基于动态核感知的无人机视角路面病害检测方法
|
|
张建刚,李肖,冯丹丹
|
Dynamic kernel perception for pavement distress detection in UAV inspection
|
|
Jiangang ZHANG,Xiao LI,Dandan FENG
|
|
| 表 3 各改进模块的消融实验结果 |
| Tab.3 Ablation study results of different improved modules |
|
| 模型 | P/% | R/% | mAP@0.5/% | mAP@0.5:
0.95/% | F1/% | FPS/(帧·s−1) | | 1）注：加粗字体表示每列中的最优值 | YOLO12n
(基线模型) | 78.0 | 71.7 | 76.0 | 48.3 | 74.7 | 212.7 | | +A | 81.7 | 69.7 | 75.9 | 48.5 | 75.2 | 217.5 | | +B | 91.2 | 74.7 | 82.1 | 55.4 | 82.1 | 197.4 | | +C | 89.8 | 80.1 | 84.6 | 60.4 | 84.6 | 236.1 | | +A+B | 89.4 | 76.5 | 82.3 | 55.1 | 82.4 | 201.3 | | +A+C | 90.3 | 78.4 | 84.0 | 58.5 | 83.9 | 239.7 | | +B+C | ${{\boldsymbol{92.3}}}^{1)} $ | 78.3 | 85.0 | 60.7 | 84.7 | 189.8 | | +A+B+C | 91.9 | 80.0 | 84.5 | 59.6 | 85.5 | 197.2 | | +A+B+C+D | 90.4 | 80.4 | 84.9 | 61.9 | 85.1 | 225.8 |
|
|
|