基于改进YOLOv7-tiny的铝型材表面缺陷检测方法

王浚银,文斌,沈艳军,张俊,王子豪

Surface defect detection method for aluminum profiles based on improved YOLOv7-tiny

Junyin WANG,Bin WEN,Yanjun SHEN,Jun ZHANG,Zihao WANG

表 6 不同算法指标对比实验结果

Tab.6 Comparative experimental results on different algorithm indicators

模型 AP/% mAP@0.5/% R/% Q/106 FLOPs/109 V/MB FPS/（帧·s−1） 不导电 脏点 漏底 凹陷 Faster-RCNN 97.4 72.8 95.8 81.8 86.9 62.3 41.30 214.0 315.0 16 YOLOv3-tiny 52.8 69.0 71.8 81.2 68.7 88.4 8.67 13.0 17.4 40 YOLOv5-s 99.3 79.9 99.4 91.2 92.5 88.8 7.00 16.0 14.4 45 改进的YOLOv5-s[28] 98.8 72.8 99.5 89.0 90.0 88.6 7.20 18.6 15.3 38 YOLOv7-tiny 99.5 69.6 99.8 92.2 90.3 88.5 6.02 13.2 12.3 50 YOLOv8-s 99.4 78.6 99.5 93.5 92.8 89.4 11.10 28.7 22.5 40 本研究模型 99.2 82.7 99.7 96.2 94.5 91.8 6.15 8.6 12.7 45