A Method for Extracting Planar Image Features Based on a Multi-Scale Feature Backtracking Network in Surface Machining

Abstract

This paper is supported by the Sichuan Science & Technology Program under Grant 2022YFG0070. Image feature extraction technology plays a very important role in the measurement and control process of industrial product processing, but currently it has problems such as low recognition accuracy. Aiming at the structural characteristics of the feature pyramid network, combined with the characteristics and requirements of image feature extraction in the measurement and control processing process, this paper proposes a planar feature extraction method based on a multi-scale feature backtracking network structure, that is, by introducing weights based on background judgment in the feature pyramid structure, accurate prediction of target location is achieved. We used the PCB Defect Dataset to verify and test the algorithm. The test results show that compared with other models, our model has improved detection accuracy by 10.7%.