Industrial image defect detection method and system based on multi-task twin network

Abstract

The invention relates to the technical field of image processing, and particularly discloses an industrial image defect detection method based on a multi-task twin network. According to the method, a to-be-detected image and a corresponding template image are jointly used as network input, related features of differences between the to-be-detected image and the corresponding template image are calculated, a segmentation network structure is used for carrying out multi-task auxiliary training, a backbone network can clearly and effectively distinguish background information in the FPN feature extraction process, and then FPN features are used for carrying out detection network training, thereby completing a defect dichotomy detection task. According to the invention, a twin network structure is adopted to provide effective guarantee for the generalization ability of the model, the multi-task segmentation network assists in detection network training, so network training is easier, performance is better, and support is provided for dichotomy detection tasks.

Description

technical field [0001] The invention relates to the technical field of image detection, in particular to an industrial image defect detection method and system based on a multi-task twin network. Background technique [0002] With the rapid development of computer vision technology and the wide application of artificial intelligence technology, machine vision technology is more and more used in industrial scenes, and it is widely used in all aspects of production. Defect detection computer vision is one of the scenes where the demand for computer vision has increased sharply in the industrial field in recent years. This task is mainly reflected in the fact that computer vision can quickly and automatically detect product appearance images through data collected by image sensors, and return whether the product has defects and Defect related information. In the defect detection process, the accuracy index of product classification is very important, which directly affects the...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide an industrial image defect detection method and system based on a multi-task twin network, which has solved the defects of insufficient model generalization and insufficient depth of twin network defect detection technology in the above-mentioned prior art

Images