Lightweight PCB defect detection algorithm and deployment based on ASF-YOLO

With the advancement in the precision of printed circuit boards (PCBs), their impact on electronic products has become increasingly significant. PCBs often exhibit various surface defects such as short circuits, open circuits, and burrs, which can have devastating effects on electronic products. To address these defects, this study proposes a lightweight detection model, ASC-YOLO. The dataset used was sourced from a public dataset on Roboflow, comprising 9,669 images. The Adown module enhances the model’s ability to recognize defects of different sizes and types while effectively reducing computation and parameter volume. SENetV2 improves the network’s learning capacity through multi-branch fully connected layers. Finally, PConv is introduced to reduce redundant computation and memory access, efficiently extracting spatial features. Compared to the original model, the parameters were reduced by 28%, and the accuracy reached 94.8%. Moreover, the model was deployed on mobile terminals, achieving high detection accuracy and efficiency. This advancement makes fully automated, high-precision AI-based detection possible.