Image defect classification method, device and electronic equipment

Abstract

The embodiment of the invention discloses an image defect classification method, device and electronic equipment, and relates to the technical field of data processing. The method includes: performing scaling processing on a target image to be subjected to defect detection to obtain a thumbnail corresponding to the target image; based on the extracted rough features, obtaining the thermal force corresponding to the target image by way of reverse gradient Figure; use the fine feature extraction branch in the skeleton network to perform feature extraction on the defect submap to obtain the fine features of the target image; through the branch feature weight adjuster, the coarse features and the fine features Weight adjustment is performed to obtain final target features, and the target features are input to the integrated classifier, so as to predict the defect category of the target image in the integrated classifier. Through the solution of the present application, the efficiency of image defect classification is improved.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to the image defect classification technology. Background technique [0002] In the defect detection and classification tasks, there are the following requirements and problems: [0003] 1) Defect detection requirements with complex scenarios: In the defect detection and classification requirements, such as the defect detection on the surface of the workpiece, the background and lighting conditions of the same part workpiece in different processes are inconsistent, resulting in the need for defect detection under different backgrounds need. [0004] 2) There are large defects and small defects in the image, and the existing image classification methods are not well suited for large and small defects at the same time. For large defects, features can be preserved in small images; while in small images after zooming, small defects will become smaller, and features are not eas...

AI Technical Summary

Problems solved by technology

[0004] 2) There are large defects and small defects in the image, and the existing image classification methods cannot be well applied to large defects and small defects at the same time

However, target detection requires manual labeling to calibrate the defect category and the position information of each defect in the graph

[0012] Existing scheme (a). Defect classification method without positioning algorithm: limited classification effect; Existing scheme (b). Defect classification method based on specific defect localization algorithm: scenes strongly dependent on pictures and defect localization algorithms, different scenes Pictures, which require different defect location algorithms; existing scheme (c). Target detection algorithm: requires a lot of manual labeling

Images