基于细节增强和跨尺度注意力融合的PCB缺陷检测算法

刘春娟,包易航,吴小所,闫浩文,朱铁栓,赵浩然

PCB defect detection algorithm based on detail enhancement and cross-scale attention fusion

Chunjuan LIU,Yihang BAO,Xiaosuo WU,Haowen YAN,Tieshuan ZHU,Haoran ZHAO

表 5 不同PCB缺陷检测算法的性能对比结果

Tab.5 Performance comparison of different PCB defect detection algorithms

算法 NP P/% R/% AP50/% mAP50/% GFLOPs RF/(帧·s−1) 缺孔 鼠咬 断路 短路 毛刺 残铜 Faster-RCNN 136 851 734 88.3 81.6 93.0 81.4 82.4 91.9 80.4 85.3 85.7 271.4 21.7 SSD 26 275 485 79.2 71.8 85.6 71.1 75.6 81.2 65.4 79.0 76.3 62.7 90.2 YOLOv8m 25 843 234 95.7 88.2 99.4 92.3 90.6 97.8 84.0 96.6 93.5 78.7 79.8 YOLOv9m 20 159 266 95.3 86.8 99.3 89.7 87.7 97.0 84.2 94.6 92.1 77.0 80.2 YOLOv10m 15 316 642 94.4 86.7 99.3 89.4 87.1 96.5 83.4 94.7 91.7 58.9 83.1 RT-DETR-l 31 996 326 96.9 93.8 97.5 96.8 92.2 95.7 93.8 99.3 95.9 103.5 46.2 RT-DETR-R18 19 879 464 96.3 91.9 96.5 96.1 91.7 93.6 91.5 98.8 94.7 57.0 70.8 DCFA-DETR 20 401 024 97.8 94.3 97.8 97.7 94.6 96.9 95.4 99.5 97.0 61.6 84.4