Research on Single-Stage Anchor-Free PCB Processing Defect Detection Method Based on Improved FCOS Algorithm

The processing defects of printed circuit board (PCB) are important factors causing the failure of electrical equipment. The processing defect detection of PCB must be followed up in real time. This paper Aims at the problems of too many manual designs, complex calculation process and unbalanced positive and negative samples in the PCB defect detection method based on two-stage Anchor, a PCB defect detection method based on Single-Stage Anchor-free is proposed. This method improves the FCOS algorithm, extracts features through the fully convolutional network, predicts the location of PCB defects through different levels of the feature pyramid networks, improves the accuracy of defect detection through three branch tasks of classification, regression and centrality, and adds an attention module based on swing transformer to the head. The whole process is Single-Stage, and PCB defect detection in the way of pixel level prediction. This method does not rely on the Anchor mechanism at all, and avoids the complex operation of Anchor. The experimental results show that this method can accurately detect all kinds of PCB defects, such as mouse bite, open circuit, short circuit, spur etc. The average detection accuracy of this method reaches 96.2%, and the detection speed is greatly improved compared with the method based on Two-Stage Anchor improves the detection speed. This method has certain theoretical value for the research of Anchor-free PCB defect detection method.