Defect detection network construction method, anomaly detection method and system, and storage medium

Abstract

The invention relates to a defect detection network construction method, an anomaly detection method and system, and a storage medium. The construction method comprises: obtaining a sample image in apreset open data set and a reference image of a standard product of a to-be-detected object; configuring a convolutional neural network module or a plurality of convolutional neural network modules with different scales and forming a defect detection network; training a main feature extraction model by using the sample image to obtain corresponding network parameters, and inputting the reference image into the main feature extraction model and the slave feature extraction model to obtain a corresponding first feature vector and a corresponding second feature vector respectively; and constructing a loss function of the slave feature extraction model according to the first feature vector and the second feature vector, and training and learning to obtain network parameters of the slave feature extraction model, thereby configuring and forming a defect detection network. The defect detection network can complete automatic extraction of features according to input image information, effectively reduces dependence on experience of workers in the process of product defect detection, and has practical value.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a defect detection network construction method, an anomaly detection method and device, and a storage medium. Background technique [0002] In the process of quality control of industrial products, it is better to have over-inspection than omission of inspection. The surface flaw detection method in industrial applications can often be regarded as the application of image-based object surface anomaly detection technology, which can be understood as "a multi-classification problem of unbalanced data sets under unsupervised or weak supervision", which is different from balanced learning. Anomaly detection is generally in an unsupervised form, which seems to be a binary classification but is actually a multi-classification (the reasons for the anomalies are different). Without knowing how many classes there are in fact and without real labels, anomaly detection often does ...

AI Technical Summary

Problems solved by technology

At present, there are still the following problems in the industrial deployment of object surface anomaly detection applications: it is easy to obtain good product samples in the industry, and it is difficult to obtain defective samples in the production process, resulting in unbalanced sample data caused by many good product samples and few defective samples; Defects such as scratches are related to the shooting angle during the shooting process, and the defects of the object can only be reproduced in a specific direction; although the causes of sample defects are different, they are relatively similar in image presentation, making it difficult to distinguish which one belongs to class one defect

However, most similar methods face the curse of dimensionality, that is, common similarity measures tend to fail on high-dimensional data

For high-dimensional data, the purpose is to find a good space / representation, and then finding anomalies becomes a simple problem of measuring similarity. Another problem brought about by high-dimensional data is scalability. It is well known that the system for measuring similarity The computational overhead is high, for example, the complexity of the distance measurement is above O(n). At this time, it is necessary to use data structure optimization or dynamic programming to reduce the complexity, which brings difficulties to control the data dimension

[0004] The currently applied surface anomaly detection methods still have the following deficiencies: (1) The feature extraction network is unstable and cannot achieve better cross-domain migration; (2) In order to extract better features, the network structure design is more complicated, resulting in The calculation time is long, which cannot meet the requirements of real-time detection and distributed detection; (3) when the traditional method is used for anomaly detection, the feature selection depends on the engineer's algorithm design experience, which requires high human experience, and it is easy to fail in manual detection; (4) ) feature simplification, poor defect detection stability

Images