Template defect detection method

Abstract

A template defect detection method comprises the steps that small defects and good products obtained from a defect library are fused to obtain product defect data, and meanwhile corresponding labelingdata are generated, wherein the front half part of the network is a twinning network based on a residual network, and the rear half part of the network is a feature fusion and extraction network, theinput of the network is a good product and a corresponding generated defect sample, and the output of the network is a label and training model; a comparison network is obtained through training andlearning, the comparison network has the capability of comparing different positions of the two samples, and the output of the network is the data difference degree of the two input samples; and the difference degree between two products output by the network is compared so as to realize defect detection. Defect detection can be realized by comparing the difference degree between two products output by the network, the time cost of manual labeling is saved, the early-stage preparation work of model training is reduced, rapid switching detection of the products in work production is further realized, and the production efficiency of the products is improved.

Description

technical field [0001] The invention relates to the field of defect detection, in particular to a template defect detection method. Background technique [0002] For template defect detection, there are currently traditional algorithms and deep learning algorithms. The disadvantages of traditional algorithms are the long development cycle and poor generalization ability of the algorithm. For some complex samples, the development of traditional algorithms is difficult. At the same time, slight changes in product shape, size or color may make the current algorithm no longer applicable, requiring professional algorithms. People are developing new algorithms again, which leads to long maintenance cycles and high costs for traditional algorithms. [0003] The basic operation process of the current deep learning algorithm applied to template defect detection is as follows. First, a large amount of defective product data and non-defective product data need to be collected, and the...

AI Technical Summary

Problems solved by technology

The disadvantages of traditional algorithms are the long development cycle and poor generalization ability of the algorithm. For some complex samples, the development of traditional algorithms is difficult. At the same time, slight changes in product shape, size or color may make the current algorithm no longer applicable, requiring professional algorithms. People are developing new algorithms again, which leads to long maintenance cycles and high costs for traditional algorithms

[0003] The basic operation process of the current deep learning algorithm applied to template defect detection is as follows. First, a large amount of defective product data and non-defective product data need to be collected, and then the corresponding data labels are manually marked. Finally, the training model is applied to the industrial production line. However, before The two-step work consumes a lot of manpower and time, which limits the rapid replacement of deep network models, which reduces production efficiency to a certain extent

[0004] At the same time, for some products, the yield rate is extremely high and the defective rate is extremely low. It is extremely unrealistic to collect a large amount of defect data, otherwise it will cause serious errors of sample data imbalance. Applying Deep Learning Algorithms

Images