Semantic Segmentation of Synthetic Aperture Radar Images Based on U-Net and Capsule Network

Shaodi Jing; Lingjuan Yu; Yuehong Hu; Zezhou Yang; Zhongliang Lu; Xiaochun Xie

doi:10.3788/LOP202158.2010009

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 20 >Page 2010009 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 20, 2010009 (2021)

Semantic Segmentation of Synthetic Aperture Radar Images Based on U-Net and Capsule Network

Shaodi Jing¹, Lingjuan Yu^1、*, Yuehong Hu², Zezhou Yang¹, Zhongliang Lu¹, and Xiaochun Xie³

Author Affiliations

¹School of Information Engineering, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, China

²Guangzhou Wayful Technology Development Co., Ltd., Guangzhou, Guangdong 510200, China

³School of Physics and Electronic Information, Gannan Normal University, Ganzhou, Jiangxi 341000, China

show less

DOI: 10.3788/LOP202158.2010009 Cite this Article Set citation alerts

Shaodi Jing, Lingjuan Yu, Yuehong Hu, Zezhou Yang, Zhongliang Lu, Xiaochun Xie. Semantic Segmentation of Synthetic Aperture Radar Images Based on U-Net and Capsule Network[J]. Laser & Optoelectronics Progress, 2021, 58(20): 2010009 Copy Citation Text

show less

Fig. 1. Structure and parameters of the U-Net

Download full size

Fig. 2. Structure and parameters of the capsule network

Download full size

Fig. 3. Network based on U-Net and capsule network for semantic segmentation of SAR image

Download full size

Fig. 4. San Francisco Bay data. (a) RGB image; (b) label

Download full size

Fig. 5. Part of training samples in the San Francisco Bay data set

Download full size

Fig. 6. Oberpfaffenhofen data. (a) RGB image; (b) label

Download full size

Fig. 7. Part of training samples in the Oberpfaffenhofen data set

Download full size

Fig. 8. Segmentation results of different methods on the San Francisco Bay data set. (a) Method 1; (b) method 2; (c) method 3; (d) method 4

Download full size

Fig. 9. Segmentation results of different methods on the Oberpfaffenhofen data set. (a) Method 1; (b) method 2; (c) method 3; (d) method 4

Download full size

Method	X_Precision	X_Recall	X_F1-score	X_IOU
Method 1	87.46	95.72	91.40	84.17
Method 2	92.87	95.11	93.98	88.64
Method 3	89.31	96.30	92.67	86.35
Method 4	93.86	96.31	95.07	90.60

Table 1. Segmentation performance of different methods on the San Francisco Bay data set unit: %

Method		X_Precision	X_Recall	X_F1-score	X_IOU
Use transfer learning	Method 2 vs. method 1	6.19	-0.64	2.82	5.31
Use transfer learning	Method 4 vs. method 3	5.10	0.01	2.58	4.92
Use capsule network	Method 3 vs. method 1	2.12	0.61	1.39	2.59
Use capsule network	Method 4 vs. method 2	1.07	1.26	1.16	2.21

Table 2. Percentage improvement of the segmentation performance (San Francisco Bay data set) unit: %

Method	Method 1	Method 2	Method 3	Method 4
Training time /s	16	13	17	14

Table 3. Training time of different methods on the San Francisco Bay data set

Method	X_Precision	X_Recall	X_F1-score	X_IOU
Method 1	87.60	91.50	89.51	81.01
Method 2	90.89	91.53	91.21	83.84
Method 3	89.16	93.07	91.07	83.61
Method 4	92.06	93.18	92.62	86.24

Table 4. Segmentation performance of different methods on the Oberpfaffenhofen data set unit: %

Method		X_Precision	X_Recall	X_F1-score	X_IOU
Use transfer learning	Method 2 vs. method 1	3.76	0.03	1.90	3.49
Use transfer learning	Method 4 vs. method 3	3.25	0.12	1.70	3.15
Use capsule network	Method 3 vs. method 1	1.78	1.72	1.74	3.21
Use capsule network	Method 4 vs. method 2	1.29	1.80	1.55	2.86

Table 5. Percentage improvement of the segmentation performance (Oberpfaffenhofen data set) unit: %

Method	Method 1	Method 2	Method 3	Method 4
Training time /s	37	29	40	32

Table 6. Training time of different methods on the Oberpfaffenhofen data set

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information