特征优选下的遥感影像面向对象分类规则构建
|
摘要
针对分类规则及其阈值的确定主要依赖人工经验、通用性差问题,该文提出了特征优选支持的面向对象分类规则构建方法。该方法利用面向对象技术,首先使用Boruta算法对先验样本数据集进行特征选择,然后根据隶属度函数构建分类规则集,最后引入置信区间概念,确定分类规则的阈值。以德国波兹坦地区的航空影像、数字表面模型(DSM)以及地面真实参考影像为实验数据,构建城市建筑、城市绿地(包括草地和树木)这两种地类的分类规则,利用不同数量的训练样本,开展面向对象分类实验,与支持向量机(SVM)监督分类方法进行对比分析。实验结果表明,在相同的优选特征下,利用置信区间确定阈值得到的分类规则,提取效果及分类精度更好,尤其在训练样本量少的情况下,该方法得到的分类精度比SVM高30%～40%。
Aiming at the problem that the classification rules and its threshold values which are mainly based on an iterative trial-and-error approach has a poor versatility,this paper proposed a method to construct classification rules automatically supported by feature optimization.A three-step workflow had been introduced,firstly,the features of prior sample data set were selected by using the Boruta algorithm.Secondly,the classification rule set was built using the membership function.Thirdly,the threshold values of the classification rules were determined using the statistic confidence interval.The test site is located in Potsdam city,Germany.The test data included aerial imagery,digital surface model(DSM),and the real label image.Urban building and the green space were taken for example.Urban building and the green space were extracted using their classification rules with different sample sizes.The results demonstrated a better effect and accuracy compared with support vector machine(SVM)using the same optimal feature,and the classification accuracy of this method was 30% ～40% higher than that of SVM in the case of a few training samples.
Aiming at the problem that the classification rules and its threshold values which are mainly based on an iterative trial-and-error approach has a poor versatility,this paper proposed a method to construct classification rules automatically supported by feature optimization.A three-step workflow had been introduced,firstly,the features of prior sample data set were selected by using the Boruta algorithm.Secondly,the classification rule set was built using the membership function.Thirdly,the threshold values of the classification rules were determined using the statistic confidence interval.The test site is located in Potsdam city,Germany.The test data included aerial imagery,digital surface model(DSM),and the real label image.Urban building and the green space were taken for example.Urban building and the green space were extracted using their classification rules with different sample sizes.The results demonstrated a better effect and accuracy compared with support vector machine(SVM)using the same optimal feature,and the classification accuracy of this method was 30% ～40% higher than that of SVM in the case of a few training samples.
引文
[1] BLASCHKE T.Object based image analysis for remote sensing[J].ISPRS Journal of Photogrammetry&Remote Sensing,2010,65(1):2-16.
[2] FORESTIER G,PUISSANT A,WEMMERT C,et al.Knowledge-based region labeling for remote sensing image interpretation[J].Computers Environment&Urban Systems,2012,36(5):470-480.
[3]郭亚鸽,于信芳,江东,等.面向对象的森林植被图像识别分类方法[J].地球信息科学学报,2012,14(4):514-522.(GUO Yage,YU Xinfang,JIANG Dong,et al.Study on forest classification based on object oriented techniques[J].Journal of Geo-information Science,2012,14(4):514-522.)
[4]孟晋杰,王建华.高分一号影像数据城市生态用地监测[J].测绘科学,2016,41(9):33-37.(MENG Jinjie,WANG Jianhua.Monitoring of urban ecological land use based on Gaofen-1data[J].Science of Surveying and Mapping,2016,41(9):33-37.)
[5]邵蔚,张友静,王玲玲,等.土地利用分类规则集的构建与应用研究[J].遥感信息,2012(1):88-94.(SHAO Wei,ZHANG Youjing,WANG Lingling,et al.Study on rule set construction and application to land use classification[J]. Remote Sensing Information,2012(1):88-94.)
[6]杨文涛,汪明,史培军,等.基于地形因子分割、分类的面向对象滑坡快速识别方法[J].自然灾害学报,2015,24(4):1-6.(YANG Wentao,WANG Ming,SHI Peijun,et al.Object-oriented rapid identification of landslides based on terrian factors segmentation and classification[J].Journal of Natural Disasters,2015,24(4):1-6.)
[7] KARYDAS C G,GITAS I Z.Development of an IKONOS image classification rule-set for multi-scale mapping of mediterranean rural landscapes[J].International Journal of Remote Sensing,2011,32(24):9261-9277.
[8] KE Changqing,XIA Lu,TANG Guodong,et al.Detecting urban vegetation using a object-oriented method with QuickBird imagery[C]∥2008IEEE International Geoscience&Remote Sensing Symposium.Piscataway,New Jersey,USA:IEEE,2008:138-141.
[9] ZIAEI Z,PRADHAN B,MANSOR S B.A rule-based parameter aided with object-based classification approach for extraction of building and roads from WorldView-2images[J].Geocarto International,2014,29(5):554-569.
[10]李显巨.基于新型遥感数据的典型地质环境信息智能识别[D].武汉:中国地质大学,2016.(LI Xianju.Intelligent identification of typical geo-environmental information using new remote sensing data[D].Wuhan:China University of Geosciences,2016.)
[11] VU T T,YAMAZAKI F,MATSUOKA M.Multiscale solution for building extraction from lidar and image data[J].International Journal of Applied Earth Observations&Geoinformation,2009,11(4):281-289.
[12]朱秀芳,潘耀忠,张锦水,等.训练样本对TM尺度小麦种植面积测量精度影响研究(Ⅰ):训练样本与分类方法间分类精度响应关系研究[J].遥感学报,2007,11(6):826-837.(ZHU Xiufang,PAN Yaozhong,ZHANG Jinshui,et al.The effects of training samples on the wheat planting area measure accuracy in TM scale(Ⅰ):the accuracy response of different classifiers to training samples[J].Journal of Remote Sensing,2007,11(6):826-837.)
[13]肖艳,姜琦刚,王斌,等.基于Relief F和PSO混合特征选择的面向对象土地利用分类[J].农业工程学报,2016,32(4):211-216.(XIAO Yan,JIANG Qigang,WANG Bin,et al.Object based land-use classification based on hybrid feature selection method of combining Relief F and PSO[J].Transactions of the Chinese Society of Agricultural Engineering,2016,32(4):211-216.)
[14]高国龙,杜华强,韩凝,等.基于特征优选的面向对象毛竹林分布信息提取[J].林业科学,2016,52(9):77-85.(GAO Guolong,DU Huaqiang,HAN Ning,et al.Mapping of moso bamboo forest using object-based approach based on the optimal features[J].Scientia Silvae Sinicae,2016,52(9):77-85.)
[15]SUN Xiaoyan,DU Huangqiang,HAN Ning,et al.Synergistic use of Landsat TM and SPOT5imagery for object-based forest classification[J].Journal of Applied Remote Sensing,2014,8(1):083550.
[16]吴波,朱勤东,高海燕,等.面向对象影像分类中基于最大化互信息的特征选择[J].国土资源遥感,2009(3):30-34.(WU Bo,ZHU Qindong,GAO Haiyan,et al.Feature selection based on maximal mutual information criterion in object-oriented classification[J].Remote Sensing for Land&Resources,2009(3):30-34.)
[17]张秀英,冯学智,江洪.面向对象分类的特征空间优化[J].遥感学报,2009,13(4):664-669.(ZHANG Xiuying,FENG Xuezhi,JIANG Hong.Feature set optimization in object-oriented methodology[J].Journal of Remote Sensing,2009,13(4):664-669.)
[18]BOYD R K,BASIC C,BETHEM R.Trace quantitative analysis by mass spectrometry[M].Chichester,UK:John Wiley&Sons,Ltd.,2008.
[19]RAU J Y,JHAN J P,RAU R J.Semiautomatic objectoriented landslide recognition scheme from multisensor optical imagery and DEM[J].IEEE Transactions on Geoscience&Remote Sensing,2014,52(2):1336-1349.
[20]LI Congcong,WANG Jie,WANG Lei,et al.Comparison of classification algorithms and training sample sizes in urban land classification with landsat thematic mapper imagery[J].Remote Sensing,2014,6(2):964-983.
[2] FORESTIER G,PUISSANT A,WEMMERT C,et al.Knowledge-based region labeling for remote sensing image interpretation[J].Computers Environment&Urban Systems,2012,36(5):470-480.
[3]郭亚鸽,于信芳,江东,等.面向对象的森林植被图像识别分类方法[J].地球信息科学学报,2012,14(4):514-522.(GUO Yage,YU Xinfang,JIANG Dong,et al.Study on forest classification based on object oriented techniques[J].Journal of Geo-information Science,2012,14(4):514-522.)
[4]孟晋杰,王建华.高分一号影像数据城市生态用地监测[J].测绘科学,2016,41(9):33-37.(MENG Jinjie,WANG Jianhua.Monitoring of urban ecological land use based on Gaofen-1data[J].Science of Surveying and Mapping,2016,41(9):33-37.)
[5]邵蔚,张友静,王玲玲,等.土地利用分类规则集的构建与应用研究[J].遥感信息,2012(1):88-94.(SHAO Wei,ZHANG Youjing,WANG Lingling,et al.Study on rule set construction and application to land use classification[J]. Remote Sensing Information,2012(1):88-94.)
[6]杨文涛,汪明,史培军,等.基于地形因子分割、分类的面向对象滑坡快速识别方法[J].自然灾害学报,2015,24(4):1-6.(YANG Wentao,WANG Ming,SHI Peijun,et al.Object-oriented rapid identification of landslides based on terrian factors segmentation and classification[J].Journal of Natural Disasters,2015,24(4):1-6.)
[7] KARYDAS C G,GITAS I Z.Development of an IKONOS image classification rule-set for multi-scale mapping of mediterranean rural landscapes[J].International Journal of Remote Sensing,2011,32(24):9261-9277.
[8] KE Changqing,XIA Lu,TANG Guodong,et al.Detecting urban vegetation using a object-oriented method with QuickBird imagery[C]∥2008IEEE International Geoscience&Remote Sensing Symposium.Piscataway,New Jersey,USA:IEEE,2008:138-141.
[9] ZIAEI Z,PRADHAN B,MANSOR S B.A rule-based parameter aided with object-based classification approach for extraction of building and roads from WorldView-2images[J].Geocarto International,2014,29(5):554-569.
[10]李显巨.基于新型遥感数据的典型地质环境信息智能识别[D].武汉:中国地质大学,2016.(LI Xianju.Intelligent identification of typical geo-environmental information using new remote sensing data[D].Wuhan:China University of Geosciences,2016.)
[11] VU T T,YAMAZAKI F,MATSUOKA M.Multiscale solution for building extraction from lidar and image data[J].International Journal of Applied Earth Observations&Geoinformation,2009,11(4):281-289.
[12]朱秀芳,潘耀忠,张锦水,等.训练样本对TM尺度小麦种植面积测量精度影响研究(Ⅰ):训练样本与分类方法间分类精度响应关系研究[J].遥感学报,2007,11(6):826-837.(ZHU Xiufang,PAN Yaozhong,ZHANG Jinshui,et al.The effects of training samples on the wheat planting area measure accuracy in TM scale(Ⅰ):the accuracy response of different classifiers to training samples[J].Journal of Remote Sensing,2007,11(6):826-837.)
[13]肖艳,姜琦刚,王斌,等.基于Relief F和PSO混合特征选择的面向对象土地利用分类[J].农业工程学报,2016,32(4):211-216.(XIAO Yan,JIANG Qigang,WANG Bin,et al.Object based land-use classification based on hybrid feature selection method of combining Relief F and PSO[J].Transactions of the Chinese Society of Agricultural Engineering,2016,32(4):211-216.)
[14]高国龙,杜华强,韩凝,等.基于特征优选的面向对象毛竹林分布信息提取[J].林业科学,2016,52(9):77-85.(GAO Guolong,DU Huaqiang,HAN Ning,et al.Mapping of moso bamboo forest using object-based approach based on the optimal features[J].Scientia Silvae Sinicae,2016,52(9):77-85.)
[15]SUN Xiaoyan,DU Huangqiang,HAN Ning,et al.Synergistic use of Landsat TM and SPOT5imagery for object-based forest classification[J].Journal of Applied Remote Sensing,2014,8(1):083550.
[16]吴波,朱勤东,高海燕,等.面向对象影像分类中基于最大化互信息的特征选择[J].国土资源遥感,2009(3):30-34.(WU Bo,ZHU Qindong,GAO Haiyan,et al.Feature selection based on maximal mutual information criterion in object-oriented classification[J].Remote Sensing for Land&Resources,2009(3):30-34.)
[17]张秀英,冯学智,江洪.面向对象分类的特征空间优化[J].遥感学报,2009,13(4):664-669.(ZHANG Xiuying,FENG Xuezhi,JIANG Hong.Feature set optimization in object-oriented methodology[J].Journal of Remote Sensing,2009,13(4):664-669.)
[18]BOYD R K,BASIC C,BETHEM R.Trace quantitative analysis by mass spectrometry[M].Chichester,UK:John Wiley&Sons,Ltd.,2008.
[19]RAU J Y,JHAN J P,RAU R J.Semiautomatic objectoriented landslide recognition scheme from multisensor optical imagery and DEM[J].IEEE Transactions on Geoscience&Remote Sensing,2014,52(2):1336-1349.
[20]LI Congcong,WANG Jie,WANG Lei,et al.Comparison of classification algorithms and training sample sizes in urban land classification with landsat thematic mapper imagery[J].Remote Sensing,2014,6(2):964-983.