基于全局信息感知的轻量级螺纹钢表面缺陷检测算法

肖剑,杨小苑,何昕泽,陈林,胡欣

Lightweight rebar surface defect detection algorithm based on global information perception

Jian XIAO,Xiaoyuan YANG,Xinze HE,Lin CHEN,Xin HU

表 2 轻量级主干网络对比实验结果

Tab.2 Results of comparison experiment on lightweight backbone networks

模型 P/% R/% mAP@0.5/% mAP@0.5-0.95/% Np/M FLOPs/G FPS/(帧·s−1) YOLOv8n(baseline) 88.3 83.1 90.2 55.9 3.0 8.1 133 YOLOv8n-GhostNet 85.5 82.8 87.6 52.2 1.7 5.1 140 YOLOv8n-GhostNetV2 81.4 84.8 86.3 49.6 2.4 6.4 134 YOLOv8n-MobileViT 87.8 83.0 89.8 55.4 2.2 12.8 132 YOLOv8n-MobileNetV3 82.7 80.2 84.5 41.1 2.3 5.7 133 YOLOv8n-MobileNetV4 85.0 83.7 88.8 41.7 5.7 22.5 128 YOLOv8n-ShuffleNetV2 81.3 80.0 84.2 41.5 1.9 5.2 139 YOLOv8n-EfficientNetV2 82.9 78.8 84.5 42.0 2.1 2.6 148 YOLOv8n-VanillaNet 84.9 83.2 86.5 51.8 2.0 5.7 147 YOLOv8n-FasterNet 85.1 83.9 87.8 50.4 4.2 10.7 130 YOLOv8n-StarNet 82.0 78.9 83.7 45.9 2.2 6.5 137 CACGLUFormer 87.5 87.4 92.9 56.2 2.4 6.8 148