An optical element defect detection method and system based on electromagnetic field prior constraint

Abstract

The application provides the technical field of optical element surface defect detection and relates to an optical element defect detection method and system based on electromagnetic field prior constraints, which comprises the following steps: S1, acquiring multispectral original image data of an optical element surface; S2, processing the multispectral original image data through a physical information feature extraction module to generate a physical feature map; S3, injecting the physical feature map into a backbone network of a physically enhanced neural network; S4, performing forward propagation on the physically enhanced neural network to generate a defect prediction probability map; S5, performing physical constraint verification on the defect prediction probability map and calculating a physical loss based on electromagnetic field theory; and S6, constructing a target function based on the physical loss and a traditional data-driven loss, optimizing network parameters of the physically enhanced neural network through back propagation, and outputting a defect segmentation result. The application has the advantages that the physical rationality and robustness of optical element defect detection are greatly improved.

Images