显著性偏振参量深度稀疏特征学习的目标检测方法 
下载： 1132次
Object Detection by Deep Sparse Feature Learning of Salient Polarization Parameters
1 安徽新华学院信息工程学院, 安徽 合肥 230088
2 中国人民解放军陆军炮兵防空兵学院偏振光成像探测技术安徽省重点实验室, 安徽 合肥 230031
图 & 表
图 1. 目标检测框架
Fig. 1. Object detection framework
下载图片 查看原文
图 2. 显著性偏振参量图像目标检测算法
Fig. 2. Object detection algorithm for salient polarization parameter image
下载图片 查看原文
图 3. 测试数据(0°方向)。(a)民航飞机;(b)草地坦克;(c)沙地卡车
Fig. 3. Test data (0° polarization direction). (a) Airplane; (b) tank; (c) truck
下载图片 查看原文
图 4. 显著性参量图像选择结果。(a1)~(c1)飞机1,飞机2,卡车图像;(a2)~(c2)显著性选择结果
Fig. 4. Results of salient parameter image selection. (a1)-(c1) Images of airplane 1, airplane 2, and truck; (a2)-(c2) salient selected results
下载图片 查看原文
图 5. 目标检测结果。(a1)~(c1)飞机1,飞机2,卡车图像;(a2)~(c2)显著性图像检测结果
Fig. 5. Results of object detection. (a1)-(c1) Images of airplane 1, airplane 2, and truck; (a2)-(c2) salient image detection results
下载图片 查看原文
表 1偏振解析显著性参量图像选择前后评价结果
Table1. Results before and after polarization analysis and salient parameter image selection
| Criteria | Airplane 1 | Airplane 2 | Truck |
|---|
| 0° | Salient | 0° | Salient | 0° | Salient |
|---|
| En | 3.75 | 6.84 | 5.29 | 6.54 | 6.22 | 6.77 | | 0.31 | 0.29 | 0.36 | 0.39 | 0.87 | 0.61 | | σ | 5.55 | 29.94 | 30.22 | 25.65 | 75.69 | 44.19 |
|
查看原文
表 2不同图像目标检测结果比较
Table2. Comparison of detection results of different image objects
| Object | mAP | AP |
|---|
| Airplane 1 | Airplane 2 | Tank 1 | Truck | Tank 2 |
|---|
| Polarization angle of 0° | 63.88 | 69.3 | 61.2 | 71.4 | 66.4 | 51.1 | | Polarization angle of 60° | 63.08 | 68.5 | 59.8 | 70.6 | 65.9 | 50.6 | | Polarization angle of 120° | 63.42 | 69.0 | 60.3 | 70.8 | 66.3 | 50.7 | | I | 63.72 | 70.2 | 60.1 | 72.3 | 65.8 | 50.2 | | Q | 60.64 | 64.4 | 57.4 | 68.8 | 61.3 | 51.3 | | U | 53.7 | 64.9 | 37.6 | 62.9 | 54.4 | 48.7 | | P | 60.46 | 70.3 | 45.3 | 70.1 | 67.1 | 49.5 | | A | 52.24 | 59.4 | 37.7 | 58.8 | 60.7 | 44.6 | | Ex | 61.18 | 70.0 | 57.6 | 67.0 | 63.5 | 47.8 | | Ey | 59.56 | 64.7 | 56.8 | 56.2 | 67.9 | 52.2 | | ΔE | 57.74 | 64.6 | 53.6 | 55.3 | 65.4 | 49.8 | | β | 31.85 | N | N | N | 23.4 | 40.3 |
|
查看原文
表 3不同模型的检测结果比较
Table3. Comparison of detection results of different models
| Model | Time /s | mAP/average | Airplane 1 | Airplane 2 | Tank |
|---|
| AP/score | AP/score | AP/score |
|---|
| Faster R-CNN | 0.7 | 66.1/0.741 | 68.1/0.842 | 59.5/0.619 | 70.6/0.762 | | Proposed | 24 | 67.9/0.819 | 70.3/0.897 | 61.2/0.680 | 72.3/0.881 |
|
查看原文
王美荣, 徐国明, 袁宏武. 显著性偏振参量深度稀疏特征学习的目标检测方法[J]. 激光与光电子学进展, 2019, 56(19): 191101. Meirong Wang, Guoming Xu, Hongwu Yuan. Object Detection by Deep Sparse Feature Learning of Salient Polarization Parameters[J]. Laser & Optoelectronics Progress, 2019, 56(19): 191101.
PDF全文
