Semi-supervised industrial image defect segmentation method based on adversarial generative network

Abstract

The invention discloses a semi-supervised industrial image defect segmentation method based on an adversarial generative network. A neural network is trained by using a small number of marked negativesamples with defects and a large number of positive samples without defects so as to obtain a segmentation network capable of automatically identifying the defects. In the construction process of theneural network, a segmentation network based on D-LinkNet and a reconstruction network based on U-net are respectively used, and feature spaces of a negative sample and a positive sample are separated in a cross training mode, so that the segmentation network can correctly segment defects in the negative sample. According to the method, dependence on industrial defect sample images can be greatlyreduced, and errors of a segmentation model during defect segmentation can be greatly reduced.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a semi-supervised industrial image defect segmentation method based on an adversarial generation network. Background technique [0002] In the process of industrial production, in order to ensure the quality of industrial products, one of the most important tasks is to detect defects on the product surface. Usually, the defects of product surface quality can only be effectively identified by trained workers. This method is inefficient and has low accuracy. In special cases, it will seriously restrict the production of enterprises. With the introduction of Industry 4.0 and the development of artificial intelligence technology, deep learning represented by data drive has been widely used. Deep learning can locate and identify defects only by relying on an appropriate number of marked pictures, and can quickly meet the needs of flexible production and rapid iteration of p...

AI Technical Summary

Problems solved by technology

This brings great challenges to the application of deep learning-based

Images