Product appearance defect detection method based on random defect model

Abstract

The invention discloses a product appearance defect detection method based on a random defect model. The method comprises the following steps: for a to-be-detected product, collecting good product pictures without appearance defects to form a good product picture set; based on the good product picture set, utilizing a two-dimensional random medium model to generate an autocorrelation model graph meeting set autocorrelation model parameters; generating a defect probability distribution picture set meeting set defect morphological parameters based on the self-correlation model graph; fusing the generated good product pictures and the defect probability distribution diagram to obtain a defect picture set and a corresponding marking picture set; training a deep learning neural network by using the defect picture set and the labeled picture set; and performing appearance defect detection on a to-be-detected product by using the trained deep learning neural network. According to the method, the defect picture set can be automatically generated for deep learning network training only by using the good product pictures, and the method can be applied to product quality detection of an industrial production line.

Description

technical field [0001] The present invention relates to the technical field of product quality inspection, and more specifically, to a product appearance defect detection method based on a random defect model. Background technique [0002] In the current automated production process, product quality testing is widely used in various fields such as electronic components, semiconductor devices, packaging and printing, food and beverage, and medical testing. [0003] In the prior art, the detection of product appearance defects based on deep learning algorithms requires manually labeled qualified product pictures and appearance defect pictures for training neural networks. This approach usually requires long-term repeated training of the model for iterative upgrades. During the iterative upgrade process, it is necessary to collect a large number of defect pictures and require manual defect labeling. Moreover, the number of defect pictures and the accuracy of manual annotation...

AI Technical Summary

Problems solved by technology

Moreover, the number of defect pictures and the accuracy of manual annotation directly affect the detection effect

However, in the actual electronic product production line, the number of good products will be far greater than the number of defective products, and the shape distribution of the actually collected defective products is usually not typical, for example, it does not have the characteristics of random distribution

In addition, due to the characteristics of fast replacement and short appearance change cycle of electronic products, the existing deep learning-based algorithms cannot be quickly deployed to actual engineering applications, and it is difficult to implement them in practical applications.

Images