基于Sentinel-2A影像特征优选的随机森林土地覆盖分类
|
摘要
中南半岛地处热带、亚热带地区,由于水热条件适宜,植被生长旺盛,土地利用强度高,地表覆盖类型的光谱特征时空变异复杂,使用传统的基于光谱特征的遥感分类精度难以保证。Sentinel-2A卫星遥感数据具有较丰富的光谱波段和较高的空间分辨率,为土地覆盖遥感分类提供了多维特征空间。但多维特征参与分类容易造成信息冗余,从而导致分类速度和精度降低。因此,如何充分利用Sentinel-2A数据丰富的光谱和空间信息,并通过高维特征空间降维进行特征优选对于提高分类精度具有重要意义。本文以中南半岛典型地区土地覆盖分类为例,利用Sentinel-2A多波段光谱特征,归一化植被指数(NDVI)、比值植被指数(RVI)、差值植被指数(DVI)、归一化水体指数(NDWI)等指数特征以及对比度、相关性、能量、均值、熵等纹理特征,在随机森林模型框架下,采用平均不纯度减少方法对不同特征在土地覆盖分类中的重要程度进行识别;利用袋外(OOB)误差方法,对重要特征组合进行了优选;利用优选特征进行随机森林土地覆盖分类,并与原始随机森林分类结果进行对比。结果表明:Sentinel-2A影像的光谱特征和纹理特征在土地覆盖分类中具有较为重要的作用,光谱特征中短波红外、可见光、植被红边波段重要性较大,纹理特征中均值、能量法重要性较高。选择重要性列前9位的特征参与分类时,OOB精度达到最高;继续增加特征会使模型复杂度过高,容易发生过拟合而使得分类精度不增反降。通过特征优选高效利用了Sentinel-2A丰富的光谱和纹理信息,其总体分类精度达87.53%,Kappa系数达0.8461,优于原始随机森林方法,一定程度上提高了热带亚热带地区复杂土地覆盖分类精度。
Due to the suitable hydrothermal conditions, vigorous vegetation growth, high land use intensity and complex spatiotemporal variation of spectral characteristics of surface cover types, it is difficult to guarantee the accuracy of remote sensing classification using traditional spectral characteristics in tropical and subtropical regions. Multi-spectral, high spatial resolution Sentinel-2 A imageries provide a new source of data for land-cover classification. In order to improve the speed and accuracy of land-cover classification using Sentinel-2 A images, we propose a classification method with feature-optimized random forests. In this study, we took the Mun River Basin of Indo-China Peninsula as our research area and made full use of the rich spectral characteristics, normalized vegetation index(NDVI), ratio vegetation index(RVI), difference vegetation index(DVI), normalized water body index(NDWI), and texture features including contrast, correlation, energy, mean, and entropy, of Sentinel-2 A images for the analyses. We used the average impurity reduction method in random forests to evaluate the importance of different spectral features, indices, and texture features. Combining the out-of-bag(OOB) error to select features, the results of land-cover classification with feature-optimized random forests were obtained. They show that the spectral features and texture features of Sentinel-2 A images play an important role in our classification compared with the original random forest land-cover classification results. The short-wave infrared, visible, and vegetation red-edge bands are of greater importance in spectral features, and the mean and energy are of high importance in texture features.The accuracy of OOB is the highest when the top 9 important features are selected. Sentinel-2 A images have good adaptability in tropical and subtropical region land-cover classification. It can effectively improve the accuracy of land-cover classification in tropical and subtropical regions.The accuracy of our classification method reaches 87.53%, and the Kappa coefficient reaches0.8461, better than the original random forest method. The random forest method based on feature optimization not only has a fast classification speed, but also can guarantee high classification accuracy under the condition that the sample is representative, especially suitable for the landcover classification of medium and high spatial resolution images of Sentinel-2 A.
Due to the suitable hydrothermal conditions, vigorous vegetation growth, high land use intensity and complex spatiotemporal variation of spectral characteristics of surface cover types, it is difficult to guarantee the accuracy of remote sensing classification using traditional spectral characteristics in tropical and subtropical regions. Multi-spectral, high spatial resolution Sentinel-2 A imageries provide a new source of data for land-cover classification. In order to improve the speed and accuracy of land-cover classification using Sentinel-2 A images, we propose a classification method with feature-optimized random forests. In this study, we took the Mun River Basin of Indo-China Peninsula as our research area and made full use of the rich spectral characteristics, normalized vegetation index(NDVI), ratio vegetation index(RVI), difference vegetation index(DVI), normalized water body index(NDWI), and texture features including contrast, correlation, energy, mean, and entropy, of Sentinel-2 A images for the analyses. We used the average impurity reduction method in random forests to evaluate the importance of different spectral features, indices, and texture features. Combining the out-of-bag(OOB) error to select features, the results of land-cover classification with feature-optimized random forests were obtained. They show that the spectral features and texture features of Sentinel-2 A images play an important role in our classification compared with the original random forest land-cover classification results. The short-wave infrared, visible, and vegetation red-edge bands are of greater importance in spectral features, and the mean and energy are of high importance in texture features.The accuracy of OOB is the highest when the top 9 important features are selected. Sentinel-2 A images have good adaptability in tropical and subtropical region land-cover classification. It can effectively improve the accuracy of land-cover classification in tropical and subtropical regions.The accuracy of our classification method reaches 87.53%, and the Kappa coefficient reaches0.8461, better than the original random forest method. The random forest method based on feature optimization not only has a fast classification speed, but also can guarantee high classification accuracy under the condition that the sample is representative, especially suitable for the landcover classification of medium and high spatial resolution images of Sentinel-2 A.
引文
[1]张景,姚凤梅,徐永明,等.基于MODIS的土地覆盖遥感分类特征的评价与比较[J].地理科学, 2010, 30(2):248-253.[Zhang J,Yao F M, Xu Y M, et al. Comparison and evaluation of classification features in land cover based on remote sensing[J]. Scientia Geographica Sinica, 2010, 30(2):248-253.]
[2]宋军伟,张友静,李鑫川,等.基于GF-1与Landsat-8影像的土地覆盖分类比较[J].地理科学进展, 2016, 35(2):255-263.[Song J W, Zhang Y J, Li X C, et al. Comparison between GF-1and Landsat-8 images in land cover classification[J]. Progress in Geography, 2016, 35(2):255-263.]
[3] Guan X, Huang C, Liu G, et al. Mapping rice cropping systems in Vietnam using an NDVI-based time-series similarity measurement based on DTW distance[J]. Remote Sensing, 2016, doi:10.3390/rs8010019.
[4]管续栋,黄翀,刘高焕,等.基于DTW距离的时序相似性方法提取水稻遥感信息:以泰国为例[J].资源科学, 2014, 36(2):267-272.[Guan X D, Huang C, Liu G H, et al. Extraction of paddy rice area using a DTW distance based similarity measure:Taking Thailand as an example[J]. Resources Science, 2014, 36(2):267-272.]
[5] Lefebvre A, Sannier C, Corpetti T. Monitoring urban areas with Sentinel-2A Data:Application to the update of the Copernicus high resolution layer imperviousness degree[J]. Remote Sensing,2016, DOI:10. 3390/rs8070606.
[6] Immitzer M, Vuolo F, Atzberger C. First experience with Sentinel-2 Data for crop and tree species classifications in central Europe[J]. Remote Sensing, 2016, DOI:10. 3390/rs8030166.
[7]时丽娜,藉雪峰,王星.利用Sentinel-2A数据的西藏阿里冰崩范围提取[J].内蒙古煤炭经济, 2017,(2):157-160.[Shi L N,Xue X F, Wang X. Extraction of the ice depression range of Tibet Ali using Sentinel-2A data[J]. Inner Mongolia Coal Economy,2017,(2):157-160.]
[8] Rujoiu-Mare M, Olariu B, Mihai B, et al. Land cover classification in Romanian Carpathians and Sub Carpathians using multidate Sentinel-2 remote sensing imagery[J]. European Journal of Remote Sensing, 2017, 50(1):496-508.
[9] Hayes M M, Miller S N, Murphy M A. High-resolution land cover classification using Random Forest[J]. Remote Sensing Letters,2014, 5(2):112-121.
[10]姚登举,杨静,詹晓娟.基于随机森林的特征选择算法[J].吉林大学学报(工学版), 2014, 44(1):137-141.[Yao D J, Yang J,Zhan X J. Feature selection algorithm based on random forest[J].Journal of Jilin University(Engineering Science), 2014, 44(1):137-141.]
[11] Pal M. Random forest classifier for remote sensing classification[J]. International Journal of Remote Sensing, 2005, 26(1):217-222.
[12]张晓羽,李凤日,甄贞,等.基于随机森林模型的陆地卫星-8遥感影像森林植被分类[J].东北林业大学学报, 2016, 44(6):53-57.[Zhang X Y, Li F R, Zhen Z, et al. Forest vegetation classification of landsat8 remote sensing image based on random forests model[J]. Journal of Northeast Forestry University, 2016, 44(6):53-57.]
[13]宫攀.基于MODIS数据关键物候特征参数的东北地区植被覆盖分类[J].资源科学, 2010, 32(6):1154-1160.[Gong P. Vegetation classification based on phenology indices derived from MODIS Data in Northeastern China[J]. Resources Science, 2010, 32(6):1154-1160.]
[14]张喆,丁建丽,李鑫,等. TVDI用于干旱区农业旱情监测的适宜性[J].中国沙漠, 2015, 35(1):220-227.[Zhang Z, Ding J L, Li X,et al. Suitability of TVDI used to monitor agricultural drought in arid area[J]. Journal of Desert Research, 2015, 35(1):220-227.]
[15]周晓双,姚霞,田永超,等.基于高光谱的稻麦叶面积指数监测研究[C].北京:2014年中国作物学会学术年会论文集, 2014.[Zhou X S, Yao X, Tian Y C, et al. Monitoring of Rice Leaf Area Index Based on Hyperspectral[C]. Beijing:China Crop Society Academic Annual Meeting, 2014.]
[16]陈文倩,丁建丽,李艳华,等.基于国产GF-1遥感影像的水体提取方法[J].资源科学, 2015, 37(6):1166-1172.[Chen W Q,Ding J L, Li Y H, et al. Extraction of water information based on China-made GF-1 remote sense image[J]. Resources Science,2015, 37(6):1166-1172.]
[17]李智峰,朱谷昌,董泰锋.基于灰度共生矩阵的图像纹理特征地物分类应用[J].地质与勘探, 2011, 47(3):456-461.[Li Z F, Zhu G C, Dong T F. Application of GLCM-based texture features to remote sensing image classification[J]. Geology and Prospecting,2011, 47(3):456-461.]
[18]胡玉福,邓良基,匡先辉,等.基于纹理特征的高分辨率遥感图像土地利用分类研究[J].地理与地理信息科学, 2011, 27(5):42-45.[Hu Y F, Deng L J, Kuang X H, et al. Study on land use classification of high resolution remote sensing image based on texture feature[J]. Geography and Geo-Information Science, 2011, 27(5):42-45.]
[19] Breiman L. Random forests, machine learning 45[J]. Journal of Clinical Microbiology, 2001, 2:199-228.
[20] Gislason P O, Benediktsson J A, Sveinsson J R. Random forests for land cover classification[J]. Pattern Recognition Letters, 2006,27(4):294-300.
[21] Fan H. Land-cover mapping in the Nujiang Grand Canyon:Integrating spectral, textural, and topographic data in a random forest classifier[J]. International Journal of Remote Sensing, 2013, 34(21):7545-7567.
[22]张洪敏,张艳芳,田茂,等.基于主成分分析的生态变化遥感监测:以宝鸡市城区为例[J].国土资源遥感, 2018(1):203-209.[Zhang H M, Zhang Y F, Tian M, et al. Remote sensing monitoring of ecological change based on principal component analysis:A case study of Baoji City[J]. Remote sensing of land and resources,2018(1):203-209.]
[23]田野,赵春晖,季亚新.主成分分析在高光谱遥感图像降维中的应用[J].哈尔滨师范大学自然科学学报, 2007, 23(5):58-60.[Tian Y, Zhao C H, Ji Y X. Application of principal component analysis in dimensionality reduction of hyperspectral remote sensing images[J]. Journal of Natural Science of Harbin Normal University, 2007, 23(5):58-60.]
[24]李冰,卢小平,李新社,等.特征优选的GF-2影像湿地地表覆盖要素提取[J].测绘与空间地理信息, 2018(9).49-52.[Li B, Lu X P, Li X S, et al. Surface coverage factor extraction of Gf-2 image of wetland with optimized features[J]. Mapping and Spatial Geographic Information, 2018(9):49-52.]
[25] Robin Genuer, Jean-Michel Poggi, Christine Tuleau-Malot. Variable selection using Random Forests[J]. Pattern Recognition Letters, 2010, 31(14):2225-2236.
[26]张磊,宫兆宁,王启为,等. Sentinel-2影像多特征优选的黄河三角洲湿地信息提取[J].遥感学报, 2018, 23(2):313-326.[Zhang L, Gong Z N, Wang Q W, et al. Sentinel-2 image multi-feature optimization for information extraction of Yellow River delta wetland[J]. Journal of Remote Sensing, 2018, 23(2):313-326.]
[2]宋军伟,张友静,李鑫川,等.基于GF-1与Landsat-8影像的土地覆盖分类比较[J].地理科学进展, 2016, 35(2):255-263.[Song J W, Zhang Y J, Li X C, et al. Comparison between GF-1and Landsat-8 images in land cover classification[J]. Progress in Geography, 2016, 35(2):255-263.]
[3] Guan X, Huang C, Liu G, et al. Mapping rice cropping systems in Vietnam using an NDVI-based time-series similarity measurement based on DTW distance[J]. Remote Sensing, 2016, doi:10.3390/rs8010019.
[4]管续栋,黄翀,刘高焕,等.基于DTW距离的时序相似性方法提取水稻遥感信息:以泰国为例[J].资源科学, 2014, 36(2):267-272.[Guan X D, Huang C, Liu G H, et al. Extraction of paddy rice area using a DTW distance based similarity measure:Taking Thailand as an example[J]. Resources Science, 2014, 36(2):267-272.]
[5] Lefebvre A, Sannier C, Corpetti T. Monitoring urban areas with Sentinel-2A Data:Application to the update of the Copernicus high resolution layer imperviousness degree[J]. Remote Sensing,2016, DOI:10. 3390/rs8070606.
[6] Immitzer M, Vuolo F, Atzberger C. First experience with Sentinel-2 Data for crop and tree species classifications in central Europe[J]. Remote Sensing, 2016, DOI:10. 3390/rs8030166.
[7]时丽娜,藉雪峰,王星.利用Sentinel-2A数据的西藏阿里冰崩范围提取[J].内蒙古煤炭经济, 2017,(2):157-160.[Shi L N,Xue X F, Wang X. Extraction of the ice depression range of Tibet Ali using Sentinel-2A data[J]. Inner Mongolia Coal Economy,2017,(2):157-160.]
[8] Rujoiu-Mare M, Olariu B, Mihai B, et al. Land cover classification in Romanian Carpathians and Sub Carpathians using multidate Sentinel-2 remote sensing imagery[J]. European Journal of Remote Sensing, 2017, 50(1):496-508.
[9] Hayes M M, Miller S N, Murphy M A. High-resolution land cover classification using Random Forest[J]. Remote Sensing Letters,2014, 5(2):112-121.
[10]姚登举,杨静,詹晓娟.基于随机森林的特征选择算法[J].吉林大学学报(工学版), 2014, 44(1):137-141.[Yao D J, Yang J,Zhan X J. Feature selection algorithm based on random forest[J].Journal of Jilin University(Engineering Science), 2014, 44(1):137-141.]
[11] Pal M. Random forest classifier for remote sensing classification[J]. International Journal of Remote Sensing, 2005, 26(1):217-222.
[12]张晓羽,李凤日,甄贞,等.基于随机森林模型的陆地卫星-8遥感影像森林植被分类[J].东北林业大学学报, 2016, 44(6):53-57.[Zhang X Y, Li F R, Zhen Z, et al. Forest vegetation classification of landsat8 remote sensing image based on random forests model[J]. Journal of Northeast Forestry University, 2016, 44(6):53-57.]
[13]宫攀.基于MODIS数据关键物候特征参数的东北地区植被覆盖分类[J].资源科学, 2010, 32(6):1154-1160.[Gong P. Vegetation classification based on phenology indices derived from MODIS Data in Northeastern China[J]. Resources Science, 2010, 32(6):1154-1160.]
[14]张喆,丁建丽,李鑫,等. TVDI用于干旱区农业旱情监测的适宜性[J].中国沙漠, 2015, 35(1):220-227.[Zhang Z, Ding J L, Li X,et al. Suitability of TVDI used to monitor agricultural drought in arid area[J]. Journal of Desert Research, 2015, 35(1):220-227.]
[15]周晓双,姚霞,田永超,等.基于高光谱的稻麦叶面积指数监测研究[C].北京:2014年中国作物学会学术年会论文集, 2014.[Zhou X S, Yao X, Tian Y C, et al. Monitoring of Rice Leaf Area Index Based on Hyperspectral[C]. Beijing:China Crop Society Academic Annual Meeting, 2014.]
[16]陈文倩,丁建丽,李艳华,等.基于国产GF-1遥感影像的水体提取方法[J].资源科学, 2015, 37(6):1166-1172.[Chen W Q,Ding J L, Li Y H, et al. Extraction of water information based on China-made GF-1 remote sense image[J]. Resources Science,2015, 37(6):1166-1172.]
[17]李智峰,朱谷昌,董泰锋.基于灰度共生矩阵的图像纹理特征地物分类应用[J].地质与勘探, 2011, 47(3):456-461.[Li Z F, Zhu G C, Dong T F. Application of GLCM-based texture features to remote sensing image classification[J]. Geology and Prospecting,2011, 47(3):456-461.]
[18]胡玉福,邓良基,匡先辉,等.基于纹理特征的高分辨率遥感图像土地利用分类研究[J].地理与地理信息科学, 2011, 27(5):42-45.[Hu Y F, Deng L J, Kuang X H, et al. Study on land use classification of high resolution remote sensing image based on texture feature[J]. Geography and Geo-Information Science, 2011, 27(5):42-45.]
[19] Breiman L. Random forests, machine learning 45[J]. Journal of Clinical Microbiology, 2001, 2:199-228.
[20] Gislason P O, Benediktsson J A, Sveinsson J R. Random forests for land cover classification[J]. Pattern Recognition Letters, 2006,27(4):294-300.
[21] Fan H. Land-cover mapping in the Nujiang Grand Canyon:Integrating spectral, textural, and topographic data in a random forest classifier[J]. International Journal of Remote Sensing, 2013, 34(21):7545-7567.
[22]张洪敏,张艳芳,田茂,等.基于主成分分析的生态变化遥感监测:以宝鸡市城区为例[J].国土资源遥感, 2018(1):203-209.[Zhang H M, Zhang Y F, Tian M, et al. Remote sensing monitoring of ecological change based on principal component analysis:A case study of Baoji City[J]. Remote sensing of land and resources,2018(1):203-209.]
[23]田野,赵春晖,季亚新.主成分分析在高光谱遥感图像降维中的应用[J].哈尔滨师范大学自然科学学报, 2007, 23(5):58-60.[Tian Y, Zhao C H, Ji Y X. Application of principal component analysis in dimensionality reduction of hyperspectral remote sensing images[J]. Journal of Natural Science of Harbin Normal University, 2007, 23(5):58-60.]
[24]李冰,卢小平,李新社,等.特征优选的GF-2影像湿地地表覆盖要素提取[J].测绘与空间地理信息, 2018(9).49-52.[Li B, Lu X P, Li X S, et al. Surface coverage factor extraction of Gf-2 image of wetland with optimized features[J]. Mapping and Spatial Geographic Information, 2018(9):49-52.]
[25] Robin Genuer, Jean-Michel Poggi, Christine Tuleau-Malot. Variable selection using Random Forests[J]. Pattern Recognition Letters, 2010, 31(14):2225-2236.
[26]张磊,宫兆宁,王启为,等. Sentinel-2影像多特征优选的黄河三角洲湿地信息提取[J].遥感学报, 2018, 23(2):313-326.[Zhang L, Gong Z N, Wang Q W, et al. Sentinel-2 image multi-feature optimization for information extraction of Yellow River delta wetland[J]. Journal of Remote Sensing, 2018, 23(2):313-326.]