DR Image Defect Detection Method of Pressed Joints Based on Improved YOLOv8

To address the challenges of high complexity and workload in manual DR defect detection, this article presents EBW-YOLO, a DR image defect detection algorithm based on YOLOv8n. YOLOv8n, known for its real-time performance and high accuracy, serves as the benchmark network. The proposed method incorporates an efficient channel attention (ECA) mechanism into the C2f module of the shallow backbone to enhance key defect feature extraction, while a bidirectional feature pyramid network (BiFPN) is employed in the neck to optimize multi-scale feature processing and fusion. Additionally, the normalized Wasserstein distance(NWD) loss function replaces the conventional CIoU loss to improve small target detection. According to the training and verification tests on the self-built data set, the improved convolutional network EBW-YOLO is applied to the detection and identification of the defects of the crimped splints. The values of [email protected] and [email protected]:0.95 reach 91.9% and 74.1% respectively, and the FPS reaches 50, meeting the requirements of actual engineering inspection.