JP2005156334A - Pseudo defective image automatic creation device and imaging inspection device - Google Patents
Pseudo defective image automatic creation device and imaging inspection device Download PDFInfo
- Publication number
- JP2005156334A JP2005156334A JP2003394788A JP2003394788A JP2005156334A JP 2005156334 A JP2005156334 A JP 2005156334A JP 2003394788 A JP2003394788 A JP 2003394788A JP 2003394788 A JP2003394788 A JP 2003394788A JP 2005156334 A JP2005156334 A JP 2005156334A
- Authority
- JP
- Japan
- Prior art keywords
- pseudo
- defective
- image
- defective image
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
- 230000002950 deficient Effects 0.000 title claims abstract description 126
- 238000007689 inspection Methods 0.000 title claims abstract description 27
- 238000003384 imaging method Methods 0.000 title 1
- 238000013528 artificial neural network Methods 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 230000002194 synthesizing effect Effects 0.000 claims abstract description 5
- 230000009467 reduction Effects 0.000 claims description 22
- 230000008859 change Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 57
- 230000013016 learning Effects 0.000 abstract description 20
- 230000015572 biosynthetic process Effects 0.000 abstract description 2
- 238000003786 synthesis reaction Methods 0.000 abstract description 2
- 238000000034 method Methods 0.000 description 18
- 230000008569 process Effects 0.000 description 11
- 238000012545 processing Methods 0.000 description 9
- 238000004519 manufacturing process Methods 0.000 description 6
- 238000001514 detection method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000007781 pre-processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000001419 dependent effect Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
- 230000004304 visual acuity Effects 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Landscapes
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
Description
本発明は、画像情報をニューラルネットワーク（Neural Network）の入力とし、その出力結果からその画像に写る製品（被検体）の良否を判定する画像検査装置において、初期学習時に必要とする不良画像を擬似的に作成する場合に、自動的且つ大量に作成する疑似不良画像自動作成装置、及び前記疑似不良画像自動作成装置を具備する画像検査装置に関する。 According to the present invention, an image inspection apparatus that uses image information as an input to a neural network and determines the quality of a product (subject) reflected in the image from the output result is used to simulate a defective image required during initial learning. The present invention relates to an automatic pseudo-defective image creation device that automatically creates a large number of images when automatically creating the image, and an image inspection device including the pseudo-defective image automatic creation device.
かつて、工業製品等の検査は、作業者の目視による検査が主流であったが、このような作業方法は、作業者の経験に依存する部分が大きかった。そのため、特に、作業者が検査の初心者である場合などは、不良品を識別することが困難となり、その結果として、粗悪品を流通させてしまうこととなった。また、十分な経験を有する作業者でも、検査環境が変化する、例えば、作業者が長時間の労働を行うことにより、視力等が低下する等の体調の変化により、工業製品の品質を維持することができなくなる可能性もあった。
このような中で、近年、教師付き学習機能を有するニューラルネットワーク（或いは、ニューロ）を用いることにより、被検体が良品であるか、不良品であるかを検査する検査装置が使われている。ニューラルネットワークを用いた検査装置は、予め、良品サンプルデータと不良品サンプルデータをニューラルネットワークに入力し、学習させることにより、被検体が良品であるか否かを判定している。
In the past, inspection of industrial products or the like was mainly performed by visual inspection by workers, but such a work method largely depends on the experience of the workers. Therefore, in particular, when the worker is a beginner of inspection, it becomes difficult to identify defective products, and as a result, poor products are distributed. Also, even for workers with sufficient experience, the quality of industrial products is maintained by changes in the physical condition such as changes in the inspection environment, for example, visual acuity etc. due to workers working long hours There was also a possibility that it could not be done.
Under such circumstances, in recent years, an inspection apparatus for inspecting whether a subject is a non-defective product or a defective product by using a neural network (or a neuro) having a supervised learning function has been used. An inspection apparatus using a neural network inputs non-defective product sample data and defective product sample data into the neural network in advance and determines whether or not the subject is non-defective.
このような従来の発明として、例えば、予め作成された疑似的な異常音響データをニューラルネットワークに入力して、その学習を行わせて、設備における音響及び振動の周波数波形に基づく認識をして診断を行うことを特徴とする設備における音響及び振動の診断方法に関する発明がある（特許文献１参照）。 As such a conventional invention, for example, pseudo abnormal acoustic data prepared in advance is input to a neural network, and learning is performed, and diagnosis is performed based on recognition based on frequency waveforms of sound and vibration in equipment. There is an invention relating to a method for diagnosing sound and vibration in equipment that is characterized by performing (see Patent Document 1).
また、例えば、ニューラルネットワークによる話者認識システムを構成するに際し、学習用音声サンプルの数を多くすることなく、変動が加わった入力パターンに対しても正しい認識ができるようにすることが可能な話者認識システムに関する発明がある（特許文献２参照）。 In addition, for example, when configuring a speaker recognition system using a neural network, it is possible to correctly recognize an input pattern with fluctuations without increasing the number of learning speech samples. There is an invention related to a person recognition system (see Patent Document 2).
また、被検体の欠陥種別を自動的に判別でき、検査結果を迅速且つ正確にフィードバックすることが可能な欠陥種別判別装置、及び欠陥種別の判別結果に基づいて、被検体の製造プロセスラインの管理を行うプロセス管理システムに関する発明がある（特許文献３参照）。
ニューラルネットワークを用いた検査装置の良否判定の精度に関する課題は、今なお引き継がれている課題であり、様々な方法により、この問題を解決している。
一般的に、ニューラルネットワークを用いた検査装置の良否判定の精度は、ニューラルネットワークの学習数に依存する部分が大きいため、ニューラルネットワークにサンプルを多く入力し、学習させた方が、検査装置の判定精度はそれだけ高いものとなる。
しかしながら、実際には、工業製品等の製造技術は非常に高精度であり、不良品が製造されることは非常に稀である。また、実稼働している製造工程の一部を改悪して、故意に不良品を製造することにより不良品のサンプルを得る方法は、十分な不良品サンプルを得ることができないばかりでなく、製造ライン全体にも影響するので、非現実的な方法である。
そのため、実在する不良品のデータではなく、擬似的な不良品データをニューラルネットワークに入力し、学習させることにより、検査装置の判定精度を向上させる必要がある。或いは、少ない不良品データでも検査装置の判定精度を向上させる技術が必要である。
The problem regarding the accuracy of the quality determination of the inspection apparatus using the neural network is still a problem that has been handed over, and this problem is solved by various methods.
In general, the accuracy of pass / fail judgment of an inspection device using a neural network is largely dependent on the number of learnings in the neural network. Therefore, it is better to input more samples into the neural network and learn it. The accuracy is higher.
However, in practice, manufacturing techniques for industrial products and the like are very accurate, and it is very rare that defective products are manufactured. Moreover, the method of obtaining a defective sample by deliberately manufacturing a defective product by altering a part of the production process that is actually in operation can not only obtain a defective sample, but also manufacture Since it affects the entire line, it is an unrealistic method.
Therefore, it is necessary to improve the determination accuracy of the inspection apparatus by inputting and learning pseudo defective product data, not actual defective product data, to the neural network. Alternatively, a technique for improving the determination accuracy of the inspection apparatus even with a small amount of defective product data is required.
このような問題を解決するために、例えば、特許文献２に示す話者認識システムに関する発明は、計算によって学習パターンに雑音を重畳し、擬似的にパターン数を増やすに際し、加える雑音の大きさの上限を、実際の音声サンプルによる学習パターンから算出した標準偏差を基準に設定する手段を有することにより、学習用音声サンプルの数を多くすることなく、変動が加わった入力音声に対しても、正しい認識ができるようにしている。 In order to solve such a problem, for example, the invention relating to the speaker recognition system disclosed in Patent Document 2 superimposes noise on a learning pattern by calculation, and increases the number of patterns in a pseudo manner. By having a means to set the upper limit on the basis of the standard deviation calculated from the learning pattern of actual speech samples, it is correct even for input speech that has been changed without increasing the number of training speech samples. It can be recognized.
また、特許文献３に示す欠陥種別判別装置及びプロセス管理システムに関する発明は、被検体の欠陥検査で発見された個々の欠陥の種別を判定する欠陥種別判定装置において、被検体表面の欠陥から得られる情報、欠陥検出光学系の種類及び欠陥検出の画像処理方式等からなる欠陥情報と、そのような欠陥情報となる欠陥名称を表した教師信号とで学習させたニューロ処理ユニットを備えたことにより、欠陥部の欠陥情報から欠陥の種別を自動的に判別することが可能となり、欠陥の発生した製造プロセスへの迅速なフィードバックが可能となる。その結果として、発生する欠陥を最小限に抑えることが可能となる。しかも、熟練した作業者であっても、判別が困難な多数の測定結果及び測定条件を設定することができるので、再現性の高い欠陥種別判定結果を得ることが可能となる。また、ニューロ処理ユニットに、人間でしか識別することができない疑似欠陥を学習させたので、疑似欠陥の自動判定が可能となった。 Further, the invention relating to the defect type discriminating apparatus and the process management system disclosed in Patent Document 3 is obtained from the defect on the surface of the object in the defect type determining apparatus that determines the type of each defect found in the defect inspection of the object. By providing a neuro processing unit trained with information, defect information consisting of defect detection optical system type and defect detection image processing method, etc., and a teacher signal representing the defect name as such defect information, It becomes possible to automatically determine the type of the defect from the defect information of the defective portion, and it is possible to provide quick feedback to the manufacturing process in which the defect has occurred. As a result, the generated defects can be minimized. In addition, even a skilled worker can set a large number of measurement results and measurement conditions that are difficult to discriminate, so that a highly reproducible defect type determination result can be obtained. Further, since the neuro processing unit is made to learn a pseudo defect that can be identified only by humans, it is possible to automatically determine the pseudo defect.
しかしながら、上記に示した発明は、例えば、特許文献２で示した発明は、擬似的にパターン数を増やすに際し、加える雑音の大きさの上限を、実際の音声サンプルによる学習パターンから算出した標準偏差を基準に設定することから、ある程度の音声サンプルをニューラルネットワークに投入しておく必要があるといった問題がある。
つまり、２つ以上の不良音声サンプルがなければ、標準偏差を求めることができず、また、現実的に有効な疑似不良音声サンプルを作成するためには、高々数個のデータではなく、ある程度の数のデータを入力する必要がある。
However, in the invention shown above, for example, in the invention shown in Patent Document 2, when increasing the number of patterns in a pseudo manner, the upper limit of the amount of noise to be added is a standard deviation calculated from a learning pattern based on an actual speech sample. Therefore, there is a problem that it is necessary to input a certain amount of audio samples to the neural network.
In other words, without two or more bad voice samples, the standard deviation cannot be obtained, and in order to create a practically effective pseudo bad voice sample, not a few data at all, You need to enter a number of data.
また、例えば、特許文献３で示した発明は、ニューロ処理ユニット自体に、被検体表面の欠陥から得られる情報、欠陥検出光学系の種類及び欠陥検出の画像処理方式等からなる欠陥情報と、そのような欠陥情報となる欠陥名称を表した教師信号とで学習させることにより、ニューラルネットワーク部の構成が複雑となってしまう。そのため、装置内部の各機構は互いに依存性が強いものとなり、装置を改良する必要性が発生した場合でも、その改良は困難なものとなってしまう可能性があるので、装置内部の各機構は、できる限り相互依存性が少ない方が好ましい。 In addition, for example, the invention shown in Patent Document 3 includes, in the neuro processing unit itself, defect information including information obtained from defects on the surface of the object, the type of defect detection optical system, the image processing method for defect detection, and the like. Learning with a teacher signal representing a defect name as such defect information complicates the configuration of the neural network unit. Therefore, each mechanism inside the device is strongly dependent on each other, and even if the need to improve the device arises, it may be difficult to improve, so each mechanism inside the device It is preferable to have as little interdependence as possible.
本発明は上記事情を鑑みてなされたものであり、画像情報をニューラルネットワークの入力とし、その出力結果からその画像の写る製品の良否を判定する画像検査装置において、初期学習時に必要とする不良画像を擬似的に作成する場合に、自動的且つ大量に作成することが可能な疑似不良画像自動作成装置、及び前記疑似不良画像自動作成装置を具備する画像検査装置を提供することを目的とする。 The present invention has been made in view of the above circumstances. In an image inspection apparatus that uses image information as input to a neural network and determines the quality of a product in which the image is captured based on the output result, a defective image required at the time of initial learning. It is an object of the present invention to provide an automatic pseudo-defective image creating apparatus capable of creating a large number of images automatically and in large quantities, and an image inspection apparatus including the pseudo-defective image automatic creating apparatus.
前記課題を解決するために、請求項１記載の発明は、被検体の良品画像を入力する良品画像入力手段と、被検体の不良画像を入力する不良品画像入力手段と、被検体の不良画像から良品との差分データを抽出する不良画像抽出手段と、前記良品画像入力手段から入力された良品画像と前記差分データとを合成し、疑似不良画像を作成する疑似不良画像作成手段とを有することを特徴とする。 In order to solve the above problems, the invention described in claim 1 is a non-defective image input means for inputting a non-defective image of a subject, a non-defective image input means for inputting a defective image of the subject, and a defective image of the subject. A defect image extracting means for extracting difference data from a non-defective product from the non-defective product, and a pseudo-defective image creating means for synthesizing the non-defective image input from the good image input means and the difference data to create a pseudo-defective image. It is characterized by.
請求項２記載の発明は、前記擬似不良画像作成手段は、乱数値を発生させる乱数発生手段と、良品画像と差分データを合成する際に、前記乱数発生手段にて発生させた乱数値を組み合わせて合成するための条件を設定する擬似データ作成条件設定手段とを有することを特徴とする。 The invention according to claim 2 is characterized in that the pseudo-defective image creating means combines a random number generating means for generating a random value and a random value generated by the random number generating means when the good image and the difference data are combined. And pseudo-data creation condition setting means for setting conditions for combining them.
請求項３記載の発明は、前記擬似データ作成条件設定手段は、擬似不良画像の作成個数を設定する作成個数設定手段を有することを特徴とする。 The invention according to claim 3 is characterized in that the pseudo data creation condition setting means has a creation number setting means for setting the number of creation of pseudo defective images.
請求項４記載の発明は、前記不良画像抽出手段は、被検体の不良画像から、仮想的に良品の状態を算出する仮想良品状態算出手段と、不良画像と前記仮想良品状態算出手段にて求められた仮想的な良品の状態との差から差分データを抽出する第２の不良画像抽出手段とを有することを特徴とする。 According to a fourth aspect of the present invention, the defective image extraction means is obtained by a virtual non-defective product state calculating unit that virtually calculates a good product state from a defective image of a subject, a defective image and the virtual non-defective product state calculating unit. And a second defective image extracting means for extracting difference data from a difference from the virtual non-defective product state.
請求項５記載の発明は、前記擬似不良画像作成手段は、前記差分データを合成する際に、合成することが不可能である書き込み禁止範囲を設定する書き込み禁止範囲設定手段と、前記差分データを拡大もしくは縮小して合成するために、拡大・縮小率を設定する拡大・縮小率設定手段と、前記差分データの明るさを変更して合成するために、明るさを設定する明るさ設定手段と、前記書き込み禁止範囲設定手段で設定した書き込み禁止範囲には差分データの合成を行わない第１の合成手段と、前記拡大・縮小率設定手段で設定した拡大・縮小率で差分データの合成を行う第２の合成手段と、前記明るさ設定手段で設定した明るさで差分データの合成を行う第３の合成手段とを有することを特徴とする。 According to a fifth aspect of the present invention, the pseudo-defective image creating means combines the difference data with a write-inhibited range setting means for setting a write-inhibited range that cannot be combined, and the difference data. An enlargement / reduction ratio setting means for setting an enlargement / reduction ratio in order to synthesize by enlarging or reducing; and a brightness setting means for setting brightness to synthesize by changing the brightness of the difference data; The first combining unit that does not synthesize the difference data in the write-inhibited range set by the write-inhibited range setting unit and the difference data are combined at the enlargement / reduction ratio set by the enlargement / reduction rate setting unit. It has a 2nd synthetic | combination means and a 3rd synthetic | combination means which synthesize | combines difference data with the brightness set by the said brightness setting means.
請求項６記載の発明は、ニューラルネットワークによる画像検査装置であって、請求項１〜５のいずれか１項に記載の擬似不良画像自動作成装置を具備することを特徴とする。 A sixth aspect of the present invention is an image inspection apparatus using a neural network, and includes the pseudo-defective image automatic creating apparatus according to any one of the first to fifth aspects.
本発明によれば、被検体の良品画像を入力する良品画像入力手段と、被検体の不良画像を入力する不良品画像入力手段と、被検体の不良画像から良品との差分データを抽出する不良画像抽出手段と、前記良品画像入力手段から入力された良品画像と前記差分データとを合成し、疑似不良画像を作成する疑似不良画像作成手段とを有することにより、人手を介さずに、擬似不良画像を大量に作成することが可能となる。 According to the present invention, a non-defective image input unit that inputs a non-defective image of a subject, a non-defective image input unit that inputs a defective image of the subject, and a defect that extracts difference data from a non-defective product from the defective image of the subject. By including image extraction means and pseudo-defective image creation means for synthesizing the non-defective image input from the non-defective image input means and the difference data to create a pseudo-defective image, pseudo-defectiveness is obtained without human intervention. A large amount of images can be created.
本発明の主たる目的は、工業製品等の検査において、より精度の高い検査装置を提供することであり、擬似不良画像自動作成装置を具備する画像検査装置の形態が、最も好ましい形態である。しかしながら、画像の加工などの分野においても、擬似不良画像自動作成装置のノウハウは、有効に利用することが可能である。 The main object of the present invention is to provide a highly accurate inspection apparatus for inspection of industrial products and the like, and the form of the image inspection apparatus provided with the pseudo-defective image automatic creating apparatus is the most preferable form. However, even in the field of image processing and the like, the know-how of the pseudo-defective image automatic creation device can be used effectively.
次に、添付図面を参照しながら、本発明の実施形態を説明する。
図１は、疑似不良画像作成装置を具備する画像検査装置の内部構成を示すブロック図であり、ニューラルネットワーク１０を用いて製品の良否を検査することが可能である。
良品の学習は、良品画像入力部１１にて、予め保存してある良品画像を読み込み、フィルタ処理や特徴量の抽出を行う前処理部１６を経由してニューラルネットワーク１０に入力し、良品データとして学習を行なわせる。
不良品の学習は、（不良品の画像数 ≧ 必要とする学習数）であれば、良品と同様に、予め保存してある不良品画像を読み込み、前処理部１６を経由してニューラルネットワーク１０に入力し、不良品データとして学習を行なわせる。
Next, embodiments of the present invention will be described with reference to the accompanying drawings.
FIG. 1 is a block diagram showing an internal configuration of an image inspection apparatus including a pseudo-defective image creation apparatus, and it is possible to inspect the quality of a product using a neural network 10.
In order to learn non-defective products, the non-defective image input unit 11 reads pre-stored non-defective images, and inputs them to the neural network 10 via the preprocessing unit 16 that performs filtering and extraction of feature amounts. Have students learn.
If the defective product is learned (number of images of defective product ≧ number of required learnings), the defective product image stored in advance is read in the same way as the non-defective product, and the neural network 10 is passed through the preprocessing unit 16. To learn as defective product data.
しかしながら、通常はオリジナルの不良データを必要学習数だけ入手することは困難である。
そのため、不良画像から良品との差分データ１７を抽出する不良画像抽出部１３と、良品画像部１１から良品画像を読み込み、差分データ１７を合成して疑似不良画像を生成する疑似不良画像作成部１５と、差分データ１７を合成位置等どのように合成するかの条件を、乱数発生部１８の乱数値を組み合わせて疑似不良画像作成部１５に指示する疑似データ条件設定部１４を備えている。
However, it is usually difficult to obtain the required number of original defective data.
Therefore, a defect image extraction unit 13 that extracts the difference data 17 from the defective image from the defect image, and a pseudo defect image creation unit 15 that reads the non-defective image from the non-defective image unit 11 and combines the difference data 17 to generate a pseudo defect image. And a pseudo data condition setting unit 14 for instructing the pseudo defect image creating unit 15 on conditions for how to synthesize the difference data 17 such as a synthesis position in combination with the random number value of the random number generating unit 18.
図２は、疑似不良画像作成処理の流れを示すフローチャートである。
まず初めに、不良画像入力部１２においてオリジナルの不良データを読み出し、不良画像抽出部１３で良品との差分データ１７を抽出する（ステップＳ１）。
次に、擬似データ条件設定部１４において擬似データ作成個数ｎを設定し、書き込み開始座標Ｙ［ｎ］、及びＸ［ｎ］を求める（ステップＳ２、Ｓ３）。
次に、差分データ１７に対して、設定条件に従い読み込んだ良品画像データ上に合成処理を行う（ステップＳ４）。こうして、擬似データ作成個数ｎ分、ステップＳ４〜Ｓ６の処理を繰り返し、ｎ＝０になったら（ステップＳ６／Ｙｅｓ）、処理を終了する。
FIG. 2 is a flowchart showing the flow of the pseudo-defective image creation process.
First, the original defect data is read out by the defective image input unit 12, and the difference data 17 from the non-defective product is extracted by the defective image extraction unit 13 (step S1).
Next, the pseudo data creation number n is set in the pseudo data condition setting unit 14, and the writing start coordinates Y [n] and X [n] are obtained (steps S2 and S3).
Next, a composition process is performed on the non-defective image data read in accordance with the setting conditions for the difference data 17 (step S4). In this way, the processing of steps S4 to S6 is repeated for the number of pseudo data created n, and when n = 0 (step S6 / Yes), the processing is terminated.
次に、具体例を用いて本実施例の動作を説明する。
図３は、液晶のしみ不良画像を等高線で表し、更にその一部を断面にした図である。
また、図４は、良品画像を等高線で表し、更にその一部を断面にした図である。
良品画像の場合、等高線は図４−（ｂ）に示すように楕円状になる。しかしながら、図３のようにしみ不良画像では、乱れた等高線となり（図３−（ｂ）参照）、その断面を見ると、しみの部分がへこんだ状態になる（図３−（ｃ）、（ｄ）参照）。
一方で、図４の良品の場合は、断面図は、滑らかな曲線となる（図４−（ｃ）、（ｄ）参照）。
Next, the operation of this embodiment will be described using a specific example.
FIG. 3 is a view in which a stain defect image of liquid crystal is represented by contour lines, and a part thereof is shown in cross section.
FIG. 4 is a view in which a non-defective image is represented by contour lines and a part thereof is shown in cross section.
In the case of a non-defective image, the contour lines are elliptical as shown in FIG. However, as shown in FIG. 3, the defective spot image has a distorted contour line (see FIG. 3- (b)), and when the cross section is viewed, the spot portion is dented (FIG. 3- (c), ( d)).
On the other hand, in the case of the non-defective product of FIG. 4, the cross-sectional view becomes a smooth curve (see FIGS. 4- (c) and (d)).
図３、図４で示した、良品の場合においては滑らかな曲線になる性質を利用して、不良画像抽出部１３では、図５の点線で示すように仮想の良品曲線を求め、その良品曲線と不良値との差ｄｚ［ｙ］［ｘ］を求め、これを差分データ１７とする。
ここで、ｘ，ｙは、しみを囲むｘ及びｙ方向の０から始まる座標を示し、その例を図７に示す（図２のステップＳ１参照）。
The defective image extraction unit 13 obtains a virtual non-defective curve as shown by the dotted line in FIG. 5 by using the property that becomes a smooth curve in the case of the non-defective product shown in FIGS. The difference dz [y] [x] between the error value and the defective value is obtained and used as difference data 17.
Here, x and y indicate coordinates starting from 0 in the x and y directions surrounding the stain, and an example thereof is shown in FIG. 7 (see step S1 in FIG. 2).
次に、疑似データ条件設定部１４において、１つの差分データ１７から作成する疑似不良作成数ｎを求める（図２のステップＳ２）。疑似不良データ数ｎは、プログラム上の固定値でもよいし、キーボード等から入力する形式を採ってもよい。
疑似不良作成数ｎを求めると、作成する設定値毎に乱数発生部１８から乱数値を求め、その乱数値から、以下の例のように書き込みが始まる（図２のステップＳ３）。
Next, the pseudo data condition setting unit 14 obtains the number of pseudo defects created n from one difference data 17 (step S2 in FIG. 2). The number of pseudo defective data n may be a fixed value on a program, or may be input from a keyboard or the like.
When the pseudo defect creation number n is obtained, a random value is obtained from the random number generator 18 for each set value to be created, and writing starts from the random value as in the following example (step S3 in FIG. 2).
（書き込み開始座標Ｙ［ｎ］）
← （乱数値／書き込み開始座標Ｙの最大値余り），
（書き込み開始座標Ｘ［ｎ］）
← （乱数値／書き込み開始座標Ｘの最大値余り）
ここで、本実施例では、製品の不良形態が位置のみ変動する製品を扱った場合とする。
(Write start coordinate Y [n])
← (Random value / Remainder of maximum value of writing start coordinate Y),
(Write start coordinate X [n])
← (Random value / Remainder of maximum value of writing start coordinate X)
Here, in this embodiment, it is assumed that a product in which the defective form of the product varies only in position is handled.
次に、疑似不良画像作成部１５において、良品画像の座標Ｙ［ｎ］，Ｘ［ｎ］を書き込み開始点として、式（１）のように、図７−（ａ）の差分データ１７の分元データから差し引いて、疑似不良画像を作成する（図２のステップＳ４）。 Next, the pseudo-defective image creation unit 15 uses the coordinates Y [n], X [n] of the non-defective image as the writing start point, and the difference data 17 shown in FIG. A pseudo-defective image is created by subtracting from the original data (step S4 in FIG. 2).
（座標Ｙ［ｎ］＋ｙ、Ｘ［ｎ］＋ｘの疑似画像データ）
＝ （座標Ｙ［ｎ］＋ｙ、Ｘ［ｎ］＋ｘの元画像データ）
−（座標Ｙ［ｎ］＋ｙ、Ｘ［ｎ］＋ｘの差分データ） ・・・（１）
(Pseudo image data of coordinates Y [n] + y, X [n] + x)
= (Original image data of coordinates Y [n] + y, X [n] + x)
-(Difference data of coordinates Y [n] + y, X [n] + x) (1)
例えば、図６は、良品画像を疑似不良画像にした場合の等高線図と、座標Ｙ[ｎ]＋１０のＸ方向の断面図である。
同様にして、ｎ個の疑似不良画像を繰り返し作成し（図２のステップＳ５、Ｓ６）、ニューラルネットワーク１０の学習データとして使用する。
For example, FIG. 6 is a contour map when a non-defective image is a pseudo-defective image and a cross-sectional view in the X direction of coordinates Y [n] +10.
Similarly, n pseudo-defective images are repeatedly created (steps S5 and S6 in FIG. 2) and used as learning data for the neural network 10.
実施例１の第１の効果は、装置のテスト段階で、疑似不良データを大量に作成し、学習データとして使用することにより、本番の検査時は、早期に高い精度の検査結果を得ることが可能となる。
その理由は、ニューラルネットワークの判定精度は、学習数に依るところが大きいが、実際、不良データは入手しにくく、その問題に対して擬似的に不良データを大量作成できるようにしたからである。
The first effect of the first embodiment is that a large amount of pseudo-defective data is created and used as learning data at the test stage of the apparatus, so that a high-accuracy inspection result can be obtained early during the actual inspection. It becomes possible.
The reason is that the determination accuracy of the neural network largely depends on the number of learning, but in fact, it is difficult to obtain defective data, and a large amount of defective data can be created in a pseudo manner for the problem.
第２の効果は、人の手を介在させることなく、大量に疑似データを作成できる点である。
その理由は、不良がもつパラメータを乱数により作り出すことで、自動的に疑似データを大量作成できるようにしたからである。
The second effect is that a large amount of pseudo data can be created without any human intervention.
The reason is that a large amount of pseudo-data can be automatically created by generating random parameters using random numbers.
次に、本発明の他の実施例について、図面を参照して詳細に説明する。
図８は、疑似不良画像作成処理の流れを示すフローチャートである。なお、本図での処理の流れは、図２と比較して、書き込み禁止範囲の設定や、拡大・縮小率の設定、不良の明るさの設定、拡大・縮小処理、明るさの変更処理が加えられている点で、図２と異なっている。
Next, another embodiment of the present invention will be described in detail with reference to the drawings.
FIG. 8 is a flowchart showing the flow of the pseudo-defective image creation process. Compared to FIG. 2, the processing flow in this figure is different from that in FIG. 2 in that a write-protection range setting, an enlargement / reduction ratio setting, a defective brightness setting, an enlargement / reduction process, and a brightness change process are performed. It is different from FIG. 2 in that it is added.
まず初めに、不良画像入力部１２においてオリジナルの不良データを読み出し、不良画像抽出部１３で良品との差分データ１７を抽出する（ステップＳ１１）。
次に、擬似データ条件設定部１４において擬似データ作成個数ｎを求め、書き込み禁止範囲ＡＹ［ｎ］［ａ］，ＡＸ［ｎ］［ａ］を求める（ステップＳ１２、Ｓ１３）。
First, the original defect data is read out by the defective image input unit 12, and the difference data 17 from the non-defective product is extracted by the defective image extraction unit 13 (step S11).
Next, the pseudo data condition setting unit 14 obtains the number n of pseudo data to be created, and obtains the write prohibition ranges AY [n] [a], AX [n] [a] (steps S12 and S13).
次に、書き込み開始座標Ｙ［ｎ］、及びＸ［ｎ］を求め（ステップＳ１４）、書き込み開始位置が書き込み禁止範囲ＡＹ［ｎ］［ａ］，ＡＸ［ｎ］［ａ］にないことを確認する（ステップＳ１５）。
書き込み開始位置が、書き込み禁止位置にある場合（ステップＳ１５／Ｙｅｓ）は、再び書き込み開始位置を求める（ステップＳ１４）。
書き込み開始位置が、書き込み禁止位置にない場合（ステップＳ１５／Ｎｏ）は、次の処理に進む。
Next, write start coordinates Y [n] and X [n] are obtained (step S14), and it is confirmed that the write start position is not in the write prohibition ranges AY [n] [a] and AX [n] [a]. (Step S15).
When the write start position is at the write prohibit position (step S15 / Yes), the write start position is obtained again (step S14).
If the write start position is not at the write prohibit position (step S15 / No), the process proceeds to the next process.
次に、乱数発生部１８で得られる乱数値により、次式（２）、（３）のように、拡大・縮小率Ｚ［ｎ］、不良の明るさｍ［ｎ］を求め、設定する（ステップＳ１６、Ｓ１７）。
Ｚ［ｎ］←（乱数値／１０の余り）の値で図７（ｂ）より求める・・・（２）
ｍ［ｎ］←（乱数値／１０の余り）の値で図７（ｃ）より求める・・・（３）
ただし、拡大・縮小率、及び不良の明るさは、それぞれ１０段階とする。
Next, the enlargement / reduction ratio Z [n] and the defect brightness m [n] are obtained and set as shown in the following equations (2) and (3) based on the random value obtained by the random number generator 18 ( Steps S16 and S17).
Obtained from FIG. 7 (b) with the value of Z [n] ← (the remainder of random number / 10) (2)
Obtained from FIG. 7 (c) with the value of m [n] ← (the remainder of random number / 10) (3)
However, the enlargement / reduction ratio and the brightness of the defect are each in 10 stages.
次に、上で設定した拡大・縮小率Ｚ［ｎ］を用いて、差分データ１７を拡大・縮小する（ステップＳ１８）。
簡単のため、拡大・縮小率がＺ［ｎ］＝２の場合は、２倍の拡大とし、差分データ１７をＸ及びＹ方向に同じデータが２回続く処理を行い、全体を２×２倍する。
また、拡大・縮小率がＺ［ｎ］＝０．５の場合には、１／２の縮小として、差分データ１７をＸ及びＹ方向ともデータを間引く処理を行う。
Next, the difference data 17 is enlarged / reduced using the enlargement / reduction ratio Z [n] set above (step S18).
For simplicity, when the enlargement / reduction ratio is Z [n] = 2, the enlargement is doubled, the difference data 17 is processed twice in the X and Y directions, and the entire data is 2 × 2 times. To do.
When the enlargement / reduction ratio is Z [n] = 0.5, the difference data 17 is thinned out in both the X and Y directions as 1/2 reduction.
次に、不良の明るさｍ［ｎ］においては、差分データ１７にｍ［ｎ］を乗算する処理を行い（ステップＳ１９）、疑似データ作成を行う（ステップＳ２０）。
このことは、式（１）が、以下に示す式（４）のようになることを意味し、ｍ［ｎ］が１よりも大きい場合は、不良箇所が暗くなり、１より小さい場合には不良箇所が明るくなる。
Next, for defective brightness m [n], a process of multiplying the difference data 17 by m [n] is performed (step S19), and pseudo data is created (step S20).
This means that the formula (1) becomes as shown in the following formula (4). When m [n] is larger than 1, the defective portion becomes dark, and when m [n] is smaller than 1, The defective part becomes brighter.
（座標Ｙ［ｎ］＋ｙ、Ｘ［ｎ］＋ｘの疑似画像データ）
＝ （座標Ｙ［ｎ］＋ｙ、Ｘ［ｎ］＋ｘの元画像データ）
−（座標ｙ、ｘの差分データ１７）×ｍ［ｎ］ ・・・（４）
(Pseudo image data of coordinates Y [n] + y, X [n] + x)
= (Original image data of coordinates Y [n] + y, X [n] + x)
− (Difference data 17 of coordinates y and x) × m [n] (4)
このようにして、擬似データ作成個数ｎ分だけ、ステップＳ１９〜Ｓ２１の処理を繰り返し、ｎ＝０になったら（ステップＳ２２／Ｙｅｓ）、処理を終了する。 In this way, the processing of steps S19 to S21 is repeated for the number of pseudo data created n, and when n = 0 (step S22 / Yes), the processing is terminated.
実施例２の効果は、様々な種類の擬似不良画像を作成することができる点である。
例えば、同様の欠陥が画像の中央と画像の端側にある場合に、画像の中央にある欠陥は不良であると判定したい場合は、擬似不良画像を作成する際に、画像の中央部には差分データの書き込みを許可し、一方で、画像の端側にある欠陥は不良でないと判定したい場合には、擬似不良画像を作成する際に、端側には差分データの書き込みを許可しないというような設定をすることにより、多くのパターン（擬似不良画像）を作成することが可能となる。
また、差分データに対して、拡大・縮小の設定をおこなったり、色の明るさに変化を持たせることにより、多くのパターン（擬似不良画像）を作成することが可能となる。
The effect of the second embodiment is that various types of pseudo-defective images can be created.
For example, if the same defect is at the center of the image and the edge of the image, and you want to determine that the defect at the center of the image is defective, when creating a pseudo-defective image, If you want to allow writing of difference data while determining that the defect on the edge side of the image is not defective, do not allow writing of difference data on the edge side when creating a pseudo-bad image With this setting, many patterns (pseudo-defective images) can be created.
In addition, it is possible to create a large number of patterns (pseudo-defective images) by setting the enlargement / reduction for the difference data or changing the brightness of the color.
（効果）
以上の説明から明らかなように、被検体の良品画像を入力する良品画像入力手段と、被検体の不良画像を入力する不良品画像入力手段と、被検体の不良画像から良品との差分データを抽出する不良画像抽出手段と、前記良品画像入力手段から入力された良品画像と前記差分データとを合成し、疑似不良画像を作成する疑似不良画像作成手段とを有することにより、人手を介さずに、擬似不良画像を大量に作成することが可能となる。
(effect)
As is clear from the above description, the non-defective image input means for inputting the non-defective image of the subject, the defective image input means for inputting the defective image of the subject, and the difference data between the non-defective product from the defective image of the subject. By having a defective image extraction means for extracting, and a non-defective image input means from the non-defective image input means and the pseudo-defective image creating means for synthesizing the difference data and creating a pseudo-defective image, without human intervention It is possible to create a large number of pseudo-defective images.
また、前記擬似不良画像作成手段は、乱数値を発生させる乱数発生手段と、良品画像と差分データを合成する際に、前記乱数発生手段にて発生させた乱数値を組み合わせて合成するための条件を設定する擬似データ作成条件設定手段とを有することにより、擬似不良画像を大量に作成することが可能となる。 In addition, the pseudo-defective image creating means includes a random number generating means for generating a random value, and a condition for combining the non-defective image and the difference data generated by combining the random number values generated by the random number generating means. It is possible to create a large number of pseudo-defective images.
また、前記擬似データ作成条件設定手段は、擬似不良画像の作成個数を設定する作成個数設定手段を有することにより、所望する数の擬似不良画像を作成することが可能となる。 Further, the pseudo data creation condition setting means has creation number setting means for setting the number of created pseudo defect images, so that a desired number of pseudo defect images can be created.
また、前記不良画像抽出手段は、被検体の不良画像から、仮想的に良品の状態を算出する仮想良品状態算出手段と、不良画像と前記仮想良品状態算出手段にて求められた仮想的な良品の状態との差から差分データを抽出する第２の不良画像抽出手段とを有することにより、良品データを読み込まずに、擬似不良画像を作成することが可能となる。 In addition, the defective image extraction unit includes a virtual non-defective product state calculating unit that virtually calculates a non-defective product state from the defective image of the subject, and a virtual non-defective product obtained by the defective image and the virtual non-defective product state calculating unit. By having the second defective image extracting means for extracting the difference data from the difference from the state, it becomes possible to create a pseudo-defective image without reading the good product data.
また、前記擬似不良画像作成手段は、前記差分データを合成する際に、合成することが不可能である書き込み禁止範囲を設定する書き込み禁止範囲設定手段と、前記差分データを拡大もしくは縮小して合成するために、拡大・縮小率を設定する拡大・縮小率設定手段と、前記差分データの明るさを変更して合成するために、明るさを設定する明るさ設定手段と、前記書き込み禁止範囲設定手段で設定した書き込み禁止範囲には差分データの合成を行わない第１の合成手段と、前記拡大・縮小率設定手段で設定した拡大・縮小率で差分データの合成を行う第２の合成手段と、前記明るさ設定手段で設定した明るさで差分データの合成を行う第３の合成手段とを有することにより、様々なパターンの擬似不良画像を大量に作成することが可能となる。 The pseudo-defective image creating means is configured to synthesize the difference data by enlarging or reducing the difference data, and a write prohibition range setting means for setting a write prohibition range that cannot be combined when the difference data is synthesized. An enlargement / reduction ratio setting means for setting an enlargement / reduction ratio, a brightness setting means for setting brightness to change and synthesize the brightness of the difference data, and the write prohibition range setting A first combining unit that does not combine the difference data in the write-inhibited range set by the unit, and a second combining unit that combines the difference data with the enlargement / reduction ratio set by the enlargement / reduction rate setting unit In addition, it is possible to create a large number of pseudo-defective images of various patterns by including the third combining unit that combines the difference data with the brightness set by the brightness setting unit. .
また、ニューラルネットワークによる画像検査装置であって、請求項１〜５のいずれか１項に記載の擬似不良画像自動作成装置を具備することにより、擬似不良画像を大量に作成し、学習データとしてニューラルネットワークに学習させることが可能となる。また、ニューラルネットワークに大量のデータを学習させることにより、本番の検査時には、早期に高い精度の検査結果を得ることが可能となる。 An image inspection apparatus using a neural network, comprising the pseudo-defective image automatic creation device according to any one of claims 1 to 5 to create a large number of pseudo-defective images, and neural as learning data It is possible to let the network learn. In addition, by making a neural network learn a large amount of data, it is possible to obtain a highly accurate test result at an early stage during the actual test.
１０ ニューラルネットワーク
１１ 良品画像入力部
１２ 不良画像入力部
１３ 不良画像抽出部
１４ 疑似データ条件設定部
１５ 疑似不良画像作成部
１６ 前処理部
１７ 差分データ
１８ 乱数発生部
ＡＸ［ｎ］、ＡＹ［ｎ］ 書き込み禁止範囲
Ｘ［ｎ］、Ｙ［ｎ］ 書き込み開始位置
ｍ［ｎ］ 不良の明るさ
Ｚ［ｎ］ 拡大・縮小率
ｄｚ 良品曲線と不良値との差
ｎ 疑似不良作成数
DESCRIPTION OF SYMBOLS 10 Neural network 11 Good product image input part 12 Defective image input part 13 Defective image extraction part 14 Pseudo data condition setting part 15 Pseudo defective image creation part 16 Pre-processing part 17 Difference data 18 Random number generation part AX [n], AY [n] Write inhibition range X [n], Y [n] Write start position m [n] Defect brightness Z [n] Enlargement / reduction ratio dz Difference between non-defective curve and defect value n Number of pseudo defects created
Claims (6)
被検体の不良画像を入力する不良品画像入力手段と、
被検体の不良画像から良品との差分データを抽出する不良画像抽出手段と、
前記良品画像入力手段から入力された良品画像と前記差分データとを合成し、疑似不良画像を作成する疑似不良画像作成手段とを有することを特徴とする疑似不良画像自動作成装置。 A non-defective image input means for inputting a non-defective image of the subject;
A defective product image input means for inputting a defective image of the subject;
A defective image extracting means for extracting difference data with a non-defective product from a defective image of a subject;
An apparatus for automatically creating a pseudo-defective image, comprising: a pseudo-defective image creating unit that synthesizes a non-defective image input from the non-defective image input unit and the difference data to create a pseudo-defective image.
良品画像と差分データを合成する際に、前記乱数発生手段にて発生させた乱数値を組み合わせて合成するための条件を設定する擬似データ作成条件設定手段とを有することを特徴とする請求項１記載の擬似不良画像自動作成装置。 The pseudo-defective image creating means includes a random number generating means for generating a random value,
2. A pseudo data creation condition setting unit that sets a condition for combining a random image value generated by the random number generation unit when combining a non-defective image and difference data. The pseudo-faulty image automatic creation device described.
不良画像と前記仮想良品状態算出手段にて求められた仮想的な良品の状態との差から差分データを抽出する第２の不良画像抽出手段とを有することを特徴とする請求項１〜３のいずれか１項に記載の擬似不良画像自動作成装置。 The defective image extraction means includes a virtual non-defective product state calculating means for virtually calculating a good product state from a defective image of the subject;
4. A second defective image extracting unit that extracts difference data from a difference between a defective image and a virtual good product state obtained by the virtual good product state calculating unit. The pseudo-faulty image automatic creation device according to any one of the above.
前記差分データを拡大もしくは縮小して合成するために、拡大・縮小率を設定する拡大・縮小率設定手段と、
前記差分データの明るさを変更して合成するために、明るさを設定する明るさ設定手段と、
前記書き込み禁止範囲設定手段で設定した書き込み禁止範囲には差分データの合成を行わない第１の合成手段と、
前記拡大・縮小率設定手段で設定した拡大・縮小率で差分データの合成を行う第２の合成手段と、
前記明るさ設定手段で設定した明るさで差分データの合成を行う第３の合成手段とを有することを特徴とする請求項１〜４のいずれか１項に記載の擬似不良画像自動作成装置。 The pseudo-defective image creating means, when synthesizing the difference data, write prohibition range setting means for setting a write prohibition range that cannot be combined,
An enlargement / reduction ratio setting means for setting an enlargement / reduction ratio to synthesize the difference data by enlarging or reducing,
Brightness setting means for setting brightness in order to change and synthesize the brightness of the difference data;
A first combining unit that does not synthesize difference data in the write prohibited range set by the write prohibited range setting unit;
Second combining means for combining difference data at the enlargement / reduction ratio set by the enlargement / reduction ratio setting means;
5. The pseudo-defective image automatic creating apparatus according to claim 1, further comprising: a third combining unit configured to combine the difference data with the brightness set by the brightness setting unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2003394788A JP2005156334A (en) | 2003-11-25 | 2003-11-25 | Pseudo defective image automatic creation device and imaging inspection device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2003394788A JP2005156334A (en) | 2003-11-25 | 2003-11-25 | Pseudo defective image automatic creation device and imaging inspection device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| JP2005156334A true JP2005156334A (en) | 2005-06-16 |
Family
ID=34720719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP2003394788A Withdrawn JP2005156334A (en) | 2003-11-25 | 2003-11-25 | Pseudo defective image automatic creation device and imaging inspection device |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JP2005156334A (en) |
Cited By (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010164506A (en) * | 2009-01-19 | 2010-07-29 | Panasonic Corp | Inspection method |
| CN102129563A (en) * | 2010-01-15 | 2011-07-20 | 松下电器产业株式会社 | Sensory testing device and sensory testing method |
| JP2011214903A (en) * | 2010-03-31 | 2011-10-27 | Denso It Laboratory Inc | Appearance inspection apparatus, and apparatus, method and program for generating appearance inspection discriminator |
| JP2014182092A (en) * | 2013-03-21 | 2014-09-29 | Jx Nippon Oil & Energy Corp | Abnormality detection method and abnormality detection device |
| KR101688458B1 (en) * | 2016-04-27 | 2016-12-23 | 디아이티 주식회사 | Image inspection apparatus for manufactured articles using deep neural network training method and image inspection method of manufactured articles thereby |
| JP6216024B1 (en) * | 2016-11-15 | 2017-10-18 | 株式会社Ｐｒｅｆｅｒｒｅｄ Ｎｅｔｗｏｒｋｓ | Trained model generation method and signal data discrimination device |
| JP2018205123A (en) * | 2017-06-05 | 2018-12-27 | 学校法人梅村学園 | Image generation device and image generation method of generating an inspection-purpose image for making performance adjustment of image inspection system |
| WO2019022170A1 (en) * | 2017-07-26 | 2019-01-31 | 横浜ゴム株式会社 | Defect inspecting method and defect inspecting device |
| WO2019039757A1 (en) * | 2017-08-24 | 2019-02-28 | 주식회사 수아랩 | Method and device for generating training data and computer program stored in computer-readable recording medium |
| WO2019107614A1 (en) * | 2017-11-30 | 2019-06-06 | 전자부품연구원 | Machine vision-based quality inspection method and system utilizing deep learning in manufacturing process |
| WO2019188040A1 (en) * | 2018-03-29 | 2019-10-03 | 日本電気株式会社 | Image processing device, image processing method, and image processing program |
| JP2020027424A (en) * | 2018-08-10 | 2020-02-20 | 東京エレクトロンデバイス株式会社 | Learning data generating device, discrimination model generating device, and program |
| JP2020041889A (en) * | 2018-09-10 | 2020-03-19 | 日本電気硝子株式会社 | Method and system for inspecting workpiece |
| JP2020052044A (en) * | 2018-09-21 | 2020-04-02 | 古河電気工業株式会社 | Image determination device, image inspection device, manufacturing system of electric wire with terminal, and image determination method |
| JP2020061007A (en) * | 2018-10-11 | 2020-04-16 | 富士通株式会社 | Learning program, learning method and learning device |
| WO2020255292A1 (en) * | 2019-06-19 | 2020-12-24 | 株式会社島津製作所 | Bone section image analysis method and learning method |
| JP2021002270A (en) * | 2019-06-24 | 2021-01-07 | 株式会社Ｊｖｃケンウッド | Image recognition learning device, image recognition learning method, image recognition learning program and terminal device |
| KR20210003661A (en) * | 2019-07-02 | 2021-01-12 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| US20210080400A1 (en) * | 2019-09-12 | 2021-03-18 | Aisin Seiki Kabushiki Kaisha | Image restoration apparatus, image restoration method, image restoration program, restorer generation apparatus, restorer generation method, restorer generation program, determiner generation apparatus, determiner generation method, determiner generation program, article determination apparatus, article determination method, and article determination program |
| WO2021070675A1 (en) * | 2019-10-08 | 2021-04-15 | キヤノン株式会社 | Teacher-data generating method, trained learning model, and system |
| JP2021089219A (en) * | 2019-12-05 | 2021-06-10 | 東洋製罐グループホールディングス株式会社 | Image inspection system and image inspection method |
| KR102272497B1 (en) * | 2020-07-31 | 2021-07-02 | (주)딥인사이트 | Object-oriented data augmentation method |
| CN113138198A (en) * | 2021-04-27 | 2021-07-20 | 环旭(深圳)电子科创有限公司 | System and method for generating defect image of electronic element |
| WO2021152416A1 (en) * | 2020-01-31 | 2021-08-05 | 株式会社半導体エネルギー研究所 | Training data generation device and defect discrimination system |
| WO2021161853A1 (en) * | 2020-02-12 | 2021-08-19 | 株式会社モルフォ | Analysis device and analysis method |
| KR20210116913A (en) * | 2020-03-18 | 2021-09-28 | 라온피플 주식회사 | Apparatus and method for generating a defect image |
| WO2021193347A1 (en) * | 2020-03-26 | 2021-09-30 | パナソニックＩｐマネジメント株式会社 | Data generation system, data generation method, data generation device, and additional learning requirement assessment device |
| CN113538631A (en) * | 2020-04-20 | 2021-10-22 | 贤智研究株式会社 | Computer program, method and apparatus for generating virtual defect image by artificial intelligence model generated based on user input |
| KR20210128190A (en) * | 2020-04-16 | 2021-10-26 | 한국세라믹기술원 | Method For Virtual Data Generation of Crack Patterns |
| KR20210153586A (en) * | 2019-07-02 | 2021-12-17 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| KR20220008530A (en) * | 2020-07-14 | 2022-01-21 | 한국생산기술연구원 | System and method of defect inspection using generation and transformation of input and output data based on deep learning |
| JPWO2022044150A1 (en) * | 2020-08-26 | 2022-03-03 | ||
| JP7043645B1 (en) | 2021-03-03 | 2022-03-29 | Dmg森精機株式会社 | Board inspection method |
| WO2022065271A1 (en) * | 2020-09-25 | 2022-03-31 | ファナック株式会社 | Image creation device |
| KR20220074021A (en) * | 2020-11-27 | 2022-06-03 | 씨제이올리브네트웍스 주식회사 | Deep learning-based image analysis method for selecting defective products and a system therefor |
| US11373293B2 (en) | 2019-06-27 | 2022-06-28 | SCREEN Holdings Co., Ltd. | Method for building image determination model, image determination model, and image determination method |
| US11373292B2 (en) | 2018-12-28 | 2022-06-28 | Tdk Corporation | Image generation device and appearance inspection device |
| JP2022100292A (en) * | 2020-12-23 | 2022-07-05 | 株式会社リコー | Neural network training method, defect detection method and apparatus therefor, and recording medium |
| DE102022208849A1 (en) | 2021-09-30 | 2023-03-30 | Omron Corporation | IMAGE TESTING METHOD AND IMAGE TESTING DEVICE |
| WO2023100474A1 (en) * | 2021-12-02 | 2023-06-08 | 株式会社日立製作所 | System, image processing method, and program |
| EP3997663A4 (en) * | 2019-07-12 | 2023-08-09 | The Procter & Gamble Company | System and method for providing textile information and visualizing same |
| WO2023175348A1 (en) * | 2022-03-17 | 2023-09-21 | Cheyney Design & Development Ltd. | Improvements in or relating to inspection and quality control |
| WO2023190045A1 (en) * | 2022-03-29 | 2023-10-05 | パナソニックＩｐマネジメント株式会社 | Image generation system, image generation method, and program |
| WO2023238590A1 (en) * | 2022-06-10 | 2023-12-14 | 日立Astemo株式会社 | Pseudo defect image generation device |
| US11966219B2 (en) | 2022-03-30 | 2024-04-23 | Honda Motor Co., Ltd. | Pseudo defective product data generation method |
| JP7531177B2 (en) | 2020-07-08 | 2024-08-09 | 日新電機株式会社 | Effluent water quality prediction device and effluent water quality prediction method |
-
2003
- 2003-11-25 JP JP2003394788A patent/JP2005156334A/en not_active Withdrawn
Cited By (84)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010164506A (en) * | 2009-01-19 | 2010-07-29 | Panasonic Corp | Inspection method |
| CN102129563A (en) * | 2010-01-15 | 2011-07-20 | 松下电器产业株式会社 | Sensory testing device and sensory testing method |
| JP2011214903A (en) * | 2010-03-31 | 2011-10-27 | Denso It Laboratory Inc | Appearance inspection apparatus, and apparatus, method and program for generating appearance inspection discriminator |
| JP2014182092A (en) * | 2013-03-21 | 2014-09-29 | Jx Nippon Oil & Energy Corp | Abnormality detection method and abnormality detection device |
| KR101688458B1 (en) * | 2016-04-27 | 2016-12-23 | 디아이티 주식회사 | Image inspection apparatus for manufactured articles using deep neural network training method and image inspection method of manufactured articles thereby |
| JP2018081442A (en) * | 2016-11-15 | 2018-05-24 | 株式会社Ｐｒｅｆｅｒｒｅｄ Ｎｅｔｗｏｒｋｓ | Learned model generating method and signal data discrimination device |
| JP6216024B1 (en) * | 2016-11-15 | 2017-10-18 | 株式会社Ｐｒｅｆｅｒｒｅｄ Ｎｅｔｗｏｒｋｓ | Trained model generation method and signal data discrimination device |
| WO2018092747A1 (en) * | 2016-11-15 | 2018-05-24 | 株式会社Ｐｒｅｆｅｒｒｅｄ Ｎｅｔｗｏｒｋｓ | Learned model generation method, learned model generation device, signal data discrimination method, signal data discrimination device, and signal data discrimination program |
| JP7254324B2 (en) | 2017-06-05 | 2023-04-10 | 学校法人梅村学園 | IMAGE GENERATING APPARATUS AND IMAGE GENERATING METHOD FOR GENERATING INSPECTION IMAGE FOR PERFORMANCE ADJUSTMENT OF IMAGE INSPECTION SYSTEM |
| JP2018205123A (en) * | 2017-06-05 | 2018-12-27 | 学校法人梅村学園 | Image generation device and image generation method of generating an inspection-purpose image for making performance adjustment of image inspection system |
| JP7210873B2 (en) | 2017-07-26 | 2023-01-24 | 横浜ゴム株式会社 | Defect inspection method and defect inspection apparatus |
| JP2019027826A (en) * | 2017-07-26 | 2019-02-21 | 横浜ゴム株式会社 | Defect inspection method and defect inspection device |
| WO2019022170A1 (en) * | 2017-07-26 | 2019-01-31 | 横浜ゴム株式会社 | Defect inspecting method and defect inspecting device |
| US11313806B2 (en) | 2017-07-26 | 2022-04-26 | The Yokohama Rubber Co., Ltd. | Defect inspection method and defect inspection device |
| WO2019039757A1 (en) * | 2017-08-24 | 2019-02-28 | 주식회사 수아랩 | Method and device for generating training data and computer program stored in computer-readable recording medium |
| KR20190021967A (en) * | 2017-08-24 | 2019-03-06 | 주식회사 수아랩 | Method, apparatus and computer program stored in computer readable medium for generating training data |
| KR101992239B1 (en) | 2017-08-24 | 2019-06-25 | 주식회사 수아랩 | Method, apparatus and computer program stored in computer readable medium for generating training data |
| US11605003B2 (en) * | 2017-08-24 | 2023-03-14 | Sualab Co., Ltd. | Method and device for generating training data and computer program stored in computer-readable recording medium |
| WO2019107614A1 (en) * | 2017-11-30 | 2019-06-06 | 전자부품연구원 | Machine vision-based quality inspection method and system utilizing deep learning in manufacturing process |
| JPWO2019188040A1 (en) * | 2018-03-29 | 2021-03-11 | 日本電気株式会社 | Image processing equipment, image processing method and image processing program |
| US11797886B2 (en) | 2018-03-29 | 2023-10-24 | Nec Corporation | Image processing device, image processing method, and image processing program |
| WO2019188040A1 (en) * | 2018-03-29 | 2019-10-03 | 日本電気株式会社 | Image processing device, image processing method, and image processing program |
| JP2020027424A (en) * | 2018-08-10 | 2020-02-20 | 東京エレクトロンデバイス株式会社 | Learning data generating device, discrimination model generating device, and program |
| JP2020041889A (en) * | 2018-09-10 | 2020-03-19 | 日本電気硝子株式会社 | Method and system for inspecting workpiece |
| JP7054450B2 (en) | 2018-09-10 | 2022-04-14 | 日本電気硝子株式会社 | Work inspection method |
| JP7306933B2 (en) | 2018-09-21 | 2023-07-11 | 古河電気工業株式会社 | Image determination device, image inspection device, and image determination method |
| JP2020052044A (en) * | 2018-09-21 | 2020-04-02 | 古河電気工業株式会社 | Image determination device, image inspection device, manufacturing system of electric wire with terminal, and image determination method |
| JP2020061007A (en) * | 2018-10-11 | 2020-04-16 | 富士通株式会社 | Learning program, learning method and learning device |
| JP7115207B2 (en) | 2018-10-11 | 2022-08-09 | 富士通株式会社 | Learning program, learning method and learning device |
| US11373292B2 (en) | 2018-12-28 | 2022-06-28 | Tdk Corporation | Image generation device and appearance inspection device |
| JPWO2020255292A1 (en) * | 2019-06-19 | 2020-12-24 | ||
| WO2020255292A1 (en) * | 2019-06-19 | 2020-12-24 | 株式会社島津製作所 | Bone section image analysis method and learning method |
| JP7173338B2 (en) | 2019-06-19 | 2022-11-16 | 株式会社島津製作所 | Bone image analysis method and learning method |
| JP2021002270A (en) * | 2019-06-24 | 2021-01-07 | 株式会社Ｊｖｃケンウッド | Image recognition learning device, image recognition learning method, image recognition learning program and terminal device |
| US11373293B2 (en) | 2019-06-27 | 2022-06-28 | SCREEN Holdings Co., Ltd. | Method for building image determination model, image determination model, and image determination method |
| KR20210153586A (en) * | 2019-07-02 | 2021-12-17 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| KR20210003661A (en) * | 2019-07-02 | 2021-01-12 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| KR102450130B1 (en) * | 2019-07-02 | 2022-10-04 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| KR102450131B1 (en) * | 2019-07-02 | 2022-10-04 | 주식회사 마키나락스 | Systems and methods for detecting flaws on panels using images of the panels |
| US11948292B2 (en) | 2019-07-02 | 2024-04-02 | MakinaRocks Co., Ltd. | Systems and methods for detecting flaws on panels using images of the panels |
| EP3997663A4 (en) * | 2019-07-12 | 2023-08-09 | The Procter & Gamble Company | System and method for providing textile information and visualizing same |
| JP7383946B2 (en) | 2019-09-12 | 2023-11-21 | 株式会社アイシン | Image restoration device, image restoration method, image restoration program, restorer generation device, restorer generation method, restorer generation program, determiner generation device, determiner generation method, determiner generation program, article determination device, article determination method, and article judgment program |
| US20210080400A1 (en) * | 2019-09-12 | 2021-03-18 | Aisin Seiki Kabushiki Kaisha | Image restoration apparatus, image restoration method, image restoration program, restorer generation apparatus, restorer generation method, restorer generation program, determiner generation apparatus, determiner generation method, determiner generation program, article determination apparatus, article determination method, and article determination program |
| JP2021043816A (en) * | 2019-09-12 | 2021-03-18 | アイシン精機株式会社 | Image restoration apparatus, image restoration method, image restoration program, restorer generation apparatus, restorer generation method, restorer generation program, determiner generation apparatus, determiner generation method, determiner generation program, article determination apparatus, article determination method, and article determination program |
| US11815468B2 (en) * | 2019-09-12 | 2023-11-14 | Aisin Corporation | Image restoration apparatus, image restoration method, image restoration program, restorer generation apparatus, restorer generation method, restorer generation program, determiner generation apparatus, determiner generation method, determiner generation program, article determination apparatus, article determination method, and article determination program |
| JP7536517B2 (en) | 2019-10-08 | 2024-08-20 | キヤノン株式会社 | Method for generating training data, trained learning model, and system |
| WO2021070675A1 (en) * | 2019-10-08 | 2021-04-15 | キヤノン株式会社 | Teacher-data generating method, trained learning model, and system |
| JP2021089219A (en) * | 2019-12-05 | 2021-06-10 | 東洋製罐グループホールディングス株式会社 | Image inspection system and image inspection method |
| JP7520056B2 (en) | 2020-01-31 | 2024-07-22 | 株式会社半導体エネルギー研究所 | Learning data generation device, defect identification system |
| WO2021152416A1 (en) * | 2020-01-31 | 2021-08-05 | 株式会社半導体エネルギー研究所 | Training data generation device and defect discrimination system |
| JP7492240B2 (en) | 2020-02-12 | 2024-05-29 | 株式会社 東京ウエルズ | Analytical device and analytical method |
| WO2021161853A1 (en) * | 2020-02-12 | 2021-08-19 | 株式会社モルフォ | Analysis device and analysis method |
| KR20220104020A (en) | 2020-02-12 | 2022-07-25 | 가부시키가이샤 모르포 | Analytical devices and analytical methods |
| KR102372987B1 (en) | 2020-03-18 | 2022-03-11 | 라온피플 주식회사 | Apparatus and method for generating a defect image |
| KR20210116913A (en) * | 2020-03-18 | 2021-09-28 | 라온피플 주식회사 | Apparatus and method for generating a defect image |
| WO2021193347A1 (en) * | 2020-03-26 | 2021-09-30 | パナソニックＩｐマネジメント株式会社 | Data generation system, data generation method, data generation device, and additional learning requirement assessment device |
| KR102408754B1 (en) * | 2020-04-16 | 2022-06-13 | 한국세라믹기술원 | Method For Virtual Data Generation of Crack Patterns |
| KR20210128190A (en) * | 2020-04-16 | 2021-10-26 | 한국세라믹기술원 | Method For Virtual Data Generation of Crack Patterns |
| KR102430090B1 (en) | 2020-04-20 | 2022-08-11 | 세이지리서치 주식회사 | Computer program, method, and device for generating virtual defect image using artificial intelligence model generated based on user input |
| KR20210129775A (en) * | 2020-04-20 | 2021-10-29 | 세이지리서치 주식회사 | Computer program, method, and device for generating virtual defect image using artificial intelligence model generated based on user input |
| WO2021215730A1 (en) * | 2020-04-20 | 2021-10-28 | 세이지리서치 주식회사 | Computer program, method, and device for generating virtual defect image by using artificial intelligence model generated on basis of user input |
| CN113538631A (en) * | 2020-04-20 | 2021-10-22 | 贤智研究株式会社 | Computer program, method and apparatus for generating virtual defect image by artificial intelligence model generated based on user input |
| JP7531177B2 (en) | 2020-07-08 | 2024-08-09 | 日新電機株式会社 | Effluent water quality prediction device and effluent water quality prediction method |
| KR20220008530A (en) * | 2020-07-14 | 2022-01-21 | 한국생산기술연구원 | System and method of defect inspection using generation and transformation of input and output data based on deep learning |
| KR102475851B1 (en) | 2020-07-14 | 2022-12-09 | 한국생산기술연구원 | System and method of defect inspection using generation and transformation of input and output data based on deep learning |
| KR20220015910A (en) * | 2020-07-31 | 2022-02-08 | (주)딥인사이트 | Object-oriented data augmentation method |
| KR102393951B1 (en) * | 2020-07-31 | 2022-05-03 | (주)딥인사이트 | Object-oriented data augmentation method |
| KR102272497B1 (en) * | 2020-07-31 | 2021-07-02 | (주)딥인사이트 | Object-oriented data augmentation method |
| JP7392166B2 (en) | 2020-08-26 | 2023-12-05 | 三菱重工業株式会社 | Image generation device, image generation method and program |
| JPWO2022044150A1 (en) * | 2020-08-26 | 2022-03-03 | ||
| WO2022065271A1 (en) * | 2020-09-25 | 2022-03-31 | ファナック株式会社 | Image creation device |
| KR102621884B1 (en) | 2020-11-27 | 2024-01-05 | 씨제이올리브네트웍스 주식회사 | Deep learning-based image analysis method for selecting defective products and a system therefor |
| KR20220074021A (en) * | 2020-11-27 | 2022-06-03 | 씨제이올리브네트웍스 주식회사 | Deep learning-based image analysis method for selecting defective products and a system therefor |
| JP2022100292A (en) * | 2020-12-23 | 2022-07-05 | 株式会社リコー | Neural network training method, defect detection method and apparatus therefor, and recording medium |
| JP7243801B2 (en) | 2020-12-23 | 2023-03-22 | 株式会社リコー | Neural network training method and defect detection method, device and storage medium |
| JP2022134614A (en) * | 2021-03-03 | 2022-09-15 | Dmg森精機株式会社 | Substrate inspection method |
| JP7043645B1 (en) | 2021-03-03 | 2022-03-29 | Dmg森精機株式会社 | Board inspection method |
| CN113138198A (en) * | 2021-04-27 | 2021-07-20 | 环旭(深圳)电子科创有限公司 | System and method for generating defect image of electronic element |
| DE102022208849A1 (en) | 2021-09-30 | 2023-03-30 | Omron Corporation | IMAGE TESTING METHOD AND IMAGE TESTING DEVICE |
| WO2023100474A1 (en) * | 2021-12-02 | 2023-06-08 | 株式会社日立製作所 | System, image processing method, and program |
| WO2023175348A1 (en) * | 2022-03-17 | 2023-09-21 | Cheyney Design & Development Ltd. | Improvements in or relating to inspection and quality control |
| WO2023190045A1 (en) * | 2022-03-29 | 2023-10-05 | パナソニックＩｐマネジメント株式会社 | Image generation system, image generation method, and program |
| US11966219B2 (en) | 2022-03-30 | 2024-04-23 | Honda Motor Co., Ltd. | Pseudo defective product data generation method |
| WO2023238590A1 (en) * | 2022-06-10 | 2023-12-14 | 日立Astemo株式会社 | Pseudo defect image generation device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP2005156334A (en) | Pseudo defective image automatic creation device and imaging inspection device | |
| JP6936957B2 (en) | Inspection device, data generation device, data generation method and data generation program | |
| WO2019117065A1 (en) | Data generation device, data generation method and data generation program | |
| JP2019095217A (en) | Visual inspection device | |
| JP2019091249A (en) | Defect inspection device, defect inspecting method, and program thereof | |
| JP5441728B2 (en) | Sensory inspection device and sensory inspection method | |
| JP7372017B2 (en) | Steel component learning device, steel component estimation device, steel type determination device, steel component learning method, steel component estimation method, steel type determination method, and program | |
| JP2007047930A (en) | Image processor and inspection device | |
| WO2021161853A1 (en) | Analysis device and analysis method | |
| CN107194908A (en) | Image processing apparatus and image processing method | |
| JP4442042B2 (en) | Image processing program creation method and system | |
| CN114445746A (en) | Model training method, railway contact net abnormity detection method and related device | |
| JP2012018066A (en) | Device for inspecting abnormality | |
| CN106796180A (en) | For diversity and the dynamic lattice of disfigurement discovery | |
| JP2010164506A (en) | Inspection method | |
| TW202347181A (en) | System and method for defect detection | |
| CN114972375A (en) | Training method and device of image generation model, equipment and storage medium | |
| JP4704897B2 (en) | 3D measurement data inspection method using parametric tolerance | |
| JP2006113073A (en) | System and method for pattern defect inspection | |
| JP2002093875A (en) | Method of evaluating hazardous spot information on pattern of semiconductor device | |
| WO2022173037A1 (en) | Image generating method, machine learning method, image generating device, program, and training image generating method | |
| US20230204549A1 (en) | Apparatus and automated method for evaluating sensor measured values, and use of the apparatus | |
| JP2021128415A (en) | Machine learning method and information processing device for machine learning | |
| JP2003076976A (en) | Pattern matching method | |
| JP2010244320A (en) | Image processor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A621 | Written request for application examination |
Free format text: JAPANESE INTERMEDIATE CODE: A621 Effective date: 20061006 |
|
| A977 | Report on retrieval |
Free format text: JAPANESE INTERMEDIATE CODE: A971007 Effective date: 20090310 |
|
| A131 | Notification of reasons for refusal |
Free format text: JAPANESE INTERMEDIATE CODE: A131 Effective date: 20090407 |
|
| A521 | Request for written amendment filed |
Free format text: JAPANESE INTERMEDIATE CODE: A523 Effective date: 20090605 |
|
| A761 | Written withdrawal of application |
Free format text: JAPANESE INTERMEDIATE CODE: A761 Effective date: 20090904 |