基于邻域最小生成树的半监督极化SAR图像分类方法

滑文强; 王爽; 郭岩河; 谢雯

doi:10.12000/JR18104

基于邻域最小生成树的半监督极化SAR图像分类方法

doi: 10.12000/JR18104

滑文强^1,2, ,,
王爽^3,,
郭岩河^3,,
谢雯^1,

①.
西安邮电大学计算机学院西安 710121
②.
西安邮电大学陕西省网络数据分析与智能处理重点实验室西安 710121
③.
智能感知与图像理解教育部重点实验室国际智能感知与计算联合研究中心西安电子科技大学西安 710071

基金项目: 国家自然科学基金面上项目(61771379)，陕西省普通高等学校重点学科专项

详细信息

作者简介:
滑文强(1987–)，男，陕西西安人，现为西安邮电大学计算机学院讲师，研究方向为极化SAR图像处理。E-mail: huawenqiang2013@163.com

王　爽(1978–)，女，西安电子科技大学教授，博士生导师，智能信息处理研究所副所长，智能感知与图像理解教育部重点实验室成员，国家“111”计划创新引智基地成员，IEEE会员，IET会员，中国电子学会会员，中国计算机学会会员。主要从事SAR/PolSAR处理与分析、稀疏表示、机器学习等方面的研究工作。E-mail: shwang@mail.xidian.edu.cn

郭岩河(1990–)，男，福建泉州人，西安电子科技大学博士研究生，主要研究方向为极化SAR图像处理、深度学习、机器学习等。E-mail: 1153603266@qq.com

谢　雯(1989–)，陕西西安人，现为西安邮电大学通信工程学院讲师，主要研究方向为极化SAR图像处理。E-mail: xiewen236@163.com

通讯作者:
滑文强 huawenqiang2013@163.com

中图分类号: TN958
计量
- 文章访问数: 2925
- HTML全文浏览量: 1048
- PDF下载量: 195
- 被引次数: 0
出版历程
- 收稿日期: 2018-12-03
- 修回日期: 2018-12-28
- 网络出版日期: 2019-08-28

Semi-supervised PolSAR Image Classification Based on the Neighborhood Minimum Spanning Tree

HUA Wenqiang^{1,2
, ,},
WANG Shuang^3
,,
GUO Yanhe^3
,,
XIE Wen^1
,

①.
School of Computer Science and Technology, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
②.
Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing, Xi’an University of Posts and Telecommunications, Xi’an 710121, China
③.
Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education, International Research Center for Intelligent Perception and Computation, Xidian University, Xi’an, 710071, China

Funds: The National Natural Science Foundation of China (61771379), Shaanxi Key Disciplines of Special Funds Projects

More Information

Corresponding author: HUA Wenqiang, huawenqiang2013@163.com

摘要

摘要: 该文针对极化SAR图像分类中只有少量标记样本的问题，提出了一种基于邻域最小生成树的半监督极化SAR图像分类方法。该方法针对极化SAR图像以像素为分类对象的特点，结合自训练方法的思想，利用极化SAR图像像素点的空间信息，提出了基于邻域最小生成树辅助学习的样本选择策略，增加自训练过程中被选择无标记样本的可靠性，扩充标记样本数量，训练更好的分类器。最终用训练好的分类器对极化SAR图像进行测试。对3组真实的极化SAR图像进行测试，实验结果表明，该方法在只有少量标记样本的情况下能获得满意的分类结果，且分类正确率明显优于传统的分类算法。
- 极化SAR图像 /
- 地物分类 /
- 半监督 /
- 最小生成树
Abstract: In this paper, a novel semi-supervised classification method based on the Neighborhood Minimum Spanning Tree (NMST) is proposed to solve the Polarimetric Synthetic Aperture Radar (PolSAR) terrain classification when labeled samples are few. Combining the idea of self-training method and spatial information of the pixels in PolSAR image, a new help-training sample selection strategy based on spatial neighborhood information is proposed, named as NMST, to select the high reliable unlabeled samples to enlarge the training set and improve the base classifier. Finally, the PolSAR image is classified by this improved classifier. The experiments results tested on three PolSAR data sets show that the proposed method achieves a better performance than existing classification methods when the number of labeled samples is few.
- PolSAR /
- Terrain classification /
- Semi-supervised learning /
- Minimum spanning tree

HTML全文

图 1 自训练方法

Figure 1. Self-training method

下载: 全尺寸图片幻灯片

图 2 极化SAR协方差矩阵中9个元素的灰度值

Figure 2. The gray value of 9 elements in PolSAR covariance matrix

下载: 全尺寸图片幻灯片

图 3 带权无向图G及其最小生成树

Figure 3. Weighted undirected graph G and its minimum spanning tree

下载: 全尺寸图片幻灯片

图 4 基于邻域的最小生成树生成过程

Figure 4. The spanning process of neighborhood minimum spanning tree

下载: 全尺寸图片幻灯片

图 5 基于邻域最小生成树的半监督极化SAR分类方法

Figure 5. Semi-supervised PolSAR classification based on the neighborhood minimum spanning tree

下载: 全尺寸图片幻灯片

图 6 Flevoland地区AIRSAR L波段数据不同方法的分类结果

Figure 6. Classification results of the Flevoland data acquired by AIRSAR

下载: 全尺寸图片幻灯片

图 7 Flevoland地区Radarsat-2 C波段数据不同方法的分类结果

Figure 7. Classification result of the Flevoland data acquired by Radarsat-2

下载: 全尺寸图片幻灯片

图 8 旧金山地区Radarsat-2 C波段数据不同方法的分类结果

Figure 8. Classification result of the San Francisco data acquired by Radarsat-2

下载: 全尺寸图片幻灯片

图 9 迭代次数对实验结果的影响

Figure 9. The effects of number of iterations in the proposed method

下载: 全尺寸图片幻灯片

表 1 AIRSAR L波段的Felvoland地区不同分类算法的分类精度(%)

Table 1. Classification accuracy of the Flevoland area acquired by AIRSAR L band (%)

区域	方法
区域	Wishart	SVM	Self-training	本文方法
Stembeans	91.48	70.07	90.82	98.75
Rapeseed	61.83	38.02	67.14	59.58
Bare soil	97.51	86.89	70.97	96.75
Potatoes	79.47	58.38	80.27	81.99
Beet	92.35	85.61	95.05	94.60
Wheat 2	67.43	71.80	67.39	89.86
Peas	93.10	77.70	95.24	97.56
Wheat 3	82.08	82.42	94.33	97.05
Lucerne	84.53	40.77	81.67	95.06
Barley	81.96	98.29	98.62	98.39
Wheat	81.46	68.28	85.34	85.41
Grasses	66.49	65.03	81.75	80.08
Forest	84.21	61.03	77.66	94.77
Water	46.85	65.32	69.39	93.35
Building	81.77	78.91	2.18	85.58
OA	79.40	70.30	77.19	89.92

下载: 导出CSV

表 2 AIRSAR L波段的Felvoland 地区不同训练样本的分类结果

Table 2. Classification results of the Flevoland area acquired by AIRSAR L band with different number of training samples

方法	训练样本数
	4		6		8		10
	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa
Wishart	74.62	0.7215	76.19	0.7459	78.78	0.7656	80.26	0.7831
SVM	56.07	0.5423	58.12	0.5611	64.42	0.6102	70.30	0.6682
Self-training	63.36	0.6025	68.42	0.6569	73.89	0.7146	77.23	0.7489
本文方法	79.33	0.7888	83.06	0.8093	86.90	0.8416	89.92	0.8852

下载: 导出CSV

表 3 Radarsat-2 C波段的Felvoland地区不同分类算法的分类精度(%)

Table 3. Classification accuracy of the Flevoland area acquired by Radarsat-2 C band (%)

区域	方法
区域	Wishart	SVM	Self-training	本文方法
Urban	69.61	54.75	63.93	71.44
Water	98.71	96.83	99.10	98.82
Forest	91.65	65.25	73.83	83.63
Cropland	55.27	78.97	79.23	82.24
OA	78.81	73.95	79.02	84.03

下载: 导出CSV

表 4 Radarsat-2 C波段的Felvoland 地区不同训练样本的分类结果

Table 4. Classification results of the Flevoland area acquired by Radarsat-2 C band with different number of training samples

方法	训练样本数
	4		6		8		10
	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa
Wishart	69.21	0.5803	73.65	0.6239	76.81	0.6854	78.81	0.7026
SVM	50.79	0.4153	64.79	0.5471	70.05	0.5968	73.95	0.6394
Self-training	65.69	0.5233	70.41	0.5911	74.40	0.6605	79.45	0.7144
本文方法	76.71	0.6768	79.29	0.7235	82.02	0.7644	84.03	0.7882

下载: 导出CSV

表 5 Radarsat-2 C波段的旧金山地区不同分类算法的分类结果(%)

Table 5. Classification accuracy of the San Francisco area acquired by radarsat-2 C Band (%)

区域	方法
区域	Wishart	SVM	Self-training	本文方法
Water	98.70	90.04	98.04	99.92
Vegetation	91.03	78.51	84.45	91.50
Low-Density Urban	81.30	42.31	70.18	75.05
High-Density Urban	42.58	77.15	33.01	68.27
Developed	55.26	24.00	56.16	58.81
OA	73.77	62.40	68.37	78.71

下载: 导出CSV

表 6 Radarsat-2 C波段的旧金山地区不同训练样本的分类结果

Table 6. Classification results of the San Francisco area acquired by Radarsat-2 C band with different number of training samples

方法	训练样本数
	4		6		8		10
	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa	OA (%)	Kappa
Wishart	68.09	0.5181	70.44	0.5439	72.49	0.5867	73.77	0.6011
SVM	50.24	0.2817	51.25	0.2905	56.31	0.3628	62.40	0.4342
Self-training	52.34	0.3126	58.62	0.3669	63.27	0.4357	68.42	0.5308
本文方法	70.87	0.5482	73.15	0.5986	75.23	0.6284	78.71	0.6852

下载: 导出CSV

参考文献(27)

[1]	NUNZIATA F, MIGLIACCIO M, LI Xiaofeng, et al. Coastline extraction using dual-Polarimetric COSMO-SkyMed PingPong mode SAR data[J]. IEEE Geoscience and Remote Sensing Letters, 2014, 11(1): 104–108. doi: 10.1109/LGRS.2013.2247561
[2]	HE Jinglu, WANG Yinghua, LIU Hongwei, et al. A novel automatic PolSAR ship detection method based on superpixel-level local information measurement[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(3): 384–388. doi: 10.1109/LGRS.2017.2789204
[3]	BUONO A, NUNZIATA F, MIGLIACCIO M, et al. Classification of the yellow river delta area using fully polarimetric SAR measurements[J]. International Journal of Remote Sensing, 2017, 38(23): 6714–6734. doi: 10.1080/01431161.2017.1363437
[4]	RATHA D, BHATTACHARYA A, and FRERY A C. Unsupervised classification of PolSAR data using a scattering similarity measure derived from a geodesic distance[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(1): 151–155. doi: 10.1109/LGRS.2017.2778749
[5]	LI Dong and ZHANG Yunhua. Adaptive model-based classification of PolSAR data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(12): 6940–6955. doi: 10.1109/TGRS.2018.2845944
[6]	HUANG Xiayuan, ZHANG Bo, QIAO Hong, et al. Local discriminant canonical correlation analysis for supervised PolSAR image classification[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(11): 2102–2106. doi: 10.1109/LGRS.2017.2752800
[7]	REDOLFI J, SÁNCHEZ J, and FLESIA A G. Fisher vectors for PolSAR image classification[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(11): 2057–2061. doi: 10.1109/LGRS.2017.2750800
[8]	LIU Hongying, WANG Yikai, YANG Shuyuan, et al. Large polarimetric SAR data semi-supervised classification with spatial-anchor graph[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2016, 9(4): 1439–1458. doi: 10.1109/JSTARS.2016.2518675
[9]	HUA W Q, WANG S, YANG Zhao et al. Semi-supervised PolSAR image classification based on improved Tri-training[C]. 2017 IEEE International Geoscience and Remote Sensing Symposium, Fort Worth, USA, 2017:3937-3940.
[10]	ROSENBERG C, HEBERT M, and SCHNEIDERMAN H. Semi-supervised self-training of object detection models[C]. Proceedings of the 2005 7th IEEE Workshops on Applications of Computer Vision, Breckenridge, USA, 2005: 29–36.
[11]	BLUM A and MITCHELL T. Combining labeled and unlabeled data with co-training[C]. Proceedings of the 11th Conference on Computational Learning Theory, Madison, USA, 1998: 92–100.
[12]	ZHU Zhihua and LI Ming. Tri-training: Exploiting unlabeled data using three classifiers[J]. IEEE Transactions on Knowledge and Data Engineering, 2005, 17(11): 1529–1541. doi: 10.1109/TKDE.2005.186
[13]	LIU Hongying, WANG Yikai, ZHU Dexiang, et al.. Semi-supervised classification based on anchor-spatial graph for large polarimetric SAR data[C]. Proceedings of 2015 IEEE International Geoscience and Remote Sensing Symposium, Milan, Italy, 2015: 1845–1848.
[14]	LIU Hongying, ZHU Dexiang, YANG Shuyuan, et al. Semisupervised feature extraction with neighborhood constraints for polarimetric SAR classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2016, 9(7): 3001–3015. doi: 10.1109/JSTARS.2016.2532922
[15]	WU Wenjin, LI Hailei, ZHANG Lu, et al. High-resolution PolSAR scene classification with pretrained deep convnets and manifold polarimetric parameters[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(10): 6159–6168. doi: 10.1109/TGRS.2018.2833156
[16]	RASMUS A, VALPOLA H, HONKALA M, et al. Semi-supervised learning with ladder networks[J]. arXiv: 1507.02672, 2015.
[17]	CHENG Yanhua, ZHAO Xin, CAI Rui, et al. Semi-supervised multimodal deep learning for RGB-D object recognition[C]. Proceedings of the 25th International Joint Conference on Artificial Intelligence, New York, USA, 2016: 3345–3351.
[18]	HÄNSCH R and HELLWICH O. Semi-supervised learning for classification of polarimetric SAR-data[C]. Proceedings of 2009 IEEE International Geoscience and Remote Sensing Symposium, Cape Town, South Africa, 2009: 987–990.
[19]	Liu H Y, Wang Y K, Zhua D X et al.. Semi-supervised classification based on anchor-spatial graph for large polarimetric SAR data[C]. 2015 IEEE International Geoscience and Remote Sensing Symposium, Milan, Italy, 2015: 1845-1848.
[20]	HUA Wenqiang, WANG Shuang, LIU Hongying, et al. Semisupervised PolSAR image classification based on improved cotraining[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017, 10(11): 4971–4986. doi: 10.1109/JSTARS.2017.2728067
[21]	GENG Jie, MA Xiaorui, FAN Jianchao, et al. Semisupervised classification of polarimetric SAR image via superpixel restrained deep neural network[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(1): 122–126. doi: 10.1109/LGRS.2017.2777450
[22]	LASZLO M and MUKHERJEE S. Minimum spanning tree partitioning algorithm for microaggregation[J]. IEEE Transactions on Knowledge and Data Engineer, 2005, 17(7): 902–911. doi: 10.1109/TKDE.2005.112
[23]	王晓东. 计算机算法设计与分析[M]. 第4版, 北京: 电子工业出版社, 2012: 103–104. WANG Xiaodong. Design and Analysis of Algorithms[M]. 4th Ed, Beijing: China, Electronic Industry Press, 2002: 103–104.
[24]	LEE J S, GRUNES M R, AINSWORTH T L, et al. Unsupervised classification using polarimetric decomposition and the complex Wishart classifier[J]. IEEE Transactions on Geoscience and Remote Sensing, 1999, 37(5): 2249–2258. doi: 10.1109/36.789621
[25]	LEE J S, GRUNES M R, and DE GRANDI G. Polarimetric SAR speckle filtering and its implication for classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 1999, 37(5): 363–373. doi: 10.1109/36.789635
[26]	LONG Y, and LIU X. SVM lithological classification of PolSAR image in yushigou Area, Qilian Mountain[J]. Scientific Journal of Earth Science, 2013, 3(4): 128–132.
[27]	LEE J S, GRUNES M R, and KWOK R. Classification of multi-look polarimetric SAR imagery based on complex Wishart distribution[J]. International Journal of Remote Sensing, 1994, 15(11): 2299–2311. doi: 10.1080/01431169408954244