JP4012366B2 - Surface flaw detector - Google Patents

Description

【００３８】
疵個数計数回路７ｄは、弁別回路７ａが弁別し第１テーブルに書込まれた疵候補の（＋）数及び（−）数を計数する。
【００３９】
エリア内疵個数演算回路１０は、エリア内パターン化回路９の各テーブルに書込まれている疵候補個数のＯＲ（論理和）を取り、疵要素の総数（論理和での総数）を算出する。即ち第１テーブル〜第３テーブルの疵候補である（＋）数及び（−）数のＯＲ（論理和）を取る。
【００４０】
表１の疵個数テーブルに、第１〜第３テーブルの疵候補（＋）あるいは疵候補（−）の論理和すなわち、第1〜第3テーブルにおいて当該画素が（＋）あるいは（−）であれば疵個数テーブルの当該画素を疵有の（×）とし、第1〜第3テーブルにおいて当該画素がいずれも（０）であれば疵個数テーブルの当該画素を疵無の（０）と示す。この（×）の合計数６が疵要素の総数となる。結果として、第１〜第３テーブルの疵パターン情報が１個の情報に圧縮（１０画素／１エリアから１画素）される。
【００４１】
エリア内濃度情報演算回路８は、同期回路５のバッファから出力された画像信号の、エリア内に於ける最大値、最小値、及び平均値を演算する。即ちフィルタリングする前の信号の濃度情報を演算する。エリア極性検出回路１２ａは、該最大値と平均値との差Ｓ＋（明るい信号を意味する）及び平均値と最小値との差Ｓ−（暗い信号を意味する）を生成する。
【００４２】
エリア内疵有無判定回路１１は、疵個数計数回路７ｄが計数した疵候補数、エリア内疵個数演算回路１０が算出した疵要素の総数、及びエリア極性検出回路１２ａからの最大値と平均値との差Ｓ＋、及び平均値と最小値との差Ｓ−に基づき該エリアを疵ありとするか否かの判定を行なう。
【００４３】
エリア内代表選択回路１２ｂは、エリア極性検出回路１２ａが判定した極性（Ｓ＋≧Ｓ−ならＳ＋；Ｓ-＞Ｓ+ならＳ-）、エリア内濃度情報演算回路８が演算した画像信号の最大値、最小値、平均値、疵個数計数回路７ｄが計数した疵候補数、及びエリア内疵個数演算回路１０が算出した疵画素の総数から、エリア内代表信号の選択、即ち最大値をエリア内代表信号として選択するか、最小値を選択するか、あるいは平均値を選択するか、の選択を行なう。この様に回路１０、１１、１２ｂは信号をエリア単位に圧縮して処理する。エリア内疵有無判定及びエリア内代表信号の選択を表にして示す。
【００４４】
【表２】

【００４５】
表中の第1行目では、例えばフィルタ６ａが抽出した疵個数は０（弁別；７ａ、計数；７ｄ）であるが、フィルタ６ａ、６ｂ、６ｃが抽出した疵の論理和を求めたときの疵個数（論理和演算及び個数演算；１０）が１である場合には信号Ａは１となり、エリア内に疵「有」と判定（判定；１１）し、疵「有」とする旨の信号となる。そして極性信号がＳ＋（演算；１２a）である場合にはエリア内疵代表値として信号Ｃは最大値となり、極性信号がＳ−（演算；１２a）である場合には、エリア内疵代表として信号Ｃは最小値を使用する。
【００４６】
上記実施例では、最大値と平均値との差Ｓ＋、及び平均値と最小値との差Ｓ−を直接比較し、エリア内代表信号を決定する例を述べた。しかし、Ｓ＋、Ｓ−の大きさが同じでも疵の有害性が異なることがあるのでＳ＋、Ｓ−を直接比較するのではなく、重みをつけることで、優先性を持たせることもある。例えば、最小値の信号を重みをつけて優先したい場合では、Ｓ＋、Ｓ−の信号値に重み分の差を付けて比較し、最大値、最小値、平均値の選択を行う。
【００４７】
比較信号の選択基準を
ｆ（Ｓ＋、Ｓ−）＝Ｓ＋ − （Ｓ− ＋α）
α：オフセット、αは正負をとりうる
ｆ（Ｓ＋、Ｓ−）＞０の時 Ｓ＋
ｆ（Ｓ＋、Ｓ−）≦０の時 Ｓ−
として、重みを加えることで、同じ有害度でも最大値が大きくなる場合、最小値が小さくなる場合と言った、疵の特性に即した代表内代表信号の選択が可能となる。
【００４８】
総疵個数（論理和個数演算；１０）の信号Ａ、疵有無（判定；１１）の信号Ｂ及びエリア内疵代表信号(選択；１２ｂ)の信号Ｃは、疵個数２次元化合成回路１３、疵有無マップ２次元化合成回路１４及び濃淡画像２次元化合成回路１５において、他のカメラ３ｂ、３ｃ、３ｄからの信号と併せて検査対象鋼板１の全幅（ｘ方向）に対応した信号にすると共に、検査対象鋼板１の長さ方向（ｙ方向）に集積してｘｙ方向１画面分の２次元画像を合成する。
【００４９】
図２の（ｂ）は、カメラ３ａ〜３ｄの合成１画面の一例を模擬的に示す。図２の（ｂ）は図２の（ａ）で示したエリアを縮小して表示しており、点線で囲まれた最小１区画が１画素である。画面の横幅（ｘ方向幅）は４分割され、左方から１次元カメラ３ａ、３ｂ、３ｃ、３ｄの順に撮影領域を分担している。例えば１次元カメラ３ａはｘ方向（画面左側から横方向）に２５画素／５エリアを分担し、ｙ方向（縦方向）に８画素／４エリアを分担している。４台のカメラが撮影した１画面の中には、即ち画像圧縮前の撮影画面の中には、例えば図に示す様な線、面、点に相当する画像信号が含まれる。
【００５０】
図２の（ｃ）は、（ｂ）で示した横５、縦２画素の領域が、横１、縦１画素の領域に縮小（圧縮）表示される状態を示す。すなわちエリア内疵個数演算回路１０で１０画素／１エリアから１画素に画像圧縮された状態を合成１画面で示す。図２の（ｂ）で示した線、面、点に相当する画像信号は、図２の（ｃ）では線状の信号は疵と判定されて表示され、面状の信号も疵と判定されて表示されるが、点状の信号は疵とは判定されず表示されていない。
【００５１】
画像フィルタ１６は、空間フィルタであり濃淡画像２次元化合成回路１５からの濃淡の２次元分布より２次元的な疵情報を抽出する。表にｘ方向３個、ｙ方向７個の画像フィルタ係数マトリックスを示す。
【００５２】
【表３】

【００５３】
画像フィルタフィルタ１６に入力される濃淡画像２次元化合成回路１５からの注目画素を中心とした濃淡の２次元分布（ｘ方向３個、ｙ方向７個の領域に於ける最大値、最小値、平均値の２次元分布）のｘ方向３個に、表３に示す係数マトリックス１、０、−１を乗算し、これをｙ方向７個について順次行ない、得た２１個の積の総和が注目画素のフィルタ通過値である。これにより大きいが濃度は薄い疵を抽出することが出来る。弁別回路１７は画像フィルタ１６からの正極性、負極性及び無の３値信号を、信号有無の単純２値化し疵候補の有無の判定を行なう。
【００５４】
合成回路１８は、弁別回路１７からの疵候補信号（大疵情報）と疵有無マップ２次元化合成回路１４からの疵有無マップ信号（小疵情報）の合成をＯＲロジックで行ない、疵候補の２次元画像を得る。
【００５５】
疵特徴検出回路１９では、この合成回路１８からの信号と疵個数２次元化合成回路１３からの信号及び濃淡画像２次元化合成回路１５からの信号を基に、疵信号の特徴量を演算する。
【００５６】
図４に疵特徴検出回路１９の処理手順を示す。合成回路１８からの疵有無マップ２次元信号（１；処理手順に対応。以下同様）を、２次元走査（ｘ、ｙ方向）によりラベリング処理し（２）、例えば図に示す様に、ブローブ１、ブローブ２、ブローブ３として検出する（３）。各ブローブの面積を閾値（基準値）で識別し、基準値以下のブローブはノイズとして除去し（４）、疵候補を決定する（５）。例えば図に示す様に、ブローブ３はノイズとして除去する。疵特徴検出回路１９は以下の各パラメータに基づき、疵信号の特微量を演算する。
【００５７】
「形状パラメータ」；幅、長さ、面積、傾き、縦横比
「濃度パラメータ」；平均濃度、最大濃度、最小濃度、濃度の分散
「画素内疵個数パラメータ」；圧縮後の画素（当所の１エリア）内疵個数がｎ個以下の割合（微小な疵）
形状パラメータに基づく疵形状特微量の演算は、決定された各疵候補毎に行なう（６）。濃淡パラメータに基づく疵濃淡特微量の演算は、各疵候補毎に、該疵候補に対応する濃淡画像２次元化信号により行なう（７）。疵個数パラメータに基づく疵個数特微量の演算は、各疵候補毎に、該疵候補に対応する疵個数２次元化信号により行なう（８）。
【００５８】
疵種・有害度判定回路２０は、１画面毎の各疵候補について、上記の疵特微量を用いて疵の名称及び有害の度合を判定する。本実施例に於ける１画面（ｘ方向幅４０００画素、ｙ方向５００画素）は、被検査対象鋼板１のｘ方向幅２０００ｍｍ、ｙ方向長さ５００ｍｍの範囲に相当する。
【００５９】
疵種・有害度判定は、判定パーツ、即ち
「IＦ ＴＨＥＮ ルール」
「ツリー状ルール」
「テーブル状ルール」
「ニューラルネットワーク」
等のいずれかを多層状に組み合わせて、あるいはそれらを組み合わせて行なう。
図５に判定パーツの一例を概略で示す。すなわち（ａ）はツリー状ルールの一例を示し、（ｂ）はテーブル状ルールの一例を示し、（ｃ）はニューラルネットワークの一例を示す。
【００６０】
図６に、「ツリー状ルール」を多層状に組み合わせて疵種・有害度判定を実行する一例を示す。疵種判定を多層で繰り返えし、疵種類及び有害度を求めるものである。
【００６１】
図７に、４台の１次元カメラ３ａ、３ｂ、３ｃ、３ｄの撮影画像を１枚の画像に合成する画像合成システムの信号同期に関する部分の構成を示す。カメラコントローラ２２からクロック信号をカメラ３ａ〜３ｄ、シェーディング補正・Ａ／Ｄ変換回路４ａ〜４ｄ、同期回路５ａ〜５ｄ、信号処理回路２１ａ〜２１ｄ（図１にも示す）、及び合成回路１３〜１５に供給する。４台のカメラ３ａ、３ｂ、３ｃ、３ｄは、クロック信号に基づき検査対象鋼板１のｘ方向（幅方向）を撮影（走査）を同時に分担して行ない、これを繰返えす。シェーディング補正回路４ａ、４ｂ、４ｃ、４ｄは各々Ａ／Ｄ変換器を備え、カメラ３ａ〜３ｄからのアナログ画像信号をデジタル画像信号に変換した後、検査対象鋼板１の幅方向（ｘ方向）の明度差、あるいは１次元ＣＣＤカメラの幅方向感度差等による出力変化を補正する。
【００６２】
移動距離検出器２３は、被検査対象鋼板１が一定距離（例えば１ｍｍ）移動する毎にパルスを発生するエンコーダであり、同期回路５ａ〜５ｄ、信号処理回路２１ａ〜２１ｄ及び合成回路１３〜１５に距離信号ＬＴを供給する。同期回路５ａ〜５ｄは該距離信号ＬＴによりエネイブル（ゲート）信号を作成し、信号処理回路２１ａ〜２１ｄ及び合成回路１３〜１５も同様にゲート信号を作成する。
【００６３】
同期回路５ａはバッファを備え、該バッファはエネイブル期間に入力されたシェーディング補正回路４ａからのデジタル画像信号をメモリする。同期回路５ｂ〜５ｄのバッファも、各々同様にしてエネイブル期間に入力されたシェーディング補正回路４ｂ〜４ｄからのデジタル画像信号をメモリする。カメラ３ａ〜３ｄはクロック信号に基づき走査を繰返すが、エネイブル期間でなければ画像信号は各々のバッファにメモリされない。即ち距離信号ＬＴが供給された時にのみ各画像信号をメモリする。この様にしてメモリされた各カメラ対応の画像信号は、疵個数２次元化合成回路１３、疵有無マップ２次元化合成回路１４及び濃淡画像２次元化合成回路１５において合成し、検査対象鋼板１の全幅（ｘ方向）に対応した信号に一元化すると共に、該信号をｙ方向に積算してｘｙ方向１画面分の２次元画像とする。
【００６４】
以上の様に不要な情報を前処理で集約し必要な情報を残して圧縮するので、検出情報の質の低下を伴うことなく情報量を低減できる。この為オンライン処理によって精度が高い表面疵の検査を高速で処理することが可能であり、検査効率が向上する。加えて、本実施例では濃度情報（中間調）を含む疵情報を提供出来るので検査員の疵検査能力が向上する。
【００６５】
−第２実施例−
図８に本発明の第２実施例による表面疵検出装置の構成を示す。本実施例ではｙ方向に移動する帯状の検査対象鋼板１を、４台のＣＣＤカメラ３ａ、３ｂ、３ｃ、３ｄにより幅方向（ｘ方向）を４分割して撮影する。検査対象鋼板１の幅方向（ｘ方向）を照明装置２により一様に照明し、この照明光が検査対象鋼板１表面で正反射した方向に上記４台のカメラ３ａ、３ｂ、３ｃ、３ｄの視野を合わせてあるので、各カメラには一様な明るい画像が得られる。
【００６６】
図９は、図８に示す表面疵検出装置の４台のＣＣＤカメラ３ａ、３ｂ、３ｃ、３ｄのタイミングチヤートである。
【００６７】
１次元カメラ３ａ〜３ｄからのアナログ画像信号は、シェーディング補正回路４ａ〜４ｄが備えるＡ／Ｄ変換器によりデジタル変換された後、感度補正される。即ち検査対象鋼板１の幅方向（ｘ方向）の明度差、あるいはＣＣＤカメラの幅方向感度差等による出力変化を補正する。カメラ３ａ〜３ｄは、カメラコントローラ２２から供給される周波数ｆの第１クロック信号に従い検査対象鋼板１を１／４板幅づつ同時に撮影する。カメラ３ａ〜３ｄからの１次元画像信号は、シェーディング補正回路４ａ〜４ｄを介して画像メモリ２４ａ〜２４ｄに１走査線分、画素毎に記憶され、走査毎に記憶内容は書換えられる。
【００６８】
移動距離検出器２３は検査対象鋼板１の一定距離移動（ｙ方向）に対応してパルスを発生するエンコーダであり、読出しタイミング回路２５、同期回路５及び信号処理回路２１ａに距離信号ＬＴを供給する。読出しタイミング回路２５は該距離信号ＬＴによりエネイブル（ゲート）信号を作成し、同期回路５及び信号処理回路２１ａも同様にゲート信号を作成する。
【００６９】
読出しタイミング回路２５は上記エネイブル期間中、カメラコントローラ２２から供給されるカメラ３ａ〜３ｄが撮影に使用した第１クロック信号の周波数ｆを２倍した、周波数２ｆの第２クロック信号を作成し、画像メモリ２４ａ〜２４ｄ及び選択回路２６に供給する。カメラ３ａが撮影し画像メモリ２４ａに記憶されている画像信号（ＣＨ１）とカメラ３ｂが撮影し画像メモリ２４ｂに記憶されている画像信号（ＣＨ２）は、タイミング回路２５から供給される周波数２ｆの第２クロック信号によって読出され、選択回路２６はその順番にＣＨ１、２を統合し、選択回路２６が持つ第１バッファにメモリする。同様に上記エネイブル期間中、画像メモリ２４ｃに記憶されている画像信号（ＣＨ３）とカメラ３ｄが撮影し画像メモリ２４ｄに記憶されている画像信号（ＣＨ４）は、タイミング回路２５から供給される周波数２ｆの第２クロック信号によって読出され、選択回路２６はその順番にＣＨ３、４を統合し、選択回路２６が持つ第２バッファにメモリする。
【００７０】
読出しタイミング回路２５は、カメラコントローラ２２から供給されるカメラ３ａ〜３ｄが撮影に使用した第１クロック信号の周波数ｆを４倍した、周波数４ｆの第３クロック信号を作成し、選択回路２６に供給する。
【００７１】
選択回路２６は上記第１バッファにメモリされているCH1、2統合信号、及び第2バッファにメモリされているCH3、4統合信号を、該周波数４ｆの第３クロック信号で読み出し統合する。即ち画像信号を画像メモリ２４ａ、画像メモリ２４ｂ、画像メモリ２４ｃ、画像メモリ２４ｄの順（ＣＨ１、ＣＨ２、ＣＨ３、ＣＨ４の順）に読出して統合（１元化）し、同期回路５に供給する。同期回路５に入力される段階で既に４台のカメラ信号が１行のシリアルデータに統合されている。従って信号処理回路は１個（２１ａ）のみであり、また疵個数２次元化回路１３、疵有無マップ２次元化合成回路１４、濃淡画像２次元化合成回路の入力は１個のみであり、これらの回路は各々の入力信号をｙ方向に積算するのみでｘｙ方向１画面分の２次元画像を合成する。上記説明以外の信号処理は、第１実施例と同様であるので説明は省略する。
【００７２】
４台のＣＣＤカメラ３ａ、３ｂ、３ｃ、３ｄにより検査対象鋼板１をｘ方向（幅方向）に４分割して同一時期に撮影して画像メモリ２４ａ〜２４ｄに記憶し、これを撮影時の４倍遠のクロック信号で読出し統合（１元化）することにより、恰も一台のカメラで検査対象鋼板１の全幅を撮影した如く信号処理を行なうので、伝送コスト及び信号処理コストを削減出来る。
【００７３】
【発明の効果】
以上説明したように本発明によれば、オンラインの質劣化が少ない疵情報を提供し、検出信号処理の簡素化及び疵検出装置の低コスト化を図ることができる。
【図面の簡単な説明】
【図１】本発明の第１実施例の構成を示すブロック図である。
【図２】図２（ａ）は図１に示す１台の１次元カメラが走査する２本の走査線上の画素及びエリアを示す平面図であり、図２（ｂ）は４台の１次元カメラが撮影する画面の一例を示す平面図であり、図２（ｃ）は図２（ｂ）の信号を処理し圧縮した状態を示す平面図である。
【図３】図１に示すフィルタ６ａ〜６ｃ及び弁別回路７ａ〜７ｃの構成を示すブロック図である。
【図４】疵特徴検出回路１９の処理手順を示すブロック図である。
【図５】疵特徴検出のための判定ツールの一例を示し、図５（ａ）はツリー状ルールを示すブロック図であり、図５（ｂ）はテーブル状ルールを示す図であり、図５（ｃ）はニューラルネットワークを示すブロック図である。
【図６】ツリー状ルールを使用した疵種・有害度判定の一例を示すブロック図である。
【図７】図１に示す４台の１次元カメラの撮影画像を１画面に画像合成処理を行なう部分の回路構成を示すブロック図である。
【図８】本発明の第２実施例の構成を示すブロック図である。
【図９】図８に示す４台の１次元カメラの撮影画像を１画面に画像合成処理を行なう際のタイミングチャートである。
【符号の説明】
１：検査対象鋼板
２：照明装置
３ａ〜３ｄ： １次元ＣＣＤカメラ
３ａ１〜３ｄ１：カメラ信号処理回路
４ａ〜４ｄ：シェーディング補正回路（含Ａ／Ｄ変換器）
５：同期回路
６ａ、６ｂ、６ｃ：フィルタ回路
７ ａ、７ｂ、７ｃ：弁別回路
７ｄ：疵個数計数回路
８：エリア内濃度情報演算回路
９：エリア内疵パターン化回路
１０：エリア内疵個数演算回路
１１：エリア内疵有無判定回路
１２ａ：エリア極性検出回路
１２ｂ：エリアデータ選択回路
１３：疵個数２次元化合成回路
１４：疵有無マップ２次元化合成回路
１５：濃淡画像２次元化合成回路
１６：画像フィルタ
１７：弁別回路
１８：合成回路
１９：疵特徴検出回路
２０：疵種、有害度判定回路
２１a〜２１ｄ：信号処理回路（６〜１２を纏めたもの）
２２：カメラコントローラ
２３：移動距離検出器
２４ａ〜２５ｄ：メモリ
２６：選択回路[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an apparatus for detecting a surface flaw based on an image signal obtained by photographing the surface of an inspection target material such as a steel plate.
[0002]
[Prior art]
As an apparatus for on-line inspection of surface flaws of strips such as steel and aluminum plates, a method using image processing is known, and for example, Japanese Patent Laid-Open No. 7-63696 is proposed. According to this, the surface width direction of the belt-shaped inspected material to be conveyed is irradiated with illumination light, and the reflected light from the surface of the inspected material is received by the reflected light detector and converted into an electrical signal. This electric signal is quantized to each level and given to the line buffer, and the line buffer binarizes and stores it for each scanning line. The filter reads the signal of the line buffer, compresses the information based on a predetermined rule, outputs the compressed information to the image processing apparatus, and discriminates the detailed information and outputs the result to the wrinkle determination device. The image processing apparatus creates a two-dimensional image based on the compressed information, and obtains the amount of sputum. The wrinkle determination device determines the type, grade, and the like of the wrinkle based on the feature amount and the fine wrinkle information. Since this apparatus can reduce the amount of information without degrading the quality of the detected information, it is said that high-precision surface defect inspection can be performed at high speed by online processing.
[0003]
In addition, when inspecting the surface flaw of a strip-shaped inspection material with an image, if the inspection material is wide, it is difficult to cover the entire width with a single optical detector. It is known to cover the full width. In this case, it is common to provide a signal processing circuit corresponding to each photodetector, and collect the outputs of each signal processing circuit in one system and supply them to the subsequent system.
[0004]
However, since providing a plurality of signal processing circuits corresponding to each photodetector is disadvantageous in terms of cost and the like, proposals have been made to process the outputs of the plurality of photodetectors with a single signal processing circuit. According to Japanese Patent Application Laid-Open No. 61-223607, a plurality of scanning detectors that scan by sharing a material to be inspected, and scanning by these scanning detectors are sequentially scanned so as not to overlap in time. Synchronization control means, circuit means for combining the output signals of the scanning period of each scanning detector into a series of signals arranged in series in time, and discriminating surface defects of the material to be inspected from the collected signals Signal processing means. With such a configuration, a plurality of scanning detectors can be handled with a single signal processing means, and the circuit configuration is simplified.
[0005]
[Problems to be solved by the invention]
  The surface flaw inspection apparatus disclosed in Japanese Patent Laid-Open No. 7-63696 discloses a processing target area and pixels before and after the processing target area.ThresholdDiscrimination based on the value is used to obtain a flaw candidate, and the discrimination result is determined on the basis of regularity to determine whether flaws are present and compressed to 1 bit, and information on fine wrinkles is obtained, and this is performed for all pixels. Yes. However, in order to perform surface flaw inspection with high accuracy online, it is necessary to reduce the amount of information at high speed without degrading the quality of the detected information. In addition to the compressed heel information, detailed information on the heel is required.
[0006]
In addition, in the apparatus disclosed in Japanese Patent Laid-Open No. Sho 61-223607, when the wrinkle inspection of a wide inspection object is performed, the entire width of the inspection area is covered by sequentially scanning a plurality of photodetectors. For example, in a detector of the type that scans a laser beam with a rotating mirror, the servomotor that drives the rotating mirror must also be sequentially scanned, and the mechanism is complicated.
[0007]
The first object of the present invention is to provide defect information with little quality deterioration in online high-speed processing, and it is a relatively rough surface division density required by an inspector during defect inspection, but defect determination accuracy is high. Providing eyelid information as a second issue, and simplification of detection signal processing when a large surface of the eyelid inspection object is subdivided and each divided area is divided and detected by each of a plurality of reflection luminance detecting devices. The fourth problem is to reduce the cost of the wrinkle inspection device.
[0008]
[Means for Solving the Problems]
The surface flaw detection apparatus of the present invention is
  1. Light intensity detecting means (3a) for converting the luminance of reflected light on the surface of the inspection target material into an electric signal and generating serially corresponding to the width direction x of the inspection target material; analog for converting the electric signal into digital data, that is, image data / Digital conversion means (4, 5); first filter means for applying a process for enhancing the luminance difference of each pixel in digital conversion to the image data; and a second filter for applying a smoothing process for smoothing the brightness difference Means (6a to 6c); first primary defect determination data indicating whether or not the luminance level of the image data subjected to the process for enhancing the luminance difference is within a predetermined range (absence), and the smoothing Primary processing means (7a-7c) for generating second primary wrinkle determination data indicating whether or not the luminance level of the image data subjected to the digitization processing is within a predetermined range (having no), and having the width direction x and From multiple (5 × 2 = 10) pixels in the y direction orthogonal to it That areaeveryOf the first and second primary wrinkle determination dataWhether each table is within the specified range“No” (疵 有)Write and patternPrimary synthesis means (9); said areaThe number of “No” (疵 existence) in the first primary wage determination data table in FIG. 5 is counted, and the logic of “No” (疵 existence) is provided for the first and second primary wage determination data tables. A sum is generated and obtained by the logical sumPrimary determination す る counting means (7d, 10) for counting the number of “No” (疵); maximum value and average value in the area of the image data obtained by the analog / digital conversion means (4, 5) A polarity detection means (8, 12a) for generating a difference S1 between the average value and the minimum value, and detecting a polarity indicating the magnitude of the difference S1 and S2, and the primary determination soot counting means (7d, Based on the number counted by 10) and the polarity detected by the polarity detection means (8, 12a), the area is determined to be a common area or a non-exclusive area, and a signal representing this is generated. Information generating means (11); and the maximum value of the image data in the area based on the number counted by the primary determination and counting means (7d, 10) and the polarity detected by the polarity detecting means (8, 12a), Area data selection means (12b) for selecting a minimum value or an average value; A surface flaw detection device
[0009]
  2.further,in frontDAn area indicating whether the data selected by the rear data selecting means (12b) is within a predetermined range (not present) or not (not present); secondary processing means (15 to 17) for generating secondary information; informationonceArea 疵 signal output means (14, 18) for outputting the logical sum of the area 疵 primary information generated by the generating means (11) and the area 疵 secondary information generated by the secondary processing means (15 to 17) as an area 疵 signal. A surface wrinkle detecting device comprising:
[0010]
  3. Each of the light intensity detection means (3a to 3d) is a plurality of sets n for converting the luminance of the reflected light in different regions in the width direction x on the surface of the material to be inspected into electric signals; the analog / digital conversion means ( 4, 5), filter means (6a-6c), primary processing means (7a-7c), primary synthesis means, primary determination 疵 counting means (7d, 10), polarity detection means (8, 12a), area 疵 primary information The generation means (11) and the area data selection means (12b) are each a plurality of sets n addressed to the respective light intensity detection means (3a to 3d); the secondary processing means (15 to 17) A first secondary synthesizing means (15) for synthesizing the data selected by the set of area data selecting means (12b) in a series corresponding to the distribution of the different regions in the width direction x of the surface of the inspection object material; Generating area secondary information indicating whether or not the combined data is within a predetermined range (non-existent) (existing); the area input signal output means (14, 18) Rear flawsonceA second secondary synthesizing unit (14) for synthesizing the area 疵 primary information generated by the information generating unit (11) in a series corresponding to the distribution of different regions in the width direction x on the surface of the inspection object material; , A surface defect detection device, characterized by outputting a logical sum of the area information primary information and the area information secondary information as an area error signal,
[0011]
  4). In the above 1 or 2,Each of the light intensity detection means is n imaging means for converting the luminance of reflected light in different regions in the width direction x of the surface of the inspection target material into an electrical signal;The photographing means (3a-3d)RespectivelyFirst synchronizing means (22) for supplying a first synchronizing signal to the camera;RespectivelyN storage means (24a-24d) for storing electrical signals from the second synchronization means (25) for generating a second synchronization signal having a frequency n times that of the first synchronization means; the n storage means Selection means (26) for selecting the storage signals (24a-24d);And outputs an electric signal that is one-dimensional in the width direction x based on the stored signal and inputs the electric signal to the analog / digital conversion meansA surface flaw detection device characterized byThe
[0014]
DETAILED DESCRIPTION OF THE INVENTION
  1. Light intensity detecting means (3a) for converting the luminance of reflected light on the surface of the inspection target material into an electric signal and generating serially corresponding to the width direction x of the inspection target material; analog for converting the electric signal into digital data, that is, image data / Digital conversion means (4, 5); Filter means (6a to 6c) for performing processing for enhancing the luminance difference of each pixel unit of digital conversion and smoothing processing for smoothing the luminance difference on the image data; The first primary defect determination data indicating whether or not the luminance level of the image data that has been subjected to the process of emphasizing is within a predetermined range (absence), and the luminance level of the image data that has been subjected to the smoothing process Primary processing means (7a to 7c) for generating second primary wrinkle determination data representing whether or not (a) is within a predetermined range (having no wobbling); a plurality of both (5 in both the width direction x and the y direction perpendicular thereto) × 2 = 10) PixelConsist ofareaeveryOf the first and second primary wrinkle determination dataWhether each table is within the specified range“No” (疵 有)Write and patternPrimary synthesis means (9); said areaThe number of “No” (疵 existence) in the first primary wage determination data table in FIG. 5 is counted, and the logic of “No” (疵 existence) is provided for the first and second primary wage determination data tables. A sum is generated and obtained by the logical sumPrimary determination 疵 counting means (7d, 10) for counting the number of “No” (疵); polarity detecting means (8, 12a) for detecting the polarity of the peak value of the image data in the area; and the primary Based on the number counted by the judgment wrinkle counting means (7d, 10) and the polarity detected by the polarity detection means (8, 12a), it is determined whether the area is a wrinkled area or a wrinkleless area and represents this There is provided a surface flaw detection device comprising an area that generates a signal and primary information generation means (11).
[0015]
The first primary color judgment data indicating whether or not the luminance level of the image data subjected to the processing for emphasizing the luminance difference in pixel units in digital conversion is within a predetermined range (absence) or not and smoothing that smoothes the luminance difference Area for a plurality of pixels (5 × 2 = 10) by generating second primary defect determination data indicating whether the luminance level of the image data subjected to the digitization process is within a predetermined range (absence) or not (anomaly) Since the logical sum of “No” (疵 existence) of the first and second primary defect determination data is generated, the logical sum data (10 pieces) addressed to one area is subjected to defect detection (6a 1 pixel) and 疵 detection with several pixels (5 pixels for 6b and 6c), i.e., high-resolution and low-resolution 疵 detection are included. Absent. Then, the number of “No” (existence) in the area is counted, and the polarity of the peak value of the image data in the area is detected, and based on these, the area is made an exclusive area or an infinite area However, this process eliminates the noise included in the flaw detection on a pixel-by-pixel basis, and there is no flaw detection over the entire area, and the flaw presence / absence data for the entire area is highly reliable. can get. In other words, the area has a relatively rough surface division density, but the wrinkle presence / absence data reflects wrinkle detection in units of pixels in the area and is highly reliable.
[0016]
2. An area for selecting the maximum value, the minimum value, or the average value of the image data of the area based on the number counted by the primary determination wrinkle counting means (7d, 10) and the polarity detected by the polarity detection means (8, 12a) Data selection means (12b);
Secondary processing means (15-17) for generating area 疵 secondary information indicating whether the data selected by the area data selection means (12b) is within a predetermined range (疵)) or not (疵 疵);
Area 疵 signal output means for outputting a logical sum of the area 疵 primary information generated by the area 疵 information generating means (11) and the area 疵 secondary information generated by the secondary processing means (15 to 17) as an area 疵 signal (14, 18);
Is further provided.
[0017]
Based on the number of pixels primarily determined as 疵 in the area and the polarity of the peak value in the area, the maximum value, minimum value, or average value of the image data in the area is determined as a representative value, and the area 疵 is determined based on this representative value. Since secondary information is generated, this area 疵 secondary information also reflects 疵 detection in units of pixels in the area and is highly reliable. Since the logical sum of this area 次 secondary information and area と す る primary information is used as an area 疵 signal, the noise included in the 疵 detection on a pixel basis is eliminated, and there is no omission detection over the entire area. A highly reliable area signal can be obtained.
[0018]
3. Each of the light intensity detection means (3a to 3d) is a plurality of sets n for converting the luminance of the reflected light in different regions in the width direction x on the surface of the inspection target material into electrical signals;
The analog / digital conversion means (4, 5), filter means (6a to 6c), primary processing means (7a to 7c), primary synthesis means, primary determination power counting means (7d, 10), polarity detection means (8, 12a), the area 疵 primary information generation means (11), and the area data selection means (12b) each have a plurality of sets n addressed to the respective light intensity detection means (3a to 3d);
The secondary processing means (15-17) synthesizes the data selected by each set of area data selection means (12b) in a series corresponding to the distribution of the different regions in the width direction x on the surface of the inspection object material. Generating area 合成 secondary information indicating whether the synthesized data is within a predetermined range (合成)) or not (疵 疵);
The area wrinkle signal output means (14, 18) corresponds to the distribution of different areas in the width direction x of the surface of the inspection target material, the area wrinkle primary information generated by the area wrinkle information generating means (11). A surface flaw detection device comprising second secondary synthesizing means (14) that synthesizes in a row, and outputs a logical sum of the synthesized area 疵 primary information and the area 前 記 secondary information as an area 疵 signal. Provided.
[0019]
When photographing the entire width of the inspection target material (1) with a set of light intensity detection means (3a), it is necessary to separate the light intensity detection means (3a) from the inspection target material (1) in order to secure a field of view. However, there are cases where it cannot be separated due to restrictions on the installation location, or when it is separated, the installation location is susceptible to vibration, the image is blurred, the brightness of reflected light is lowered, and the resolution is lowered. Since n sets of light intensity detection means (3a to 3d) divide and share the width x direction of the material to be inspected (1), it is possible to detect wrinkles at a high resolution, and the entire width of the material to be inspected (1) Area signal is obtained.
[0020]
4). In FIG. 8, the surface of the inspection object (1) is divided into n pieces in the width x direction, and is photographed and converted into electrical signals. N photographing means (3a to 3d); the photographing means (3a to 3d) First synchronization means (22) for supplying a first synchronization signal; n storage means (24a to 24d) for storing electrical signals from the photographing means; second synchronization at a frequency n times that of the first synchronization means Second synchronizing means (25) for generating a signal; selecting means (26) for selecting storage signals of the n storage means (24a to 24d); a plurality of filter means for extracting wrinkle candidates from the selected storage signals ( 6a to 6c, 7a to 7c); concentration information calculation means (8) for calculating density information from the selected storage signal, wherein the selection means (26) is based on the second synchronization signal. By sequentially reading the stored signals of the n storage means (24a to 24d), the electric signals obtained by dividing the n imaging means (3a to 3d) are unified in the x direction. , Surface flaws detection device and performing flaw detection on the basis of the electric signal after the centralized is provided.
[0021]
The n imaging means (3a to 3d) perform imaging by sharing the width x direction of the inspection target material (1). The photographing means (3a to 3d) take pictures according to the first synchronization signal (clock signal) supplied from the first synchronization means (22). The electrical signal corresponding to the photographed image is temporarily stored in n storage means (24a to 24d). The selection means (26) is configured to store the n storage means (24a to 24d) by a second synchronization signal (clock signal) supplied by the second synchronization means (25) and having a frequency n times that of the first synchronization signal. The stored signals are sequentially read and the stored signals are aligned in the x direction. In other words, the image signals obtained by sharing and photographing in the width x direction of the inspection object material (1) are combined in the x direction so that the image signals are obtained as if they were photographed by one photographing means. Thereafter, the electric signals after the alignment are processed to detect wrinkles.
[0022]
5. In 4 above, the quality of the detection information is lowered by combining the concentration information with the candidate information corresponding to the different sizes of the defects extracted by the plurality of filter means (6a to 6c, 7a to 7c). In the surface flaw detection apparatus characterized by forming an image with a reduced amount of information without reducing the quality of the detected information, an image with a reduced amount of information can be formed.
[0023]
6). In the above 4 or 5, the predetermined corresponding to the x, y two-dimensional distribution on the surface of the inspection object material (1) from the combination of the plurality of wrinkle candidate information extracted by the plurality of filter means (6a-6c, 7a-7c). In the surface wrinkle detection apparatus characterized by determining whether there is wrinkles in the small area image, the wrinkles of the image in the predetermined small area corresponding to the x and y two-dimensional distribution on the surface of the inspection target material (1 in FIG. 1). Presence / absence determination can be performed.
[0024]
Other objects and features of the present invention will become apparent from the following description of embodiments with reference to the drawings.
[0025]
【Example】
-1st Example-
FIG. 1 shows the configuration of a surface flaw inspection apparatus according to a first embodiment of the present invention. The strip-shaped steel plate 1 to be inspected that moves in the y direction is photographed by four cameras 3a, 3b, 3c, and 3d divided into four in the x direction (width direction). The cameras 3a to 3d are one-dimensional CCD cameras and output analog image signals corresponding to the brightness of the captured image. The x direction (width direction) of the inspection target steel plate 1 is uniformly illuminated by the illumination device 2, and the four cameras 3 a, 3 b, 3 c, 3 d are directed in the direction in which this illumination light is regularly reflected on the surface of the inspection target steel plate 1. Therefore, uniform bright images can be obtained for each camera. If there is a defect portion such as a crack in the surface of the steel plate 1 to be inspected, a shadow is formed in the concave portion, and the captured image becomes dark. Therefore, a defect such as a flaw is detected based on the lightness (darkness) of the image.
[0026]
The surface image signal of the inspection target steel plate 1 photographed by the camera 3a is processed by the camera signal processing circuit 3a1. The surface scanning signals of the inspection target steel plate 1 photographed by the cameras 3b to 3d are subjected to signal processing by camera signal processing circuits 3b1 to 3d1 (not shown; similar to the camera signal processing circuit 3a1).
[0027]
Hereinafter, signal processing of the one-dimensional camera 3a will be described. The one-dimensional camera 3a supplies a signal corresponding to the brightness in the x direction of the inspection target steel plate 1 to the shading correction circuit 4 for each scan. The analog image signal from the one-dimensional camera 3 a is A / D converted and sensitivity corrected by the shading correction circuit 4. That is, the shading correction circuit 4 includes an A / D converter (8 bits), converts the analog image signal from the one-dimensional camera 3a into a digital image signal (256 gradations), and then the width direction (x Direction change) or a change in output due to a width direction sensitivity difference of a one-dimensional CCD camera or the like is corrected.
In a one-dimensional camera that outputs digital signals, a digital signal is directly input to the shading correction circuit 4, and in this case, the shading correction circuit 4 corrects the above-described correction processing.
[0028]
The synchronization circuit 5 includes one buffer (memory), and receives a digital image signal that has been subjected to A / D conversion and sensitivity correction by the shading correction circuit 4. The synchronizing circuit 5 writes a digital image signal input during the enable period (described later) into the buffer. The image signal for one scan of the one-dimensional camera 3a written in the buffer is sequentially output to the next stage by the clock signal.
[0029]
The image signal for one scan from the one-dimensional camera 3a output from the buffer of the synchronization circuit 5 is supplied to the filters 6a, 6b, 6c and the in-area density information calculation circuit 8. Different wrinkle information corresponding to the size of the wrinkles on the surface of the steel plate 1 to be inspected is extracted from the image signal for one scan by the filters 6a, 6b and 6c.
[0030]
FIG. 3 shows the configuration of the filters 6a, 6b, and 6c. The filter 6a is a digital differential filter, and corresponds to a bandpass filter high having a high frequency band in the pass region, and extracts a signal having a high frequency component (pulse-like signal; corresponding to small wrinkles). Three values obtained by multiplying the values of a total of three pixels centered on the target pixel of the serial digital image signal (256 gradations) input to the filter 6a by the coefficients +1, 0, and -1 shown in FIG. The total sum of the products is the filter pass value of the pixel of interest.
[0031]
The filter 6b is a difference filter of two different moving average filters and corresponds to a bandpass filter low having a medium frequency band in the pass region, and has a medium frequency component (a slowly changing signal; medium (Corresponding to 疵). Each of the values of a total of 5 pixels centered on the target pixel of the serial digital image signal (256 gradations) input to the filter 6b is multiplied by the coefficient 1/5 shown in FIG. A total sum is calculated, and each of the total 32 pixel values centered on the target pixel is multiplied by a coefficient 1/32 shown in FIG. 3 to calculate a total sum of the 32 products obtained. The difference between the two sums is the filter pass value of the pixel of interest.
[0032]
The filter 6c is a long-period moving average filter, corresponds to a mouth-pass filter, and extracts a signal having a low frequency component (a signal corresponding to the overall brightness; corresponding to Otsuki). Each of the values of a total of 5 pixels centered on the target pixel of the serial digital image signal (256 gradations) input to the filter 6c is multiplied by the coefficient 1/5 shown in FIG. The sum is the filter pass value of the pixel of interest.
[0033]
  The discrimination circuits 7a, 7b, and 7c determine whether or not there is a wrinkle candidate at the resolution of the original signal based on the signal from each filter. As shown in FIG. 3, the discrimination circuit 7a receives the signal from the filter 6a.ThresholdDiscrimination is made according to the value, and a ternary signal of positive polarity with wrinkles, negative polarity with wrinkles and no wrinkles is obtained. Similarly, the discrimination circuits 7b and 7c obtain ternary signals of positive polarity with wrinkles, negative polarity with wrinkles and no wrinkles based on signals from the filters 6b and 6c. A point-like signal, that is, a noise component is separated by the filter and the discrimination circuit.
[0034]
The in-area soot patterning circuit 9 writes the soot candidate signal discriminated by the discrimination circuit 7a into the first table for each area, writes the soot candidate signal discriminated by the discrimination circuit 7b into the second table for each area, and the discrimination circuit 7c. The wrinkle candidate signal discriminated by is written in the third table for each area to pattern the wrinkle candidates in the area.
[0035]
Here, the area means a range of 5 × 2 pixels. FIG. 2A shows an example of the area. For example, 5 pixels photographed in the x direction from the starting point in the mth scan of the one-dimensional camera 3a and 5 pixels photographed in the x direction from the starting point in the next mth eleventh scan are combined with 5 pixels in the x direction and the y direction. A range of 2 pixels (5 × 2 pixels) is called an “area”.
[0036]
The wrinkle candidate signal written in each table is a ternary signal of “positive polarity with wrinkles (10)”, “negative polarity with wrinkles (−)”, and “no wrinkles (0)”. Table 1 shows an example of wrinkle candidate signals written in the first table, the second table, and the third table of the in-area patterning circuit 9. In the first to third tables, wrinkle candidate patterns at the resolution of the original signal from which noise and the like are separated are written.
[0037]
[Table 1]

[0038]
The basket number counting circuit 7d counts the (+) number and (−) number of basket candidates that are discriminated by the discrimination circuit 7a and written in the first table.
[0039]
The in-area soot number calculation circuit 10 takes OR (logical sum) of the number of soot candidates written in each table of the in-area patterning circuit 9 and calculates the total number of soot elements (total number in logical sum). . That is, the OR (logical sum) of the (+) number and (−) number which are candidates for the first table to the third table is obtained.
[0040]
In the table of the number of cells in Table 1, the logical sum of the cell candidates (+) or the cell candidates (−) in the first to third tables, that is, whether the pixel is (+) or (−) in the first to third tables. For example, the pixel in the bottle number table is represented by (×), and if the pixels in the first to third tables are all (0), the pixel in the bottle number table is represented as (0). The total number 6 of (x) is the total number of heel elements. As a result, the wrinkle pattern information in the first to third tables is compressed into one piece of information (10 pixels / 1 area to 1 pixel).
[0041]
The in-area density information calculation circuit 8 calculates the maximum value, minimum value, and average value in the area of the image signal output from the buffer of the synchronization circuit 5. That is, the density information of the signal before filtering is calculated. The area polarity detection circuit 12a generates a difference S + (meaning a bright signal) between the maximum value and the average value and a difference S- (meaning a dark signal) between the average value and the minimum value.
[0042]
The in-area wrinkle presence / absence determination circuit 11 includes the number of wrinkle candidates counted by the wrinkle number counting circuit 7d, the total number of wrinkle elements calculated by the in-area wrinkle number calculation circuit 10, and the maximum and average values from the area polarity detection circuit 12a. It is determined whether or not the area is wrinkled based on the difference S + and the difference S− between the average value and the minimum value.
[0043]
The in-area representative selection circuit 12b is the polarity determined by the area polarity detection circuit 12a (S + if S + ≧ S−; S− if S +> S +), and the maximum value of the image signal calculated by the in-area density information calculation circuit 8. From the minimum value, the average value, the number of wrinkle candidates counted by the wrinkle number counting circuit 7d, and the total number of wrinkle pixels calculated by the in-area wrinkle number calculation circuit 10, the selection of the representative signal in the area, that is, the maximum value is represented in the area. It is selected whether to select as a signal, to select a minimum value, or to select an average value. In this way, the circuits 10, 11, and 12b process the signals by compressing them in area units. The table shows the determination of presence / absence of defects within the area and the selection of representative signals within the area.
[0044]
[Table 2]

[0045]
In the first row of the table, for example, the number of soots extracted by the filter 6a is 0 (discrimination; 7a, count; 7d), but the logical sum of the soots extracted by the filters 6a, 6b, 6c is obtained. When the number of 疵 (OR operation and number calculation; 10) is 1, the signal A is 1, and it is determined that 疵 is “present” (determination: 11) in the area, and 疵 is “present”. It becomes. When the polarity signal is S + (calculation; 12a), the signal C is the maximum value in the area as a representative value, and when the polarity signal is S− (calculation; 12a), the signal is designated as a representative in the area. C uses the minimum value.
[0046]
In the above embodiment, the example in which the area representative signal is determined by directly comparing the difference S + between the maximum value and the average value and the difference S− between the average value and the minimum value is described. However, even if the magnitudes of S + and S− are the same, the harmfulness of sputum may differ, so that S + and S− may not be directly compared but may be given priority by weighting. For example, when it is desired to give priority to the signal with the minimum value by weighting, the signal values of S + and S− are compared with a difference corresponding to the weight, and the maximum value, the minimum value, and the average value are selected.
[0047]
Select the comparison signal selection criteria
f (S +, S −) = S + − (S− + α)
α: Offset, α can be positive or negative
When f (S +, S-)> 0 S +
When f (S +, S−) ≦ 0 S−
By adding a weight, it is possible to select a representative signal within the representative in accordance with the characteristics of the eyelid, such as when the maximum value increases or the minimum value decreases even at the same degree of harm.
[0048]
The signal A of the total number of bags (logical number calculation; 10), the signal B of the presence / absence (determination; 11), and the signal C of the in-area representative signal (selection; 12b) are In the wrinkle presence / absence map two-dimensional synthesis circuit 14 and the grayscale image two-dimensional synthesis circuit 15, together with signals from the other cameras 3b, 3c, and 3d, a signal corresponding to the full width (x direction) of the inspection target steel plate 1 is used. At the same time, the two-dimensional images for one screen in the xy direction are synthesized by accumulating in the length direction (y direction) of the steel plate 1 to be inspected.
[0049]
FIG. 2B schematically shows an example of a combined 1 screen of the cameras 3a to 3d. FIG. 2B shows the area shown in FIG. 2A in a reduced size, and the minimum one section surrounded by a dotted line is one pixel. The horizontal width (x-direction width) of the screen is divided into four, and the imaging areas are shared in the order of the one-dimensional cameras 3a, 3b, 3c, and 3d from the left. For example, the one-dimensional camera 3a shares 25 pixels / 5 area in the x direction (from the left side of the screen to the horizontal direction) and 8 pixels / 4 area in the y direction (vertical direction). One screen shot by the four cameras, that is, a shooting screen before image compression includes image signals corresponding to lines, planes, and dots as shown in the figure, for example.
[0050]
FIG. 2C shows a state where the area of 5 pixels in the horizontal direction and 2 pixels in the vertical direction shown in FIG. 2B is reduced (compressed) to the area of 1 pixel in the horizontal direction and 1 pixel in the vertical direction. That is, a state in which the image is compressed from 10 pixels / 1 area to 1 pixel by the in-area number calculation circuit 10 is shown as a combined 1 screen. The image signals corresponding to the lines, planes, and points shown in FIG. 2B are displayed with the linear signal determined as 疵 in FIG. 2C, and the planar signal is also determined as 疵. However, the dot-like signal is not judged to be 疵 and is not displayed.
[0051]
The image filter 16 is a spatial filter, and extracts two-dimensional wrinkle information from the two-dimensional distribution of light and shade from the light and shade image two-dimensionalization synthesis circuit 15. The table shows three image filter coefficient matrices in the x direction and seven in the y direction.
[0052]
[Table 3]

[0053]
A grayscale two-dimensional distribution centered on the pixel of interest from the grayscale image two-dimensionalization synthesis circuit 15 input to the image filter 16 (maximum value, minimum value in three areas in the x direction and seven areas in the y direction, Multiply 3 in the x direction (two-dimensional distribution of mean values) by the coefficient matrix 1, 0, -1 shown in Table 3, and sequentially perform this in 7 in the y direction. This is the filter pass value of the pixel. Larger but lower concentration can be extracted. The discrimination circuit 17 simply binarizes the positive, negative, and none ternary signals from the image filter 16 and determines the presence or absence of wrinkle candidates.
[0054]
The synthesizing circuit 18 synthesizes the heel candidate signal (large cocoon information) from the discrimination circuit 17 and the heel presence / absence map signal (small heel information) from the heel presence / absence map two-dimensional synthesis circuit 14 by OR logic, and A two-dimensional image is obtained.
[0055]
The haze feature detection circuit 19 calculates the feature amount of the haze signal based on the signal from the synthesis circuit 18, the signal from the hull number two-dimensional synthesis circuit 13, and the signal from the grayscale image two-dimensional synthesis circuit 15. .
[0056]
FIG. 4 shows a processing procedure of the wrinkle feature detection circuit 19. The wrinkle presence / absence map two-dimensional signal (1; corresponding to the processing procedure; the same applies hereinafter) from the synthesis circuit 18 is labeled (2) by two-dimensional scanning (x and y directions) (2), for example, as shown in FIG. , Probe 2 and probe 3 are detected (3). The area of each probe is identified by a threshold value (reference value), and the probe below the reference value is removed as noise (4), and a cocoon candidate is determined (5). For example, as shown in the figure, the probe 3 is removed as noise. The soot feature detection circuit 19 calculates the feature amount of the soot signal based on the following parameters.
[0057]
“Shape parameters”: width, length, area, slope, aspect ratio
“Density parameter”: average density, maximum density, minimum density, dispersion of density
“Pixel number in pixels” parameter: Ratio of n or less in n pixels after compression (one area in this area) (small size)
The calculation of the wrinkle shape feature amount based on the shape parameter is performed for each determined wrinkle candidate (6). The calculation of the shade density feature amount based on the shade parameter is performed for each tile candidate by using the grayscale image two-dimensional signal corresponding to the tile candidate (7). The calculation of the cocoon number feature based on the cocoon number parameter is performed for each candy candidate using the cocoon number two-dimensional signal corresponding to the candy candidate (8).
[0058]
The cocoon type / harmfulness determination circuit 20 determines the name of the cocoon and the degree of harmfulness for each candy candidate for each screen using the above-mentioned moth characteristic amount. One screen in this embodiment (x-direction width 4000 pixels, y-direction 500 pixels) corresponds to a range of the inspected steel sheet 1 in the x-direction width 2000 mm and the y-direction length 500 mm.
[0059]
Species / harmfulness judgment is a judgment part, that is,
"IF THEN Rule"
"Tree-like rule"
"Table-like rules"
"neural network"
Etc. are combined in a multilayer form, or a combination thereof.
FIG. 5 schematically shows an example of the determination part. That is, (a) shows an example of a tree-like rule, (b) shows an example of a table-like rule, and (c) shows an example of a neural network.
[0060]
FIG. 6 shows an example in which “tree-like rules” are combined in multiple layers to execute the sow seed / harm level determination. The soot species determination is repeated in multiple layers to obtain the soot type and the degree of harm.
[0061]
FIG. 7 shows a configuration of a portion related to signal synchronization of an image synthesis system that synthesizes images captured by four one-dimensional cameras 3a, 3b, 3c, and 3d into one image. Clock signals from the camera controller 22 are converted into cameras 3a to 3d, shading correction / A / D conversion circuits 4a to 4d, synchronization circuits 5a to 5d, signal processing circuits 21a to 21d (also shown in FIG. 1), and synthesis circuits 13 to 15 To supply. The four cameras 3a, 3b, 3c, and 3d perform imaging (scanning) simultaneously in the x direction (width direction) of the inspection target steel sheet 1 based on the clock signal, and repeat this. Each of the shading correction circuits 4a, 4b, 4c, and 4d includes an A / D converter, converts analog image signals from the cameras 3a to 3d into digital image signals, and then in the width direction (x direction) of the inspection target steel sheet 1. The output change due to the brightness difference or the width direction sensitivity difference of the one-dimensional CCD camera is corrected.
[0062]
The movement distance detector 23 is an encoder that generates a pulse every time the steel plate 1 to be inspected moves by a certain distance (for example, 1 mm), and is connected to the synchronization circuits 5a to 5d, the signal processing circuits 21a to 21d, and the synthesis circuits 13 to 15. A distance signal LT is supplied. The synchronization circuits 5a to 5d create an enable (gate) signal based on the distance signal LT, and the signal processing circuits 21a to 21d and the synthesis circuits 13 to 15 similarly create gate signals.
[0063]
The synchronization circuit 5a includes a buffer, and the buffer stores a digital image signal from the shading correction circuit 4a input during the enable period. Similarly, the buffers of the synchronization circuits 5b to 5d also store the digital image signals from the shading correction circuits 4b to 4d inputted in the enable period in the same manner. The cameras 3a to 3d repeat scanning based on the clock signal, but the image signals are not stored in the respective buffers unless they are enabled. That is, each image signal is stored only when the distance signal LT is supplied. The image signals corresponding to each camera stored in this manner are synthesized in the cocoon number two-dimensional synthesis circuit 13, the heel presence / absence map two-dimensional synthesis circuit 14, and the gray image two-dimensional synthesis circuit 15, and the inspection target steel plate 1 Are integrated into a signal corresponding to the full width (x direction) of the image, and the signal is integrated in the y direction to form a two-dimensional image for one screen in the xy direction.
[0064]
As described above, unnecessary information is collected by preprocessing and compressed while leaving necessary information, so that the amount of information can be reduced without deteriorating the quality of detected information. For this reason, it is possible to process surface defects with high accuracy at high speed by online processing, and the inspection efficiency is improved. In addition, in this embodiment, since wrinkle information including density information (halftone) can be provided, the inspector's wrinkle inspection ability is improved.
[0065]
-Second Example-
FIG. 8 shows the configuration of the surface flaw detection apparatus according to the second embodiment of the present invention. In the present embodiment, the strip-shaped steel plate 1 to be inspected that moves in the y direction is photographed by dividing the width direction (x direction) into four by four CCD cameras 3a, 3b, 3c, and 3d. The lighting device 2 uniformly illuminates the width direction (x direction) of the inspection target steel plate 1, and the four cameras 3a, 3b, 3c, and 3d are directed in the direction in which this illumination light is regularly reflected on the surface of the inspection target steel plate 1. Since the fields of view are matched, uniform bright images can be obtained for each camera.
[0066]
FIG. 9 is a timing chart of the four CCD cameras 3a, 3b, 3c, and 3d of the surface flaw detection apparatus shown in FIG.
[0067]
Analog image signals from the one-dimensional cameras 3a to 3d are digitally converted by A / D converters included in the shading correction circuits 4a to 4d, and then sensitivity correction is performed. That is, the output change due to the brightness difference in the width direction (x direction) of the steel plate 1 to be inspected or the sensitivity difference in the width direction of the CCD camera is corrected. The cameras 3 a to 3 d simultaneously photograph the inspection target steel plate 1 by ¼ plate width according to the first clock signal having the frequency f supplied from the camera controller 22. One-dimensional image signals from the cameras 3a to 3d are stored in the image memories 24a to 24d for one scanning line and for each pixel via the shading correction circuits 4a to 4d, and the stored contents are rewritten for each scanning.
[0068]
The movement distance detector 23 is an encoder that generates a pulse corresponding to a certain distance movement (y direction) of the steel plate 1 to be inspected, and supplies a distance signal LT to the read timing circuit 25, the synchronization circuit 5, and the signal processing circuit 21a. . The read timing circuit 25 generates an enable (gate) signal based on the distance signal LT, and the synchronization circuit 5 and the signal processing circuit 21a similarly generate a gate signal.
[0069]
During the enabling period, the readout timing circuit 25 creates a second clock signal having a frequency 2f by doubling the frequency f of the first clock signal used by the cameras 3a to 3d supplied from the camera controller 22 for photographing. The data is supplied to the memories 24 a to 24 d and the selection circuit 26. The image signal (CH 1) captured by the camera 3 a and stored in the image memory 24 a and the image signal (CH 2) captured by the camera 3 b and stored in the image memory 24 b are the second of the frequency 2 f supplied from the timing circuit 25. The selection circuit 26 integrates CH1 and CH2 in that order and stores them in the first buffer of the selection circuit 26. Similarly, during the enabling period, the image signal (CH3) stored in the image memory 24c and the image signal (CH4) captured by the camera 3d and stored in the image memory 24d are supplied with the frequency 2f supplied from the timing circuit 25. The selection circuit 26 integrates CH3 and CH4 in that order and stores them in the second buffer of the selection circuit 26.
[0070]
The read timing circuit 25 creates a third clock signal having a frequency 4f, which is four times the frequency f of the first clock signal used for photographing by the cameras 3a to 3d supplied from the camera controller 22, and supplies the third clock signal to the selection circuit 26. To do.
[0071]
The selection circuit 26 reads and integrates the CH1 and 2 integrated signals stored in the first buffer and the CH3 and 4 integrated signals stored in the second buffer with the third clock signal having the frequency 4f. That is, the image signals are read in the order of the image memory 24 a, the image memory 24 b, the image memory 24 c, and the image memory 24 d (in the order of CH 1, CH 2, CH 3, CH 4), integrated (unified), and supplied to the synchronization circuit 5. At the stage of input to the synchronization circuit 5, four camera signals are already integrated into one line of serial data. Accordingly, there is only one signal processing circuit (21a), and only one input is provided to the number-of-haze two-dimensionalization circuit 13, the haze presence / absence map two-dimensionalization synthesis circuit 14, and the grayscale image two-dimensionalization synthesis circuit. This circuit synthesizes two-dimensional images for one screen in the xy direction only by accumulating the respective input signals in the y direction. Since signal processing other than the above description is the same as that of the first embodiment, description thereof is omitted.
[0072]
The inspection target steel sheet 1 is divided into four in the x direction (width direction) by four CCD cameras 3a, 3b, 3c, and 3d, photographed at the same time, and stored in the image memories 24a to 24d. By performing integration (unification) with a double clock signal, signal processing is performed as if the entire width of the steel plate 1 to be inspected was captured with a single camera, so transmission costs and signal processing costs can be reduced.
[0073]
【The invention's effect】
As described above, according to the present invention, it is possible to provide soot information with little online quality deterioration, simplify detection signal processing, and reduce the cost of the soot detection device.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention.
2A is a plan view showing pixels and areas on two scanning lines scanned by one one-dimensional camera shown in FIG. 1, and FIG. 2B is four one-dimensional images. It is a top view which shows an example of the screen which a camera image | photographs, FIG.2 (c) is a top view which shows the state which processed and compressed the signal of FIG.2 (b).
3 is a block diagram showing a configuration of filters 6a to 6c and discrimination circuits 7a to 7c shown in FIG.
FIG. 4 is a block diagram showing a processing procedure of a wrinkle feature detection circuit 19;
5A and 5B show an example of a judgment tool for hail feature detection, FIG. 5A is a block diagram showing a tree-like rule, and FIG. 5B is a diagram showing a table-like rule. (C) is a block diagram showing a neural network.
FIG. 6 is a block diagram showing an example of soot species / harmfulness determination using a tree-like rule;
7 is a block diagram showing a circuit configuration of a part that performs image composition processing on the image captured by the four one-dimensional cameras shown in FIG. 1; FIG.
FIG. 8 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.
FIG. 9 is a timing chart when image synthesis processing is performed on the images captured by the four one-dimensional cameras shown in FIG. 8 on one screen;
[Explanation of symbols]
1: Steel plate to be inspected
2: Lighting device
3a to 3d: One-dimensional CCD camera
3a1-3d1: Camera signal processing circuit
4a to 4d: Shading correction circuit (including A / D converter)
5: Synchronous circuit
6a, 6b, 6c: filter circuit
7a, 7b, 7c: Discrimination circuit
7d: 疵 number counting circuit
8: Area density information calculation circuit
9: Inner area patterning circuit
10: Circuit for calculating the number of tiles in the area
11: Inner area presence / absence determination circuit
12a: Area polarity detection circuit
12b: Area data selection circuit
13: Two-dimensional composition circuit
14: Wrinkle presence / absence map two-dimensional synthesis circuit
15: Gray image two-dimensional synthesis circuit
16: Image filter
17: Discrimination circuit
18: Synthesis circuit
19: Haze feature detection circuit
20: Species / harmfulness judgment circuit
21a to 21d: signal processing circuit (collecting 6 to 12)
22: Camera controller
23: Moving distance detector
24a to 25d: Memory
26: Selection circuit

Claims (4)

Light intensity detection means for converting the brightness of reflected light on the surface of the inspection target material into an electrical signal and generating serially corresponding to the width direction x of the inspection target material;
Analog / digital conversion means for converting the electrical signal into digital data, that is, image data;
A first filter means for performing a process of enhancing a luminance difference in pixel units of digital conversion on the image data; and a second filter means for performing a smoothing process for smoothing the luminance difference;
The first primary eye determination data indicating whether or not the luminance level in the pixel unit of the image data subjected to the process of enhancing the luminance difference is within a predetermined range, and the pixel unit of the image data subjected to the smoothing process Primary processing means for generating second primary wrinkle determination data indicating whether or not the luminance level is within a predetermined range;
Of each area consisting of the width direction x and both a plurality of pixels in the y direction orthogonal thereto, said first and second primary defect determination data for each table, patterned write the "No" or within a predetermined range Primary synthesis means;
Count the number of “No” in the first primary wage determination data table in the area, and generate a logical sum of “No” for the first and second primary hail determination data tables, Primary determination 疵 counting means for counting the number of “no” obtained by logical sum ;
A difference S1 between the maximum value and the average value in the area of the image data obtained by the analog / digital conversion means, and a difference S2 between the average value and the minimum value are generated, and the magnitudes of the differences S1 and S2 are calculated. Polarity detection means for detecting the polarity to be represented;
Based on the number counted by the primary determination counter and the polarity detected by the polarity detection means, the area is determined to be a non-existent area or a non-existent area, and a signal representing this is generated. Generating means;
And area data selection means for selecting a maximum value, a minimum value, or an average value of the image data of the area based on the number counted by the primary determination counter and the polarity detected by the polarity detection means. Surface wrinkle detection device. Further, secondary processing means for generating area 疵 secondary information indicating whether the data selected by the area data selection means is within a predetermined range;
An area 疵 signal output means for outputting, as an area 疵 signal, a logical sum of the area 生成 primary information generated by the area 疵primary information generating means and the area 疵 secondary information generated by the secondary processing means. The surface flaw detection device according to claim 1. Each of the light intensity detection means is a plurality of sets n for converting the luminance of reflected light in different regions in the width direction x of the surface of the inspection target material into electrical signals,
The analog / digital conversion means, filter means, primary processing means, primary synthesis means, primary determination / counting means, polarity detection means, area / primary information generation means, and area data selection means are each light intensity detection means. Multiple sets of n addresses,
The secondary processing means is a first secondary synthesizing means for synthesizing the data selected by each set of area data selecting means in a series corresponding to the distribution of regions having different width directions x on the surface of the inspection object material. , And generate area 次 secondary information indicating whether the synthesized data is within a predetermined range,
The area 疵 signal output means synthesizes the area 疵primary information generated by the area 生成primary information generation means in a series corresponding to the distribution of different regions in the width direction x on the surface of the inspection target material. 3. The surface wrinkle detection apparatus according to claim 2, further comprising a secondary combining means, wherein a logical sum of the combined area w / primary information and the area w / secondary information is output as an area w signal. Each of the light intensity detection means is n imaging means for converting the luminance of reflected light in different regions in the width direction x of the surface of the inspection target material into electrical signals,
And first synchronization means for supplying a first synchronization signal to each of the photographing means;
N storage means for storing electrical signals from each of the photographing means;
Second synchronizing means for generating a second synchronizing signal having a frequency n times that of the first synchronizing means;
Selecting means for selecting storage signals of the n storage means ,
The surface wrinkle detection device according to claim 1 or 2, wherein an electric signal that is one-dimensional in the width direction is output based on the stored signal and is input to the analog / digital conversion means .

Images