基于主动深度学习的极化SAR图像分类
|
摘要
针对极化SAR图像在监督分类时存在人工标注样本费时费力以及浅层结构学习算法的表达能力有限等问题,提出一种基于主动深度学习的极化SAR图像分类方法。首先,对测量数据进行多种极化特征提取,以便完整地描述图像信息;在此基础上,通过自动编码器对大量无标记样本进行非监督学习,提取更具可分性和不变性的深层特征;然后,利用少量标记样本训练分类器,并与自动编码器连接,以监督学习的方式微调整个网络;最后,通过主动学习,选择对当前分类器最有价值的样本(分类模糊度最大的样本)进行人工标记,并加入到训练样本中,重新训练分类器和微调网络。对RADARSAT-2和EMISAR极化SAR影像进行不同分类的实验结果表明,该方法能在更少人工标记的样本下获得较高的分类精度。
Supervised classification methods usually require adequate labeled samples which are difficult and time-consuming to obtain for polarimetric SAR images,while the expression capability of the shallow structure learning algorithm is limited. A novel supervised classification method for polarimetric SAR imagery based on active deep learning is proposed in this paper. Firstly,the features are extracted from an original image by multiple polarization target decomposition methods for fully describing the data,and the features which are separable and invariable can be extracted with unsupervised learning by auto-encoder. Then,the initial classifier is trained and fine-tune the whole model with a small number of labeled samples. Finally,the most valuable samples( the largest ambiguity samples for classifier) are selected to label by active learning. Experimental results in comparison with conventional methods for polarimetric SAR data sets of RADARSAT-2 and EMISAR show that the proposed method can achieve higher classification accuracy with a small number of labeled samples.
Supervised classification methods usually require adequate labeled samples which are difficult and time-consuming to obtain for polarimetric SAR images,while the expression capability of the shallow structure learning algorithm is limited. A novel supervised classification method for polarimetric SAR imagery based on active deep learning is proposed in this paper. Firstly,the features are extracted from an original image by multiple polarization target decomposition methods for fully describing the data,and the features which are separable and invariable can be extracted with unsupervised learning by auto-encoder. Then,the initial classifier is trained and fine-tune the whole model with a small number of labeled samples. Finally,the most valuable samples( the largest ambiguity samples for classifier) are selected to label by active learning. Experimental results in comparison with conventional methods for polarimetric SAR data sets of RADARSAT-2 and EMISAR show that the proposed method can achieve higher classification accuracy with a small number of labeled samples.
引文
[1]Qi Z X,Yeh A G O,Li X,et al.Monthly short-term detection of land development using RADARSAT-2 polarimetric SAR imagery[J].Remote Sensing of Environment,2015,164:179-196.
[2]Yu P,Qin A K,Clausi D A.Unsupervised polarimetric SAR image segmentation and classification using region growing with edge penalty[J].IEEE Transactions on Geoscience and Remote Sensing,2012,50(4):1302-1317.
[3]陈军,杜培军,谭琨.一种改进的全极化SAR图像MCSMWishart非监督分类方法[J].国土资源遥感,2015,27(2):15-21.doi:10.6046/gtzyyg.2015.02.03.Chen J,Du P J,Tan K.An improved unsupervised classification scheme for polarimetric SAR image with MCSM-Wishart[J].Remote Sensing for Land and Resources,2015,27(2):15-21.doi:10.6046/gtzyyg.2015.02.03.
[4]Samat A,Du P J,Baig M H A,et al.Ensemble learning with multiple classifiers and polarimetric features for polarized SAR image classification[J].Photogrammetric Engineering&Remote Sensing,2014,80(3):239-251,doi:10.14358/PERS.80.3.239.
[5]Zhang Y H,Zhang J X,Zhang X F,et al.Land cover classification from polarimetric SAR data based on image segmentation and decision trees[J].Canadian Journal of Remote Sensing,2015,41(1):40-50.
[6]Uhlmann S,Kiranyaz S,Gabbouj M.Semi-supervised learning for ill-posed polarimetric SAR classification[J].Remote Sensing,2014,6(6):4801-4830.
[7]Liu H Y,Wang Y K,Yang S Y,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.
[8]陈荣,曹永锋,孙洪.基于主动学习和半监督学习的多类图像分类[J].自动化学报,2011,37(8):954-962.Chen R,Cao Y F,Sun H.Multi-class image classification with active learning and semi-supervised learning[J].Acta Automatica Sinica,2011,37(8):954-962.
[9]Samat A,Gamba P,Du P J,et al.Active extreme learning machines for quad-polarimetric SAR imagery classification[J].International Journal of Applied Earth Observation and Geoinformation,2015,35:305-319.
[10]Joshi A J,Porikli F,Papanikolopoulos N.Multi-class active learning for image classification[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition.Miami,FL:IEEE,2009:2372-2379.
[11]Bengio Y.Learning deep architectures for AI[J].Foundations and Trendsin Machine Learning,2009,2(1):1-127.
[12]Cloude S R,Pottier E.A review of target decomposition theorems in radar polarimetry[J].IEEE Transactions on Geoscience and Remote Sensing,1996,34(2):498-518.
[13]Rumelhart D E,Hinton G E,Williams R J.Learning representations by back-propagating errors[J].Nature,1986,323(6088):533-536.
[14]Chen Y S,Lin Z H,Zhao X,et al.Deep learning-based classification of hyperspectral data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2014,7(6):2094-2107.
[2]Yu P,Qin A K,Clausi D A.Unsupervised polarimetric SAR image segmentation and classification using region growing with edge penalty[J].IEEE Transactions on Geoscience and Remote Sensing,2012,50(4):1302-1317.
[3]陈军,杜培军,谭琨.一种改进的全极化SAR图像MCSMWishart非监督分类方法[J].国土资源遥感,2015,27(2):15-21.doi:10.6046/gtzyyg.2015.02.03.Chen J,Du P J,Tan K.An improved unsupervised classification scheme for polarimetric SAR image with MCSM-Wishart[J].Remote Sensing for Land and Resources,2015,27(2):15-21.doi:10.6046/gtzyyg.2015.02.03.
[4]Samat A,Du P J,Baig M H A,et al.Ensemble learning with multiple classifiers and polarimetric features for polarized SAR image classification[J].Photogrammetric Engineering&Remote Sensing,2014,80(3):239-251,doi:10.14358/PERS.80.3.239.
[5]Zhang Y H,Zhang J X,Zhang X F,et al.Land cover classification from polarimetric SAR data based on image segmentation and decision trees[J].Canadian Journal of Remote Sensing,2015,41(1):40-50.
[6]Uhlmann S,Kiranyaz S,Gabbouj M.Semi-supervised learning for ill-posed polarimetric SAR classification[J].Remote Sensing,2014,6(6):4801-4830.
[7]Liu H Y,Wang Y K,Yang S Y,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.
[8]陈荣,曹永锋,孙洪.基于主动学习和半监督学习的多类图像分类[J].自动化学报,2011,37(8):954-962.Chen R,Cao Y F,Sun H.Multi-class image classification with active learning and semi-supervised learning[J].Acta Automatica Sinica,2011,37(8):954-962.
[9]Samat A,Gamba P,Du P J,et al.Active extreme learning machines for quad-polarimetric SAR imagery classification[J].International Journal of Applied Earth Observation and Geoinformation,2015,35:305-319.
[10]Joshi A J,Porikli F,Papanikolopoulos N.Multi-class active learning for image classification[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition.Miami,FL:IEEE,2009:2372-2379.
[11]Bengio Y.Learning deep architectures for AI[J].Foundations and Trendsin Machine Learning,2009,2(1):1-127.
[12]Cloude S R,Pottier E.A review of target decomposition theorems in radar polarimetry[J].IEEE Transactions on Geoscience and Remote Sensing,1996,34(2):498-518.
[13]Rumelhart D E,Hinton G E,Williams R J.Learning representations by back-propagating errors[J].Nature,1986,323(6088):533-536.
[14]Chen Y S,Lin Z H,Zhao X,et al.Deep learning-based classification of hyperspectral data[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2014,7(6):2094-2107.