轻量化YOLOv5s-OCG的轨枕裂纹检测算法

董超群,汪战,廖平,谢帅,荣玉杰,周靖淞

Lightweight YOLOv5s-OCG rail sleeper crack detection algorithm

Chaoqun DONG,Zhan WANG,Ping LIAO,Shuai XIE,Yujie RONG,Jingsong ZHOU

表 3 经典算法模型与轻量化模型对比实验结果

Tab.3 Comparison experimental results between classic algorithm models and lightweight models

Model Precision/% Recall/% mAP@0.5/% FPS/(帧·s−1) P/106 Faster-RCNN 41.9 43.8 32.6 12 137.10 SSD 40.3 54.4 38.1 31 26.23 CenterNet 43.6 56.1 41.8 70 32.62 RetinaNet 41.1 54.8 40.0 64 28.55 YOLOv3 42.5 59.7 41.9 88 9.31 YOLOv5s 44.4 59.7 42.9 94 7.03 YOLOv7-tiny 43.1 56.8 42.0 89 6.03 YOLOv5s-GhostNet 40.8 56.1 38.5 107 3.71 YOLOv5s-shufflenetv2 40.1 53.5 36.2 110 0.84 YOLOv5s-OCG 46.2 62.1 47.1 96 5.64