LSDM-PCB: A Lightweight Small Defect Detection Model for Printed Circuit Board

The defects of the Printed Circuit Board(PCB) directly affect the performance and reliability of electronic products. Therefore, detecting PCB defects is crucial. Lightweight models in PCB production inspection can effectively reduce equipment costs, but they exhibit limited feature extraction capabilities. Moreover, complex background conditions can interfere with the model’s ability to locate and recognize small defects. To address these challenges, we propose LSDM-PCB, a lightweight PCB defect detection model based on YOLOv8n. Firstly, we improve the network structure to reduce the number of model parameters while enhancing the model’s ability to capture small defects. Additionally, we adopt Receptive-Field Attention Convolution(RFAConv) as a downsampling module to enhance the model’s feature extraction by considering the importance of each feature within the receptive field. Finally, we propose a Global and Local Mixed Attention(GLMA) mechanism to strengthen multi-scale feature representation, allowing the model to focus more on small defects. Results show LSDM-PCB reduces model parameters by 74% and improves mAP
50
to 96.8%, a 2.7% enhancement compared to the baseline model YOLOv8n.