YOLO-SUMAS: Improved Printed Circuit Board Defect Detection and Identification Research Based on YOLOv8.

Abstract

Aiming at the demand for defect detection accuracy and efficiency under the trend of high-density and integration in printed circuit board (PCB) manufacturing, this paper proposes an improved YOLOv8n model (YOLO-SUMAS), which enhances detection performance through multi-module collaborative optimization. The model introduces the SCSA attention mechanism, which improves the feature expression capability through spatial and channel synergistic attention; adopts the Unified-IoU loss function, combined with the dynamic bounding box scaling and bi-directional weight allocation strategy, to optimize the accuracy of high-quality target localization; integrates the MobileNetV4 lightweight architecture and its MobileMQA attention module, which reduces the computational complexity and improves the inference speed; and combines ASF-SDI Neck structure with weighted bi-directional feature pyramid and multi-level semantic detail fusion to strengthen small target detection capability. The experiments are based on public datasets, and the results show that the improved model achieves 98.8% precision and 99.2% recall, and mAP@50 reached 99.1%, significantly better than the original YOLOv8n and other mainstream models. YOLO-SUMAS provides a highly efficient industrial-grade PCB defect detection solution by considering high precision and real-time performance while maintaining lightweight characteristics.