a plurality of nodes NM.
1 to NMp. The output layer 46 includes a plurality of nodes NO1 to NOn.

The number of nodes constituting the input layer 44 and the output layer 46 is determined according to the number of data to be input and the number of data to be output. That is, the number of nodes NI forming the input layer 44 corresponds to the number of divided response signals generated by the input vector generation circuit 40. Further, the number of nodes NO constituting the output layer 46 corresponds to the number of types of buried objects. For example, the number of nodes NI is 50
The number of nodes NO is selected to be four, and the number of nodes NM constituting the intermediate layer 45 is empirically selected to be twenty.

In the learning of the neural network 41, data of the elastic wave signal and the type of the underground object stored in the database circuit 43 in advance are used. The category vector corresponding to the type of the underground object indicates the type to be identified in the state of the node NO in the output layer 46 of the neural network 41. The node number of the output layer 46 is associated with the category to be identified on a one-to-one basis. For example, when learning the category corresponding to the node NOn of the output layer 46, the node NI of the input layer 44
A predetermined feature vector is given to 1 to NIm, respectively, and a category vector of Ci (i = n) = 1 and Ci (i ≠ n) = 0 is given to nodes NO1 to NOn of the output layer 46 as teacher signals, respectively.

FIG. 3 is a diagram for explaining a determination method in the output vector determination circuit 42 of the identification circuit 24. The description will be made assuming four types of objects A to D as the types of buried objects. Therefore, the output layer of the neural network 41 is composed of four nodes NO1 to NO4. From the neural network 41, four output signals corresponding to the types A to D are output and input to the output vector determination circuit 42. . At this time, the one having the maximum output signal level is selected as the identification candidate. In the output vector determination circuit 42,
Two thresholds T1 and T2 (T1> T2) are set in advance.

As shown in FIG. 3A, the output signal level is the maximum, the signal level of B selected as a candidate for identification is equal to or higher than T1, and the signal levels of A, C, and D are equal to T.
If it is 2 or less, B as the identification candidate is adopted as the type of the buried object. As shown in FIG. 3 (2), when the signal level of B is T2 or more and T1 or less, and the signal levels of A, C and D are T2 or less, or as shown in FIG. The signal level of B is equal to or higher than T1, and A, C, D
For example, when A is equal to or greater than T2, B as an identification candidate is rejected.

The threshold values T1 and T2 are empirically selected. For example, when the four output signal levels are close to each other, if T1 and T2 are set to relatively small values near the output signal level, the selected identification candidate can be adopted. If set to a large value, the selected identification candidate is rejected. T1 and T2 are set according to desired identification accuracy. For example, assuming that the maximum value of the output signal level is 1, T1 is set to 0.7 and T2 is set to 0.3.

FIG. 4 is a diagram showing a configuration of a display screen 47 displayed on the display device 26. As shown in FIG. The display screen 47 includes a time waveform display area 48 and frequency spectrum display areas 49 and 5.
0, an input vector display area 51, a neural network configuration display area 52, an output vector display area 53, and an embedded object name display area 54.

In the time waveform display area 48, a waveform based on the time axis response signal from the waveform storage circuit 38 is displayed, for example, a waveform for 0 to 50 msec is displayed. In the frequency spectrum display areas 49 and 50, the spectrum based on the frequency response signal from the FFT operation circuit 39 is displayed. One frequency spectrum display area 49 displays a spectrum between 0 and 20 kHz, for example, and the other frequency spectrum display area 50 displays, for example, 20H.
The spectrum between z and 5 kHz is displayed logarithmically.

In the input vector display area 51, a graph based on a plurality of divided response signals from the input vector generation circuit 40, that is, an input vector input to the neural network 41 is displayed. In the neural network configuration display area 52, the configuration of the neural network 41 described above is displayed. Output vector display area 53
, An output vector from the neural network 41 is displayed as, for example, a bar graph. In the embedded object name display area 54, a name indicating the type of the embedded object determined by the output vector determination circuit 42 is displayed.

FIG. 5 is a flowchart showing a method of identifying and learning an underground object. FIG. 6 is a diagram illustrating an example of the display screen 47. In step a1, the elastic wave signal received by the vibration receiving device 22 is input to the signal amplifier 36 of the signal processing circuit 23, converted into a digital signal by the A / D converter 37, and stored in the waveform storage circuit 38 as a time axis response signal. It is memorized. In step a2, the waveform storage circuit 38
Is subjected to a Fourier transform by the FFT operation circuit 39 to generate a frequency response signal. Further, the input vector generation circuit 40 generates a plurality of divided response signals. In step a3, the received elastic wave signal is stored in the database circuit 43. By storing the information in the database circuit 43, the neural network can be re-learned when an erroneous identification is made.

At step a4, it is determined whether the mode is the identification mode. If it is the identification mode, step a5
When the mode is not the identification mode but the learning mode, the process proceeds to step a9. In step a5, the divided response signal created by the input vector generation circuit 40 is input to the neural network 41 and subjected to neural network identification processing. In step a6, the output signal from the neural network 41 is input to the output vector determination circuit 42, and the identification result determination process is performed. At step a7, for example, a display screen 47 shown in FIG.

At step a8, it is determined whether or not the processing is completed. If not, the process returns to step a1.

In step a9 when the learning mode is set in step a4, the elastic wave signal stored in the database circuit 43 in step a3 is read out, and the elastic wave signal is used for learning of the neural network 41. .

In step a10, the neural network 41 learns the relationship between the read elastic wave signal and the type of the buried object. When the learning process ends, the process returns to step a8.

The learning mode can be executed before performing the identification, and can also be executed when the identification is performed and the result of the identification is incorrect. An error in the identification result is determined by the operator. By operating the learning mode in this way, the discrimination ability can be further improved.

FIGS. 7 and 8 are graphs showing examples of elastic waves for each type of buried object. FIG. 7 (1), (3),
(5), (7) and FIGS. 8 (1), (3), (5),
(7) is a graph showing the relationship between elastic wave intensity and time, and is shown in FIGS. 7 (2), (4), (6), (8) and FIG.
(2), (4), (6) and (8) are graphs showing the relationship between elastic wave intensity and frequency. FIGS. 7 and 8 show (1) and (2) when the buried object is a metal piece.
(4) represents the case of plastic pieces, (5) and (6) represent the case of stone pieces, and (7) and (8) represent the case of wood pieces. FIG. 7 shows a case where the soil is dry loam soil, and FIG. 8 shows a case where the soil is wet sand.

From the frequency spectra of FIG. 7 and FIG.
It can be seen that there is a response characteristic unique to the substance. Also, it is understood that the influence of the measurement conditions is small. Furthermore, it has been confirmed that the reproducibility is also excellent. from these things,
It can be seen that it is possible to stably and accurately identify the type of underground object using elastic waves.

FIG. 9 shows an oscillation device 21 and a vibration receiving device 22.
It is a block diagram which shows the oscillation / vibration receiving apparatus 55 which is another example. The oscillation / vibration device 55 is configured such that the vibration probe 28 of the oscillation device 21 and the vibration probe 32 of the vibration device 22 are the same probe.
6, a motor 57, a spring 58, a sensor 59, and a vibration / vibration receiving probe 60 are provided. Attenuated vibration generated by applying an impact to the probe 60 by the hammer 56 and the motor 57 in the same manner as in the oscillation device 21 is propagated through the excitation / vibration receiving probe 60 and transmitted to the embedded object 35. The elastic wave generated by the embedded object 35 is received by the excitation / vibration probe 60, propagates through the excitation / vibration probe 60, and is converted into an elastic wave signal by the sensor 59. The elastic wave signal is input to the signal amplifier 36.

As described above, according to the present embodiment, it is possible to identify the type of the buried object with stability and high accuracy, as compared with the case where the operator identifies the type of the buried object as in the prior art.
In addition, identification can be performed in a short time, and labor and labor can be saved.

Further, when the neural network 41 is used, the number of input data to the input layer 44 and the output layer 4
It is relatively easy to change the number of output data from 6. In addition to the divided response signal as input data to the input layer 44 of the neural network 41, geological conditions (for example, probe insertion force, soil density, etc.) and measurement conditions (for example, probe insertion angle, insertion depth, etc.) ) Can further improve the identification accuracy. In the present embodiment, an example of BPNN has been described. However, similar effects can be obtained even when a non-linear function other than the sigmoid function is used, or when any neural network other than BPNN is used.

Further, by making a determination based on the output signal level from the neural network 41 in the output vector determination circuit 42, it is possible to improve the identification accuracy. If the output vector determination circuit 42 is not provided, for example, the type of the underground object is determined by the output signal from the neural network 41 having the maximum output signal level.

Further, the identification status is displayed step by step on the display device 26, so that the operator can know the identification result and further know the process of the identification process. In addition to the display on the display device 26, the present invention also includes an example in which an informing unit that emits a buzzer sound when the identification result indicates the type of the buried object is desired, and an example in which a flashing (flashing) display is performed using a warning light. It belongs to the range.

Furthermore, by using the oscillation / vibration receiving device 55, only one probe needs to be inserted into the ground, and the operation can be simplified.

[0049]

As described above, according to the present invention, the type of material of an underground object can be identified with high accuracy and in a short period of time, and labor and labor can be saved. It becomes possible. Using a neural network makes it relatively easy to change the number of input data to the neural network and the number of output data from the neural network. Further, by providing the determination means, the identification accuracy is improved. Further, by outputting the identification result, the operator can know the identification result. Further, by using the same probe for the vibration probe and the vibration probe, the operation can be simplified. Also,
Even if erroneous identification is performed, by learning the neural network, correct identification can be performed from the next time, and the identification performance is further improved.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an underground object type identification device 27 according to an embodiment of the present invention.

FIG. 2 is a diagram showing a schematic configuration of a neural network 41 of the identification circuit 24.

FIG. 3 is a diagram illustrating a determination method in an output vector determination circuit of the identification circuit.

FIG. 4 is a diagram showing a configuration of a display screen 47 displayed on the display device 26.

FIG. 5 is a flowchart showing a method of identifying and learning the type of an underground object.

FIG. 6 is a diagram showing an example of a display screen 47.

FIG. 7 is a graph showing an example of an elastic wave of a buried object.

FIG. 8 is a graph showing an example of an elastic wave of a buried object.

FIG. 9 is a block diagram showing an oscillation / vibration receiving device 55.

FIG. 10 shows a pattern recognition device 71 as a typical conventional example.
FIG. 3 is a block diagram showing the configuration of FIG.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 21 Oscillation device 22 Vibration receiving device 23 Signal processing circuit 24 Identification circuit 26 Display device 27 Identification device of type of underground object 28 Vibration probe 29 Hammer 32 Vibration probe 33 Sensor 34 Underground 35 Buried object 36 Signal amplifier 37 A / D (analog / digital) converter 38 waveform storage circuit 39 FFT operation circuit 40 input vector generation circuit 41 neural network 42 output vector determination circuit 43 database circuit 44 input layer 45 intermediate layer 46 output layer 55 oscillation / vibration device 60 Vibration / vibration probe

──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masahiko Kano 1-18-32 Fuchinobe, Sagamihara City, Kanagawa Prefecture A-302 (72) Inventor Tatsuo Makino 1 Kawasakicho, Kakamigahara City, Gifu Prefecture Kawasaki Heavy Industries, Ltd. Gifu Factory (72) Inventor Masaka Uno 1st Kawasaki-cho, Kakamigahara-shi, Gifu Pref. In the Gifu factory of Kawasaki Heavy Industries, Ltd. (72) Inventor China Kimura 1st Kawasaki-cho, Kakamigahara-shi, Gifu pref. References JP 5-5785 (JP, A) JP 8-152472 (JP, A) JP 3-286380 (JP, A) JP 4-84754 (JP, A) 2-179471 (JP, A) Japanese Patent Laid-Open No. 2-176011 (JP, A) IEEE Conference Publication No. 409 P393-398 "Neural Networks S For The Detition Of Buried Plan T", Artificial Neuronal Networks', 26-28 June 1995 Defense Technology Journal "Underground Exploration Radar" Vol. 15 NO. 8 P9-17 August, 1995 IGARSS VOL. 1994 NO.4 P2117-2119 "WAVELETS AND NEUTRAL NETS" (58) Fields investigated (Int. Cl. ⁶ , DB name) G01N 29/00-29/28 G01S 15/00-15/96 G01V 1/00- 1/52

Claims (6)

(57) [Claims] 1. An oscillating means having a vibration probe inserted into the ground until it comes into contact with an underground object, and an oscillating means for applying vibration to the underground object; A vibration receiving tip inserted therein, the oscillation means receiving an elastic wave generated by applying vibration to the object buried underground, and a vibration receiving means for outputting as an elastic wave signal; From the elastic wave signal obtained, a signal processing means for creating an identification signal for identifying the type of the underground object, and identification means for identifying the type of the underground object represented by the identification signal based on the identification signal Comprising: an input layer having a plurality of input portions; an output layer having a number of output portions corresponding to the type of the underground object; and an intervening portion between the input layer and the output layer. Including a neural network consisting of one or more hidden layers The signal processing means generates a frequency response signal by subjecting the acoustic wave signal to analog / digital conversion and then Fourier-transforming it as the identification signal, and generates a divided response signal obtained by dividing the frequency response signal into a plurality of frequency bands. An underground object type discriminating apparatus, which is provided to each of a plurality of input units of an input layer of a neural network. 2. The method according to claim 1, wherein the identification unit includes a determination unit configured to determine a type of the underground object based on an output signal level output from the output unit of the output layer of the neural network for each type of the underground object. 2. An apparatus for identifying the type of an underground object according to claim 1, wherein: 3. The apparatus according to claim 1, further comprising an output unit that outputs an identification result of the identification unit. 4. The apparatus according to claim 3, wherein said output means is display means for displaying a result of identification by said identification means.
Apparatus for identifying the type of underground object described. 5. The apparatus according to claim 1, wherein the vibrating probe and the vibrating probe are the same probe. 6. A storage unit for sequentially storing elastic wave signals output from said vibration receiving unit, wherein said identification unit stores an elastic wave signal stored in said storage unit when an identification result by said identification unit is incorrect. 2. The underground object type discriminating apparatus according to claim 1, wherein the signal is read, and the neural network of the discriminating means is learned using the read elastic wave signal.