Defect Detection of Printed Circuit Board Based on Light-weight YOLOv5s

Abstract

For small target detection of circuit board defects, traditional detection methods have problems such as false detection, missed detection and slow detection speed. Although excellent models for detecting small targets exist in mainstream algorithms based on deep learning, their network structures are complex, computationally intensive and detection efficiency is low. In order to solve the above problems, this paper proposes a light-weight printed circuit board defect detection method based on YOLOv5s. The method uses an improved spatial pyramid pooling instead of the CSP module in the Backbone stage to provide multilevel perceptual fields while significantly reducing the computational effort. Secondly, a residual structure is introduced in Backbone and Neck to enable the network to obtain a better feature learning capability and improve the stability of training while accelerating the network convergence. Finally, the model parameters, calculation amount, mAP and AP with different defects of the improved algorithm in this paper are compared with other mainstream algorithms. The experimental results show that the improved algorithm in this paper greatly reduces the model parameters and calculation amount and improves the detection efficiency under the condition of high accuracy. Compared with other algorithms, it has obvious advantages and provides a new method for PCB defect detection.