A PCB Defect Detection Algorithm Based on Improved Yolov7-tiny

In order to solve the problems of PCB surface defect detection, such as slow speed, long detection time and missed detection, a PCB surface defect detection algorithm based on improved Yolov7-tiny was proposed. The REPVGG structure is introduced into the conv, and the multi-channel structure of the training network is transformed into the single-channel structure of the inference network by the thought of network re-parameterization, so as to improve the inference speed of the model. Se attention, a visual channel attention mechanism, is incorporated into conv, and a 1 × 1 × c weight matrix is obtained by extrusion and excitation operations, enhance the ability of model precision extraction. In CBL Network Module, PRelu activation function is used instead of LeakyRelu to optimize the extraction precision. The experimental results show that the MAP@0.5 of the model can reach 98.6% on the open source PCB defect data set of Peking University.