基于改进YOLOv8的3D打印混凝土表观缺陷检测方法

田卫,周菻鈜,李欣阳,王建明,黄余康

3D-printed concrete apparent defect detection method based on improved YOLOv8

Wei TIAN,Linhong ZHOU,Xinyang LI,Jianming WANG,Yukang HUANG

表 5 各模型对比实验结果

Tab.5 Comparative experimental results for each model

算法 AP@0.5/% $ P $/% $ R $/% mAP@0.5/% FPS/帧 voids pitted-surface fracture collapse Faster R-CNN[24] 47.9 87.2 84.2 94.8 56.9 81.4 78.5 19 YOLOv3-tiny[25] 79.2 90.0 88.7 62.7 88.8 74.1 80.1 116 YOLOv5n 78.7 91.0 88.2 68.0 88.0 78.0 81.5 157 YOLOv8 76.3 86.8 95.2 95.5 83.9 82.9 88.5 143 YOLOv8-ELB[13] 75.9 87.1 93.6 93.7 82.3 81.5 87.6 138 YOLOv8-SRA 87.2 96.3 96.4 98.9 91.3 92.6 94.7 124