基于Sentinel-1A雷达影像的思茅松林蓄积量估测
|
摘要
探讨C波段雷达影像估测森林蓄积量的应用潜力,建立思茅松林蓄积量遥感估测模型,为利用遥感技术快速、准确、大面积的估测森林蓄积量提供参考。以云南省普洱市思茅区思茅松林为研究对象,采用C波段双极化合成孔径雷达Sentinel-1A影像为数据源,提取影像不同极化方式下的后向散射系数,并分别计算4个窗口(5×5、7×7、9×9、11×11)下的9种纹理特征,共计提取75维影像特征作为备选自变量,结合45块地面蓄积量调查样地,采用随机森林算法,进行建模因子重要性分析,选择最优特征,即选取VH极化方式、5×5窗口下VH极化方式的均值和异质性、7×7和9×9窗口下VH极化方式的最大概率、11×11窗口下VH极化方式的最大概率和协同性,共7个特征因子,建立随机森林蓄积量估测模型,R~2达到0.64,RMSE为30.35m~3/hm~2,模型的估测精度达到75.46%,森林蓄积量估测效果较好。研究表明,基于C波段双极化雷达影像提取纹理特征,利用随机森林算法进行特征选择,建立的森林蓄积量估测模型具有一定的可行性和推广性。
This study explored the potential of applying C band radar image to estimate forest stock volume,and to establish the estimation model of Simao pine(Pinus kesiya var.langbianensis)forest stock volume remote sensing.At the same time,it is to provide a reference for the rapid,accurate and large area estimation of forest storage volume through remote sensing technology.By taking the Simao pine forest in Simao District,Pu'er City,Yunnan Province as the research object,the C-band dual-polarized synthetic aperture radar Sentinel-1 A radar image was used as the data source to extract the back scattering coefficient under different polarization modes of the image,and 9 kinds of texture features were calculated under 4 different sizes of windows(5×5,7×7,9×9,11×11 respectively).A total of 75 image features were selected as alternative independent variables,combined with 45 survey sample plots of ground forest stock volume.The random forest algorithm was used to select the optimal features and it's applied to analyze the importance of the modeling factor,and a total of 7 characteristic factors were calculated,namely the utilization of VH polarization mode,the mean and dissimilarity of the VH polarization mode under the 5×5 window,the maximum probability of VH polarization mode under 7×7 and 9×9 windows, and the maximum probability and homogeneity of VH polarization under 11×11 windows.These factors were utilized to establish a random forest stock estimation model,which resulted into an R~2 of 0.64 and an RMSE of 30.35 m~3/hm~2,and the estimated accuracy of the model reached 75.46%,indicating a good result of forest stock volume estimation.The research shows that, by using the random forest algorithm,the texture feature extraction based on C-band dual-polarized radar image has a good feasible and propagable for estimating the forest stock volume.
This study explored the potential of applying C band radar image to estimate forest stock volume,and to establish the estimation model of Simao pine(Pinus kesiya var.langbianensis)forest stock volume remote sensing.At the same time,it is to provide a reference for the rapid,accurate and large area estimation of forest storage volume through remote sensing technology.By taking the Simao pine forest in Simao District,Pu'er City,Yunnan Province as the research object,the C-band dual-polarized synthetic aperture radar Sentinel-1 A radar image was used as the data source to extract the back scattering coefficient under different polarization modes of the image,and 9 kinds of texture features were calculated under 4 different sizes of windows(5×5,7×7,9×9,11×11 respectively).A total of 75 image features were selected as alternative independent variables,combined with 45 survey sample plots of ground forest stock volume.The random forest algorithm was used to select the optimal features and it's applied to analyze the importance of the modeling factor,and a total of 7 characteristic factors were calculated,namely the utilization of VH polarization mode,the mean and dissimilarity of the VH polarization mode under the 5×5 window,the maximum probability of VH polarization mode under 7×7 and 9×9 windows, and the maximum probability and homogeneity of VH polarization under 11×11 windows.These factors were utilized to establish a random forest stock estimation model,which resulted into an R~2 of 0.64 and an RMSE of 30.35 m~3/hm~2,and the estimated accuracy of the model reached 75.46%,indicating a good result of forest stock volume estimation.The research shows that, by using the random forest algorithm,the texture feature extraction based on C-band dual-polarized radar image has a good feasible and propagable for estimating the forest stock volume.
引文
[1]李海奎,雷渊才,曾伟生.基于森林清查资料的中国森林植被碳储量[J].林业科学,2011,47(7):7-12.
[2]潘帅,李娟,徐新,等.利用遥感数据估测森林蓄积量[J].吉林林业科技,2013,42(4):30-32.
[3]潘磊,孙玉军.应用Sentinel-1影像纹理信息模型估测杉木林生物量[J].东北林业大学学报,2018,46(1):58-62.
[4]程武学,杨存建,周介铭,等.森林蓄积量遥感定量估测研究综述[J].安徽农业科学,2009,37(16):7746-7750.
[5]朱海珍,庞勇,杨飞,等.基于ENVISAT ASAR数据的森林蓄积量估测研究[J].地理与地理信息科学,2007,23(2):51-55.
[6]徐天蜀,岳彩荣,章皖秋,等.基于多源遥感数据的森林生物量定量评价研究[J].西南林业大学学报,2016,36(3):126-130.
[7]李浩,张超,许彦红.思茅松空间分布特征研究[J].西南林业大学学报,2015,35(1):47-51.
[8]杨魁,杨建兵,江冰茹.Sentinel-1卫星综述[J].城市勘测,2015,4(2):24-27.
[9]张永红,张继贤,林宗坚,等.地形引起的雷达辐射畸变及其校正[J].测绘科学,2002,27(4):23-26.
[10]Small D.Flattening gamma:Radiometric terrain correction for SAR imagery[J].IEEE Transactions on Geoscience & Remote Sensing,2011,49(8):3081-3093.
[11]魏钜杰.复杂地形区域合成孔径雷达正射影像制作方法研究[D].阜新:辽宁工程技术大学,2009.
[12]李成绕,薛东剑,张露,等.基于Sentinel-1A卫星SAR数据的水体提取方法研究[J].地理空间信息,2018,16(1):38-40.
[13]吴樊,王超,张红.基于纹理特征的高分辨率SAR影像居民区提取[J].遥感技术与应用,2005,20(1):148-152.
[14]胡召玲,李海权,杜培军.SAR图像纹理特征提取与分类研究[J].中国矿业大学学报,2009,38(3):422-427.
[15]Haralick R M,Shanmugam K,Dinstein I.Textural Features for Image Classification[J].Studies in Media and Communication,1973,SMC-3(6):610-621.
[16]杨柳,冯仲科,岳德鹏,等.结合纹理因子和地形因子的森林蓄积量多光谱估测模型[J].光谱学与光谱分析,2017,37(7):2140-2145.
[17]甘胜江,白艳宇,孙连海,等.融合改进K近邻和随机森林的机器学习方法[J].计算机工程与设计,2017,38(8):2251-2255.
[18]李欣海.随机森林模型在分类与回归分析中的应用[J].应用昆虫学报,2013,50(4):1190-1197.
[2]潘帅,李娟,徐新,等.利用遥感数据估测森林蓄积量[J].吉林林业科技,2013,42(4):30-32.
[3]潘磊,孙玉军.应用Sentinel-1影像纹理信息模型估测杉木林生物量[J].东北林业大学学报,2018,46(1):58-62.
[4]程武学,杨存建,周介铭,等.森林蓄积量遥感定量估测研究综述[J].安徽农业科学,2009,37(16):7746-7750.
[5]朱海珍,庞勇,杨飞,等.基于ENVISAT ASAR数据的森林蓄积量估测研究[J].地理与地理信息科学,2007,23(2):51-55.
[6]徐天蜀,岳彩荣,章皖秋,等.基于多源遥感数据的森林生物量定量评价研究[J].西南林业大学学报,2016,36(3):126-130.
[7]李浩,张超,许彦红.思茅松空间分布特征研究[J].西南林业大学学报,2015,35(1):47-51.
[8]杨魁,杨建兵,江冰茹.Sentinel-1卫星综述[J].城市勘测,2015,4(2):24-27.
[9]张永红,张继贤,林宗坚,等.地形引起的雷达辐射畸变及其校正[J].测绘科学,2002,27(4):23-26.
[10]Small D.Flattening gamma:Radiometric terrain correction for SAR imagery[J].IEEE Transactions on Geoscience & Remote Sensing,2011,49(8):3081-3093.
[11]魏钜杰.复杂地形区域合成孔径雷达正射影像制作方法研究[D].阜新:辽宁工程技术大学,2009.
[12]李成绕,薛东剑,张露,等.基于Sentinel-1A卫星SAR数据的水体提取方法研究[J].地理空间信息,2018,16(1):38-40.
[13]吴樊,王超,张红.基于纹理特征的高分辨率SAR影像居民区提取[J].遥感技术与应用,2005,20(1):148-152.
[14]胡召玲,李海权,杜培军.SAR图像纹理特征提取与分类研究[J].中国矿业大学学报,2009,38(3):422-427.
[15]Haralick R M,Shanmugam K,Dinstein I.Textural Features for Image Classification[J].Studies in Media and Communication,1973,SMC-3(6):610-621.
[16]杨柳,冯仲科,岳德鹏,等.结合纹理因子和地形因子的森林蓄积量多光谱估测模型[J].光谱学与光谱分析,2017,37(7):2140-2145.
[17]甘胜江,白艳宇,孙连海,等.融合改进K近邻和随机森林的机器学习方法[J].计算机工程与设计,2017,38(8):2251-2255.
[18]李欣海.随机森林模型在分类与回归分析中的应用[J].应用昆虫学报,2013,50(4):1190-1197.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |