Exploring Lightweight YOLOv8 Architectures for Circuit Board Defect Detection

Modern electronics relies heavily on circuit boards, which must be designed and produced using a variety of procedures in order to be filled with electrical components. This procedure includes drilling and plating holes, applying masks, painting and etching fiberglass-reinforced epoxy resin boards, and adding components to the boards. Because of variations in the materials and processes, mistakes might occur at any stage. Traditionally, conductivity, impedance, and visual testing are used in a multi-stage approach to identify and resolve these problems. With modest hardware requirements, computer vision, especially using convolutional networks, provides considerable promise to improve defect detection. Once implemented, these models may significantly increase error detection and lower board failure rates, despite the significant hardware requirements for training. Using a publicly accessible dataset, this study investigates the use of the YOLO model for fault identification in circuit board fabrication. The YOLOv8 model’s lightweight design is examined; the medium-sized model achieves a mAP@50 score of 0.990, indicating that it considerably improves fault detection.