Author Affiliations
College of Engineering Science and Technology, Shanghai Ocean University, Shanghai, 201306, Chinashow less
Fig. 1. schematic design of system
Fig. 2. Convolution diagram
Fig. 3. Algorithm hierarchy
Fig. 4. TSE and SSE derivative images
Fig. 5. Images with different rotation angles. (a) 0°; (b) 135°; (c) 270°; (d) 225°
Fig. 6. Images with different noises. (a) 1.0×105 pixel; (b) 1.5×105 pixel; (c) 2.0×105 pixel; (d) 5.0×104 pixel
Fig. 7. Images with different random contrast and brightness. (a) (75:100,125); (b) (150:100,125); (c) (60:100,105); (d) (60:100,150); (e) (78:100,100); (f) (78:100,150); (g) (67:100,100); (h) (77:100,123); (i) (77:100,197)
Fig. 8. Analysis diagram of relationship between deviation and river bank line
Fig. 9. Identification result of waterfront
Fig. 10. Comparison of mAP values change trends in 4 cases
Fig. 11. Change curve of average L values
Fig. 12. Change curve of average IOU
Fig. 13. Schematic of target recognition test. (a) Boat; (b) bottle; (c) buoy; (d) foam
Fig. 14. Change curve of recognition speed
Algorithm | mAP |
---|
YOLOv3(SSE) | 0.743 | YOLOv3-TSE | 0.772 | YOLOv3-DenseNet | 0.759 | YOLOv3-TSE-DenseNet | 0.783 |
|
Table 1. Comparison of mAP of different algorithms in YOLOv3 network
Algorithm | Network | mAP |
---|
Fast R-CNN | VGG | 0.673 | Faster R-CNN | Residual-101 | 0.697 | SSD | Residual-101 | 0.711 | YOLOv3 | DarkNet53 | 0.732 | new_YOLOv3 | Darknet53+DenseNet | 0.766 |
|
Table 2. Comparison of results of target test set