Top > Search of International Patents > ANOMALOUS ITEM DETERMINATION METHOD

ANOMALOUS ITEM DETERMINATION METHOD meetings

Foreign code F190009839
File No. (GI-H29-18)
Posted date Jul 25, 2019
Country WIPO
International application number 2018JP037352
International publication number WO 2019073923
Date of international filing Oct 5, 2018
Date of international publication Apr 18, 2019
Priority data
  • P2017-196758 (Oct 10, 2017) JP
Title ANOMALOUS ITEM DETERMINATION METHOD meetings
Abstract Provided is an anomalous item determination method with which it is possible to determine an anomalous item accurately by performing machine learning using a large amount of normal data and a small quantity of anomalous data. Data relating to a plurality of items to be determined are input into an encoder-decoder structure network, features of the items to be determined are extracted, and a discriminator determines whether the distribution of the features of the items to be determined is in accordance with a normal distribution. Updating of the encoder-decoder structure network, updating of the discriminator, and updating of an encoder are each repeated to minimize a feature extraction error. The encoder, using a feature obtained by the updating, calculates an anomaly degree of the items to be determined, subjects the anomaly degree to threshold value processing, and determines whether the items to be determined are normal items or anomalous items. The step of determining whether the distribution of the features of the items to be determined is in accordance with a normal distribution comprises a step of inputting data in accordance with a normal distribution into the discriminator and calculating an error between the data and the features of the items to be determined extracted by the encoder-decoder structure network. Using the result of determination by the discriminator allows the features of the items to be determined that are used by the encoder for anomaly degree calculation to converge so as to be distributed in accordance with a normal distribution.
Outline of related art and contending technology BACKGROUND ART
By repeated learning data into the computer, as a mathematical equation or numerical feature data contained in the statistical computer is extracted, further, using the extracted features, the machine learning as a method of identification.
Machine learning as one method, referred to as the auto-encoder of the encoder and the (self-encoder), using the network decoder structure has been known a method of extracting the feature amount. And the auto-encoder, the input and output are the same as in the neural network to learn. Encoder input once the one-dimensional feature by a small chute, the decoder input so as to reproduce among the repeated output, the feature amount is extracted which represent the input.
Using the auto-encoder of the normal product if the feature extraction, using this feature, a normal product and the defective products from an aggregate of the determination target object are mixed, the defective products is determined with high accuracy and can be extracted.
Non-Patent Document 1 is, a kind of neural network 'Convolutional Neural Network (hereinafter, also referred to as a convolutional neural network) ' discloses a technique related to. CNN is, mainly used in the field of image recognition in a neural network, a local feature extraction of the convolution layer is responsible for, for each local feature in the feature pooling layers are summarized in a feature of the repeated structure. In general, including CNN for learning of the neural network, supervised learning using a large amount of training samples is required. However, the number of samples of the defective products is learned when it is difficult to secure a sufficient number is, the learning problem that cannot be performed successfully.
Non-Patent Document 2 is, a kind of neural network 'Autoencoder (hereinafter, the auto-encoder, also referred to as self-encoder) ' discloses a technique related to. Non-Patent Document 2 discloses a neural network, a multi-layered neural network parameters are initialized after unsupervised learning, supervised learning and relearning. Non-Patent Document 2 is of the auto-encoder, compressed two-dimensional input, wherein the abstract of the input vector quantity is converted into a feature vector, the feature vector from the reproduced input. However, the auto-encoder as to how to obtain and the shape distribution, it cannot be manipulated.
Non-Patent Document 3 is, a kind of neural network 'Adversarial Autoencoder (hereinafter, also referred to as a hostile self-encoder) ' discloses a technique related to. The encoder is self-deception, a hostile to auto-encoder incorporating learning, while the features are extracted which represent the input, in order to follow the feature distribution of an arbitrary technique.
Non-Patent Document 4 is, the Hotelling's T2 is a literature which disclosed method. T2 Method, only a large amount of normal data, abnormal data or a large amount of normal data and a small amount of the feature vector used to create a model from the normal, the abnormality level of each of the unknown data is computed, is a statistical method for detecting abnormal data. However, the distribution of the feature amount of the normal distribution is in accordance with the assumption that, if the data does not follow the normal distribution, sufficient detection cannot be performed. Hotelling T2 is applied to the image recognition method in the art, according to the normal distribution needs to be selected features.
Scope of claims (In Japanese)[請求項1]
 エンコーダ、デコーダ構造のネットワークとディスクリミネータのネットワークとを用いて敵対的学習を行い、判定対象物が正常品であるか異常品であるかを判定する判定方法であって、
 複数の判定対象物のデータを前記エンコーダ、デコーダ構造ネットワークに入力して、前記判定対象物の特徴を抽出する工程と、
 前記ディスクリミネータが、前記判定対象物の前記特徴の分布は正規分布に従っているのか否かを判定する工程と、
 前記エンコーダ、デコーダ構造ネットワークの更新と、前記ディスクリミネータの更新と、前記エンコーダの更新と、をそれぞれ繰り返し、前記特徴の抽出の誤差を最小化する工程と、
 前記エンコーダが、更新によって得られた前記特徴を用いて、判定対象物の異常度を算出する工程と、
 算出した前記異常度のしきい値処理を行うことによって、前記判定対象物が正常品であるか異常品であるかを判定する工程と、
 を備えており、
 前記ディスクリミネータが、前記判定対象物の前記特徴の分布は正規分布に従っているのか否かを判定する工程は、ディスクリミネータに正規分布に従ったデータを入力し、前記データと前記エンコーダ、デコーダ構造ネットワークが抽出した前記判定対象物の前記特徴との間の誤差を算出する工程であり、
 前記ディスクリミネータの判定結果を用いていることで、前記エンコーダが異常度の算出に用いる前記判定対象物の前記特徴が正規分布に従って分布するように収束させられていることを特徴とする異常品の判定方法。

[請求項2]
 前記特徴を抽出するために前記エンコーダ、デコーダ構造ネットワークに入力する複数の判定対象物のデータが、異常品よりも正常品を多く含むデータであることを特徴とする請求項1に記載の異常品の判定方法。

[請求項3]
 前記ディスクリミネータに入力する正規分布に従った前記データは、多変量の標準正規分布に従ったランダムベクターであることを特徴とする請求項1または2に記載の異常品の判定方法。

  • Applicant
  • ※All designated countries except for US in the data before July 2012
  • GIFU UNIVERSITY
  • Inventor
  • KATO Kunihito
  • NAKATSUKA Shunsuke
  • AIZAWA hiroaki
IPC(International Patent Classification)
Specified countries National States: AE AG AL AM AO AT AU AZ BA BB BG BH BN BR BW BY BZ CA CH CL CN CO CR CU CZ DE DJ DK DM DO DZ EC EE EG ES FI GB GD GE GH GM GT HN HR HU ID IL IN IR IS JO JP KE KG KH KN KP KR KW KZ LA LC LK LR LS LU LY MA MD ME MG MK MN MW MX MY MZ NA NG NI NO NZ OM PA PE PG PH PL PT QA RO RS RU RW SA SC SD SE SG SK SL SM ST SV SY TH TJ TM TN TR TT TZ UA UG US UZ VC VN ZA ZM ZW
ARIPO: BW GH GM KE LR LS MW MZ NA RW SD SL SZ TZ UG ZM ZW
EAPO: AM AZ BY KG KZ RU TJ TM
EPO: AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR
OAPI: BF BJ CF CG CI CM GA GN GQ GW KM ML MR NE SN ST TD TG
Please contact us by E-mail or facsimile if you have any interests on this patent.

PAGE TOP

close
close
close
close
close
close