Author Affiliations
1 Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun, Jilin 130033, China1 Key Laboratory of Satellite Remote Sensing Application Technology of Jilin Province, Chang Guang Satellite Technology Co., Ltd., Changchun, Jilin 130102, China1 Shandong Institute of Space Electronic Technology, China Academy of Space Technology, Yantai, Shandong 264670, China2 Key Laboratory of Satellite Remote Sensing Application Technology of Jilin Province, Chang Guang Satellite Technology Co., Ltd., Changchun, Jilin 130102, China2 University of Chinese Academy of Sciences, Beijing 100049, Chinashow less
Fig. 1. Flow chart of automatic detection method of ships based on SWIR images
Fig. 2. (a) Scene image; (b) reflectance curves of typical features
Fig. 3. (a) Intensity image I; (b) contrast stretched image I'; (c) land mask; (d) land mask filled in holes; (e) water mask
Fig. 4. Saliency detection and target chip extraction. (a)(f) Input SWIR image; (b)(g) visible image of same scene (targets marked manually); (c)(h) saliency image; (d)(i) mask image of candidate targets; (e)(j) SWIR image chip of candidate targets
Fig. 5. Rotation and translation of main axis of ships by Radon transform
Fig. 6. Gray-scale distribution characteristics of target chips. (a) Schematic of direction and regional segmentation; (b) image in polar coordinate; (c) gray-scale distribution curve; (d) gray-scale distribution histogram
Fig. 7. S-HOG descriptor. (a) Regional segmentation; (b) histogram of oriented gradient of region B1 in Fig. 7(a)
Fig. 8. Geometric characterization of target chips. (a) Prescreened target chips; (b) grayscale distribution curve; (c) grayscale distribution histogram; (d) gradient oriented histogram
Fig. 9. Detection results comparison of visible image and SWIR image. (a)(e) Input images; (b)(f) saliency map under CSM model; (c)(g) results of saliency segmentation; (d)(h) results of target discrimination
Fig. 10. Ship detection results in various scenes by proposed method
Fig. 11. Effect of parameter value on detection performance. (a) Accuracy index versus δ; (b) Re for each constraint factor; (c) fDR for each constraint factor
Input | Nt | Ntt | Nf | P /% | /% | /% | /% |
---|
Visible | 142 | 131 | 7 | 94.93 | 5.07 | 92.25 | 93.57 | SWIR | 142 | 138 | 8 | 94.52 | 5.48 | 97.18 | 95.83 |
|
Table 1. Detection results for different input images
Method | Nt | Ntt | Nf | P /% | /% | /% | /% |
---|
Proposed method | 142 | 131 | 7 | 94.93 | 5.07 | 92.25 | 93.57 | Method in [5] | 142 | 116 | 35 | 76.82 | 23.18 | 81.69 | 79.18 | Method in [12] | 142 | 121 | 4 | 96.80 | 3.20 | 85.21 | 90.64 |
|
Table 2. Detection results for different methods