Defect detection method and defect detection device in industrial quality inspection

Abstract

The invention provides a defect detection method and a defect detection device in industrial quality inspection, and relates to the technical field of industrial quality inspection. The method includes: dividing a data set into a training set and a test set after polygon labeling; dividing the training set into input model for semantic segmentation, the segmentation model includes the reverse residual module stage and the receptive field alignment module; do gradient descent on the segmentation model and continuously update iterations to obtain the pre-segmentation model; obtain the segmentation results of the pre-segmentation model, and send Input the support vector machine for training to obtain the trained defect detection model; perform defect detection according to the trained defect detection model. The backbone network of the present invention adopts the reverse residual module and the receptive field alignment module, which can improve the final detection efficiency, solve the problem of inaccurate category classification caused by too small receptive field, and combine semantic segmentation with support vector machine to solve the problem of The semantic segmentation network is directly used for problems that are easy to overcheck when detecting defects.

Description

technical field [0001] The invention relates to the technical field of industrial quality inspection, in particular to a defect detection method in industrial quality inspection and a defect detection device in industrial quality inspection. Background technique [0002] Semantic segmentation convolutional network is a very commonly used deep learning technology in the field of industrial quality inspection. Not only can this type of model be directly used for pixel-level detection of defects, but it can also be used on the basis of ordinary target detection (only detection of target frames). Then apply this type of model to extract the edge for further analysis of the detection results. [0003] Currently, the most widely used semantic segmentation models are U-Net (an algorithm for semantic segmentation using fully convolutional networks), PSPNet (Pyramid Scene Parsing Network, Pyramid Scene Parsing Network), DeepLab (combining deep convolutional neural networks and Proba...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the above technical problems, the first purpose of the present invention is to propose a defect detection method in industrial quality inspection. The backbone network of the present invention adopts the reverse residual module and the receptive field alignment module, and the reverse residual module has parameters The characteristics of small amount and fast calculation can improve the final detection efficiency. The receptive field alignment module can solve the problem of inaccurate category classification caused by too small receptive field. At the same time, it will not increase too much calculation burden. Combined with the support vector machine, it solves the problem that the semantic segmentation network is easy to over-check when it is directly used for defect detection

Images