双重轻量化PCB缺陷检测算法研究OA

This paper proposes a dual lightweight PCB defect detection algorithm to address the issues of slow detection speed and high requirements for deployment equipment in PCB defect detection methods.Firstly,a lightweight module C3Ghost is adopted in the YOLOv5 backbone network.Then,a feature fusion network is constructed using the GSConv module and C3GS module to obtain partial semantic information lost in the backbone network and improve network detection speed.Finally,multi task global channel pruning is used to prune channels that have a small impact on network accuracy,further reducing the model's parameter and computational complexity.This algorithm was tested on the PKU-Market-PCB dataset,with an average accuracy of 98.9%,a model size of 5.2M,a model parameter count of 2393469,and a detection time of 3.3ms.Compared with the original algorithm,its model size,model parameter count,and detection time were reduced by 64%,66%,and 25%,respectively.