基于特征复用机制的航拍图像小目标检测算法
|
邓天民,程鑫鑫,刘金凤,张曦月
|
Small target detection algorithm for aerial images based on feature reuse mechanism
|
Tianmin DENG,Xinxin CHENG,Jinfeng LIU,Xiyue ZHANG
|
|
表 4 不同算法在VisDrone数据集上的平均精度和参数量对比结果 |
Tab.4 Comparative results of different algorithms in average precision and parameters on VisDrone dataset |
|
模型 | AP/% | mAP0.5/% | P/MB | 行人 | 人 | 自行车 | 汽车 | 面包车 | 卡车 | 三轮车 | 遮阳蓬三轮车 | 巴士 | 摩托车 | 1)注:加粗字体为该列最优值,包含引用文献实验结果. | Faster R-CNN[18] | 20.9 | 14.8 | 7.3 | 51.0 | 29.7 | 19.5 | 14.0 | 8.8 | 30.5 | 21.2 | 21.8 | — | Cascade R-CNN[18] | 22.2 | 14.8 | 7.6 | 54.6 | 31.5 | 21.6 | 14.8 | 8.6 | 34.9 | 21.4 | 23.2 | — | YOLOv4[19] | 24.8 | 12.6 | 8.6 | 64.3 | 22.4 | 22.7 | 11.4 | 7.6 | 44.3 | 21.7 | 30.7 | — | YOLOv5s | 39.0 | 31.3 | 11.2 | 73.5 | 35.4 | 29.5 | 20.5 | 11.1 | 43.1 | 37.0 | 33.2 | 7.03 | YOLOv5l | 47.8 | 37.7 | 17.8 | 78.2 | 42.6 | 40.3 | 26.8 | 13.1 | 54.7 | 45.3 | 40.4 | 46.15 | Tph-YOLOv5[3] | 53.31) | 42.1 | 21.1 | 83.7 | 45.2 | 42.5 | 33.0 | 16.3 | 61.1 | 51.0 | 44.9 | 60.42 | YOLOX-l[20] | 34.8 | 24.5 | 16.9 | 72.4 | 34.4 | 40.5 | 23.1 | 17.8 | 53.1 | 36.0 | 35.3 | 54.16 | YOLOv7-tiny[21] | 37.9 | 34.6 | 9.4 | 76.1 | 36.3 | 29.8 | 20.1 | 10.6 | 43.2 | 41.8 | 34.0 | 6.03 | YOLOv8s | 44.3 | 34.5 | 15.6 | 80.4 | 46.1 | 37.9 | 29.5 | 15.7 | 58.5 | 46.6 | 40.9 | 11.13 | YOLOv8l | 51.1 | 39.8 | 21.9 | 82.9 | 49.3 | 45.4 | 38.1 | 20.4 | 67.0 | 52.3 | 46.8 | 43.69 | FS-YOLO | 53.1 | 43.6 | 18.9 | 85.0 | 51.4 | 41.3 | 35.3 | 21.1 | 66.0 | 54.9 | 47.0 | 5.48 |
|
|
|