基于Pinball损失函数支持向量机的极化SAR图像鲁棒分类

张腊梅 张思雨 董洪伟 朱厦

张腊梅, 张思雨, 董洪伟, 等. 基于Pinball损失函数支持向量机的极化SAR图像鲁棒分类[J]. 雷达学报, 2019, 8(4): 448–457. doi: 10.12000/JR19055
引用本文: 张腊梅, 张思雨, 董洪伟, 等. 基于Pinball损失函数支持向量机的极化SAR图像鲁棒分类[J]. 雷达学报, 2019, 8(4): 448–457. doi: 10.12000/JR19055
ZHANG Lamei, ZHANG Siyu, DONG Hongwei, et al. Robust classification of PolSAR images based on Pinball loss support vector machine[J]. Journal of Radars, 2019, 8(4): 448–457. doi: 10.12000/JR19055
Citation: ZHANG Lamei, ZHANG Siyu, DONG Hongwei, et al. Robust classification of PolSAR images based on Pinball loss support vector machine[J]. Journal of Radars, 2019, 8(4): 448–457. doi: 10.12000/JR19055

基于Pinball损失函数支持向量机的极化SAR图像鲁棒分类

doi: 10.12000/JR19055
基金项目: 国家自然科学基金(61401124, 61871158),航空科学基金(20182077008),黑龙江省留学归国人员科学基金(LC2018029)
详细信息
    作者简介:

    张腊梅(1980–),女,博士,副教授,博士生导师。2010年在哈尔滨工业大学获得博士学位,2014至2015年在加拿大曼尼托巴大学作访问学者,现为哈尔滨工业大学副教授。主要研究方向包括SAR/极化SAR图像智能处理和应用,极化干涉SAR信息提取,目标电磁散射特性分析和仿真。在国内外著名学术刊物和国际会议上发表学术论文80多篇,授权/受理发明专利10多项,担任哈尔滨IEEE GRSS秘书。E-mail: lmzhang@hit.edu.cn

    张思雨(1997–),男,黑龙江哈尔滨人,学士。2019年在大连海事大学电子信息工程专业获得学士学位,现为哈尔滨工业大学信息与通信工程学院硕士研究生。研究方向为SAR/PolSAR图像处理与智能解译、模式识别与机器学习。E-mail: missrain2831@163.com

    董洪伟(1994–),男,山东济南人,硕士。2018年在中国农业大学理学院数学系获得硕士学位,现为哈尔滨工业大学信息与通信工程专业博士研究生。研究方向为计算机视觉、模式识别与机器学习及SAR图像图像解译。E-mail: donghongwei1994@163.com

    朱 厦(1984–),女,博士,助理研究员。2012年在法国巴黎十一大学获得国家物理学博士学位,2018年在法国国家科学院信号与系统LSS实验室作访问学者,现为北京市遥感信息研究所助理研究员,主要从事微波卫星应用总体、极化SAR信息处理和海洋遥感方面的研究工作。发表学术论文30余篇,主持国家863计划等多项科研课题,先后受国家“建设高水平大学”公派研究生项目、国家外专局“智力引进计划”资助,获省部级科技进步二等奖1项。E-mail: nudt_zs@163.com

    通讯作者:

    朱厦 nudt_zs@163.com

  • 中图分类号: TP391

Robust Classification of PolSAR Images Based on Pinball loss Support Vector Machine

Funds: The National Natural Science Foundation of China (61401124, 61871158), The Aeronautical Science Foundation of China (20182077008), The Scientific Research Foundation for the Returned Overseas Scholars of Heilongjiang Province (LC2018029)
More Information
  • 摘要: 考虑到极化合成孔径雷达(PolSAR)图像标注信息量低以及相干斑噪声难以消除的问题,该文从鲁棒统计学习的角度提出了一种基于Pin-SVM的极化SAR图像鲁棒分类方法,根据极化SAR图像的散射特性和地物的纹理特性,通过求解两类样本之间的最大分位数距离来确定分类超平面,在无需迭代的前提下得到更加鲁棒的分类结果。相比传统的基于最大间隔的极化SAR图像分类算法,该文所提算法一方面在对极化SAR图像提取到的特征中包含的噪声具有更好的鲁棒性,另一方面对于训练样本的抽样范围不敏感,即重采样具有更好的鲁棒性。利用EMISAR的Foulum地区极化SAR数据进行了算法验证,多种情况的对比实验的结果验证了该算法的有效性。

     

  • 图  1  0-1损失、Hinge损失与Pinball损失的对比

    Figure  1.  The form of 0-1 loss, Hinge loss and Pinball loss

    图  2  不同超参数$\tau $取值下Pin-SVM分类示意图

    Figure  2.  Schematic diagram of Pin-SVM classification with different hyperparameter $\tau $

    图  3  基于Pin-SVM的极化SAR图像分类流程

    Figure  3.  Flowchart of PolSAR classification based on Pin-SVM

    图  4  Foulum地区EMISAR实验数据

    Figure  4.  EMISAR experimental datas of Foulum test site

    图  5  EMISAR图像分类结果对比

    Figure  5.  Classification results comparison of EMISAR image

    图  6  不同分类器对各地物分类精度的对比图

    Figure  6.  The comparison of the classification accuracy of different terrains

    图  7  分类器预测的概率密度函数对比

    Figure  7.  Comparison of probability density functions for the prediction of classifiers

    图  8  重采样的鲁棒性评估

    Figure  8.  Robustness evaluation of resampling

    图  9  Pin-SVM的参数对分类精度的影响

    Figure  9.  The effect hyperparameters of Pin-SVM on classification accuracy

    表  1  不同分类器对测试样本的分类精度(%)

    Table  1.   Classification accuracy comparison of different classifiers

    分类器地物类型整体精度
    建筑物森林裸地细径作物阔叶作物
    Pin-SVM94.390.795.182.394.191.3
    C-SVM91.080.692.380.189.786.7
    Wishart85.471.293.784.285.684.0
    LSSVM86.280.289.378.888.284.5
    OPTELM86.582.392.175.183.884.0
    下载: 导出CSV

    表  2  重采样实验结果

    Table  2.   Experimental results of the resampling

    分类器重采样
    30次50次100次150次
    Pin-SVMw$8.96 \pm 1.11$$8.69 \pm 0.99$$8.57 \pm 0.96$$8.63 \pm 0.95$
    $\tau {\rm{ = }}1.0$b$ - 5.91 \pm 0.62$$ - 5.95 \pm 0.61$$ - 5.93 \pm 0.59$$ - 5.93 \pm 0.44$
    Pin-SVMw$9.43 \pm 1.23$$9.54 \pm 1.20$$9.42 \pm 1.15$$9.68 \pm 1.13$
    $\tau {\rm{ = 0}}{\rm{.5}}$b$ - 6.93 \pm 0.75$$ - 7.02 \pm 0.71$$ - 6.91 \pm 0.67$$ - 6.85 \pm 0.59$
    Pin-SVMw$11.43 \pm 1.63$$11.22 \pm 1.47$$11.09 \pm 1.46$$11.31 \pm 1.44$
    $\tau {\rm{ = 0}}{\rm{.2}}$b$ - 8.53 \pm 0.84$$ - 8.53 \pm 0.78$$ - 8.41 \pm 0.75$$ - 8.39 \pm 0.73$
    Pin-SVMw$12.30 \pm 1.84$$12.12 \pm 1.74$$12.23 \pm 1.61$$12.09 \pm 1.59$
    $\tau {\rm{ = 0}}{\rm{.1}}$b$ - 9.57 \pm 1.02$$ - 9.88 \pm 0.99$$ - 9.50 \pm 0.97$$ - 9.62 \pm 0.92$
    C-SVMw$13.29 \pm 1.96$$12.98 \pm 1.87$$13.42 \pm 1.73$$13.12 \pm 1.69$
    b$ - 16.15 \pm 3.04$$ - 15.81 \pm 2.74$$ - 15.92 \pm 2.45$$ - 15.48 \pm 2.28$
    下载: 导出CSV
  • [1] CHEN Qiang, KUANG Gangyao, LI J, et al. Unsupervised land cover/land use classification using PolSAR imagery based on scattering similarity[J]. IEEE Transactions on Geoscience and Remote Sensing, 2013, 51(3): 1817–1825. doi: 10.1109/tgrs.2012.2205389
    [2] LI Wenmei, CHEN Erxue, LI Zengyuan, et al. Forest aboveground biomass estimation using polarization coherence tomography and PolSAR segmentation[J]. International Journal of Remote Sensing, 2015, 36(2): 530–550. doi: 10.1080/01431161.2014.999383
    [3] FOUCHER S, LANDRY T, LÓPEZ-MARTÍNEZ C, et al. An evaluation of PolSAR speckle filters on compact-pol images[C]. Proceedings of 2012 IEEE International Geoscience and Remote Sensing Symposium, Munich, Germany, 2012: 5089–5092. doi: 10.1109/IGARSS.2012.6352466.
    [4] LIU Xunan and CHENG Bo. Polarimetric SAR speckle filtering for high-resolution SAR images using RADARSAT-2 POLSAR SLC data[C]. Proceedings of 2012 International Conference on Computer Vision in Remote Sensing, Xiamen, China, 2012: 329–334. doi: 10.1109/CVRS.2012.6421284.
    [5] 滑文强, 王爽, 侯彪. 基于半监督学习的SVM-Wishart极化SAR图像分类方法[J]. 雷达学报, 2015, 4(1): 93–98. doi: 10.12000/JR14138

    HUA Wenqiang, WANG Shuang, and HOU Biao. Semi-supervised learning for classification of polarimetric SAR images based on SVM-Wishart[J]. Journal of Radars, 2015, 4(1): 93–98. doi: 10.12000/JR14138
    [6] ZHANG Yue, ZOU Huanxin, SHAO Ningyuan, et al. Unsupervised classification of polsar imagery based on consensus similarity network fusion[C]. Proceedings of 2017 IEEE International Geoscience and Remote Sensing Symposium, Fort Worth, USA, 2017: 3266–3269. doi: 10.1109/IGARSS.2017.8127694.
    [7] BI Haixia, SUN Jian, and XU Zongben. Unsupervised PolSAR image classification using discriminative clustering[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(6): 3531–3544. doi: 10.1109/TGRS.2017.2675906
    [8] FERRO-FAMIL L, POTTIER E, and LEE J S. Unsupervised classification of multifrequency and fully polarimetric SAR images based on the H/A/Alpha-Wishart classifier[J]. IEEE Transactions on Geoscience and Remote Sensing, 2001, 39(11): 2332–2342. doi: 10.1109/36.964969
    [9] SHANG Fang and HIROSE A. Use of Poincare sphere parameters for fast supervised PolSAR land classification[C]. Proceedings of 2013 IEEE International Geoscience and Remote Sensing Symposium, Melbourne, Australia, 2013: 3175–3178. doi: 10.1109/IGARSS.2013.6723501.
    [10] DABBOOR M and SHOKR M. A new likelihood ratio for supervised classification of fully polarimetric SAR data: An application for sea ice type mapping[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2013, 84: 1–11. doi: 10.1016/j.isprsjprs.2013.06.010
    [11] ZHANG Lamei, SUN Liangjie, ZOU Bin, et al. Fully polarimetric SAR image classification via sparse representation and polarimetric features[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015, 8(8): 3923–3932. doi: 10.1109/JSTARS.2014.2359459
    [12] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]. Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 770–778. doi: 10.1109/CVPR.2016.90.
    [13] VAPNIK V N. The Nature of Statistical Learning Theory[M]. New York, US: Springer-Verlag, 2000: 267–290. doi: 10.1007/978-1-4757-3264-1.
    [14] MAGHSOUDI Y, COLLINS M J, and LECKIE D G. Radarsat-2 polarimetric SAR data for boreal forest classification using SVM and a wrapper feature selector[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2013, 6(3): 1531–1538. doi: 10.1109/JSTARS.2013.2259219
    [15] CHEN Wenshuai, HAI Dong, GOU Shuiping, et al. Classification of PolSAR images based on SVM with Self-Paced learning optimization[C]. Proceedings of 2018 IEEE International Geoscience and Remote Sensing Symposium, Valencia, Spain, 2018: 4491–4494. doi: 10.1109/IGARSS.2018.8517452.
    [16] SUYKENS J A K and VANDEWALLE J. Least squares support vector machine classifiers[J]. Neural Processing Letters, 1999, 9(3): 293–300. doi: 10.1023/a:1018628609742
    [17] WANG Yidan, YANG Liming, and YUAN Chao. A robust outlier control framework for classification designed with family of homotopy loss function[J]. Neural Networks, 2019, 112: 41–53. doi: 10.1016/j.neunet.2019.01.013
    [18] BO Liefeng, WANG Ling, and JIAO Licheng. Recursive finite newton algorithm for support vector regression in the primal[J]. Neural Computation, 2007, 19(4): 1082–1096. doi: 10.1162/neco.2007.19.4.1082
    [19] WU Yichao and LIU Yufeng. Robust truncated hinge loss support vector machines[J]. Journal of the American Statistical Association, 2007, 102(479): 974–983. doi: 10.1198/016214507000000617
    [20] LIU Dalian, SHI Yong, and TIAN Yingjie. Ramp loss nonparallel support vector machine for pattern classification[J]. Knowledge-Based Systems, 2015, 85: 224–233. doi: 10.1016/j.knosys.2015.05.008
    [21] HUANG Xiaolin, SHI Lei, and SUYKENS J A K. Support vector machine classifier with pinball loss[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(5): 984–997. doi: 10.1109/tpami.2013.178
    [22] BO Liefeng, WANG Ling, and JIAO Licheng. Feature scaling for kernel fisher discriminant analysis using leave-one-out cross validation[J]. Neural Computation, 2006, 18(4): 961–978. doi: 10.1162/neco.2006.18.4.961
    [23] SCHOLKOPF B, MIKA S, BURGES C J C, et al. Input space versus feature space in kernel-based methods[J]. IEEE Transactions on Neural Networks, 1999, 10(5): 1000–1017. doi: 10.1109/72.788641
    [24] ZHANG Lamei, ZOU Bin, CAI Hongjun, et al. Multiple-component scattering model for polarimetric SAR image decomposition[J]. IEEE Geoscience and Remote Sensing Letters, 2008, 5(4): 603–607. doi: 10.1109/lgrs.2008.2000795
    [25] CLOUDE S R and POTTIER E. An entropy based classification scheme for land applications of polarimetric SAR[J]. IEEE Transactions on Geoscience and Remote Sensing, 1997, 35(1): 68–78. doi: 10.1109/36.551935
    [26] HARALICK R M, SHANMUGAM K, and DINSTEIN I. Textural features for image classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1973, SMC-3(6): 610–621. doi: 10.1109/TSMC.1973.4309314
    [27] DEKKER R J. Texture analysis of urban areas in ERS SAR imagery for map updating[C]. Proceedings of IEEE/ISPRS Joint Workshop on Remote Sensing and Data Fusion over Urban Areas, Rome, Italy, 2001: 226–230. doi: 10.1109/DFUA.2001.985885.
    [28] 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): 2363–2373. doi: 10.1109/36.789635
  • 加载中
图(9) / 表(2)
计量
  • 文章访问数:  3854
  • HTML全文浏览量:  1255
  • PDF下载量:  150
  • 被引次数: 0
出版历程
  • 收稿日期:  2019-05-01
  • 修回日期:  2019-07-12
  • 网络出版日期:  2019-08-28

目录

    /

    返回文章
    返回