Industrial false and bad detection method and system

Abstract

The invention discloses an industrial false and bad detection method and system. The method comprises the following steps: acquiring a defect picture and corresponding defect type coded data; markingdefect positions of the pictures, and marking defect category codes, defect borders and defect real scales to obtain a training data set; training a deep learning target detection algorithm model by using the training data set to obtain a defect detection model; and identifying the defect position of the to-be-identified picture by using the defect detection model, and outputting the confidence level of the corresponding defect category code and the real scale of the defect. According to the method, a labeling method of a minimum false bad point target is provided, the real scale of the defectis predicted by adding a parallel full connection layer, and finally, the real scale information and the category information are subjected to service combination judgment through experience to improve the effect of the model on true and false bad detection and classification.

Description

technical field [0001] The invention relates to the field of industrial testing, in particular to a method and system for detecting industrial false defects. Background technique [0002] The existing automatic defect classification (auto defect classification, ADC) method is mainly implemented based on the deep learning framework, and the characteristics of different defects are learned through the image convolutional network; The defect features classify the detected defects. [0003] In ADC projects, multiple defect types may be extremely similar—especially in industrial false defect detection, where the scale distribution of defects is very wide; the scale of false defects is small, and true defects (NG) and false defects (OK) are in There is basically no distinction under the scale. If the existing data marking method is still used (the marking frame is the circumscribed rectangular frame of the detection target), and the image convolution network is used to directly ...

AI Technical Summary

Problems solved by technology

[0003] In ADC projects, multiple defect types may be extremely similar—especially in industrial false defect detection, where the scale distribution of defects is very wide; the scale of false defects is small, and true defects (NG) and false defects (OK) are in There is basically no distinction under the scale

If the existing data marking method is still used (the marking frame is the circumscribed rectangular frame of the detection target), and the image convolution network is used to directly detect and classify the true and false defects, the false defect points (false defects Points smaller than the stride of the network prediction layer feature map feature map) are likely to be missed, and the problem that true and false defects are similar at the same scale will also make it difficult for the model to distinguish

Images