Author Affiliations
1Department of Information Engineering, PLA Army Artillery Air Defense Force College, Hefei 230031, Anhui, China2Key Laboratory of Polarized Light Imaging Detection Technology of Anhui Province, Hefei 230031, Anhui, Chinashow less
Fig. 1. Schematic of light propagation
Fig. 2. Layout diagram of color focal plane polarized pixel array
Fig. 3. Structure diagram of TSF-Net
Fig. 4. Structure diagram of ANN
Fig. 5. Structure diagram of APP-Net
Fig. 6. Structure diagram of APP-Net
Fig. 7. Structure diagram of RGB-Net
Fig. 8. Process of feature extraction and feature fusion
Fig. 9. Schematic of training and test process
Fig. 10. Physical drawing of portable acquisition equipment
Fig. 11. Schematic of classification of individual camouflage polarization image dataset
Fig. 12. Two types of camouflage target test diagram. (a) Multicam type camouflage; (b) Woodland type camouflage
Fig. 13. Detection effects of different models in Multicam dataset. (a) SSD model; (b) YOLOv4 model; (c) YOLOv5 model; (d) RetinaNet model; (e) Faster R-CNN model; (f) TSF-Net model
Fig. 14. Detection effects of different models in Woodland dataset. (a) SSD model; (b) YOLOv4 model; (c) YOLOv5 model; (d) RetinaNet model; (e) Faster R-CNN model; (f) TSF-Net model
Fig. 15. Parameter verification result
Fig. 16. Verification results for different branches. (a) Detection accuracy of different v values; (b) IOU-mAP curves
Structure | GPU memoryusage /MB | Time/min | Loss |
---|
(8,16,8,3) | 1453 | 225 | 1.23×10-2 | (16,8,8,3) | 1453 | 216 | 9.51×10-3 | (96,48,32,3) | 3817 | 207 | 7.48×10-3 | (48,96,32,3) | 3817 | 229 | 5.79×10-3 | (128,96,64,32,3) | 6109 | 255 | 5.55×10-3 | (96,128,64,32,3) | 6109 | 282 | 4.27×10-3 |
|
Table 1. Training results with different structures
Category | Parameter |
---|
Camera model | FLIR BFS-U3-51S5PC-C | Resolution /(pixel×pixel) | 2448×2048 | Frame rate /(frame·s-1) | 75 | Chip model | Sony IMX250MYR,Polar-RGB | Data interface | USB3.1 Gen1 | Size and weight /(mm×mm×mm) | 29×29×30 | Mass /g | 36 | Lens interface | C-Mount |
|
Table 2. Parameters of color focal plane camera
Case | Prediction (positive) | Prediction (negative) |
---|
Ture(true) | TP | TN | Ture(false) | FP | FN |
|
Table 3. Positive and negative cases
Model | mAP /% |
---|
Multicam dateset | Woodland dataset |
---|
SSD | 70.9 | 73.5 | YOLOv4 | 71.5 | 73.4 | YOLOv5 | 73.1 | 74.6 | RetinaNet | 75.2 | 77.5 | Faster R-CNN | 77.1 | 78.9 | TSF-Net | 85.9 | 87.1 |
|
Table 4. Comparison of detection accuracy of different models
Model | mAP /% |
---|
SF | SS | SB |
---|
TSF-Net | 84.9 | 84.3 | 84.6 |
|
Table 5. Comparison of posture test results of different camouflage personnel
Model | mAP /% |
---|
M/W | W/M |
---|
Faster R-CNN | 23.5 | 15.7 | TSF-Net | 48.8 | 35.5 |
|
Table 6. Cross-validation comparison