1 基于轻量化网络改进的YOLOv5模型构建
1.1 改进YOLOv5模型
1.2 建立FasterNet主干特征提取网络
表1 不同网络对比实验Tab.1 Comparative experiments of different networks |
| 模型 | P/% | R/% | mAP@ 0.50/% | FPS (f/s) | PAR/106 | 模型大小 (MB) |
|---|---|---|---|---|---|---|
| YOLOv4 | 91.7% | 89.8% | 95.3% | 58 | 64.3 | 245.53 |
| GhostNet-YOLOv4 | 94.58% | 94.56% | 95.73% | 66 | 11.4 | 42.8 |
| YOLOv5s | 98.5% | 95.4% | 97.9% | 63 | 7.06 | 14.4 |
| YOLOv5s-Ghostv2 | 96.7% | 96.5% | 97.22% | 70 | 6.46 | 12.7 |
| F-YOLOv5s | 97.0% | 97.7% | 99.0% | 61 | 5.81 | 11.3 |
1.3 引入SimAM注意力机制
表2 注意力机制对比实验Tab.2 Comparative experiment of attention mechanism |
| 模型 | P/% | R/% | mAP@0.50/ % | mAP@ 0.50:0.95/% | FPS (f/s) | 模型大小 (MB) |
|---|---|---|---|---|---|---|
| YOLOv5s | 98.5% | 95.4% | 97.9% | 71.1% | 63 | 14.4 |
| S-YOLOv5s | 96.7% | 98.1% | 98.8% | 71.8% | 62 | 13.7 |
2 实验评价
2.1 实验环境及评价指标
2.2 实验结果分析
表3 实验对比结果Tab.3 Experimental comparison results |
| Number | Method | P/% | R/% | mAP@ 0.5/% | mAP@ 0.5: 0.95/% | F1 | FPS (f/s) | 模型 (MB) |
|---|---|---|---|---|---|---|---|---|
| A | YOLOv5s | 98.5% | 95.4% | 97.9% | 71.1% | 96.9% | 63 | 14.4 |
| B | S-YOLOv5s | 95.9% | 97.5% | 97.8% | 71.6% | 97.3% | 63 | 13.7 |
| C | F-YOLOv5s | 96.7% | 98.1% | 98.8% | 70.8% | 97.3% | 61 | 11.3 |
| D | SF-YOLOv5s | 97.5% | 97.5% | 99.3% | 70.3% | 97.5% | 64 | 11.3 |
中国指挥与控制学会会刊 