Image of FPC Defect Model Based on Global and Local Feature Fusion

Abstract

Aiming at the circuit defects caused by the manufacturing process problems of Flexible Printed Circuits (FPCs), a FPC defect image repair model based on global and local feature fusion is proposed. The model uses the Context Encoder (CE) as the framework, and com-pletes the basic repair within the defect area of the FPC by referring to the context infor-mation around the defect area of the FPC; then, by analyzing the local information repair details of the defective part, the conclusion is drawn that the receptive field is expanded, and the dilated convolution is used instead of the original ordinary convolution to im-prove the ability to extract local information. Then, we extracted global features from the entire FPC image, analyzed the prior information structure of the FPC image that is more consistent with sparsity, and integrated this information structure into the global feature extraction process to enhance the consistency and coherence of global information. In this way, the extracted local information and global information are fused to jointly guide the repair results to be more efficient. Finally, we conducted qualitative and quantitative experiments using existing popular methods and the method in this paper, respectively, and proved the effectiveness of the method in this chapter for large-scale defect repair from many aspects, meeting the requirements of repairing FPC image defects in actual industrial scenarios.