基于时间序列遥感影像及DTW算法的塞罕坝林场树种识别研究
|
摘要
森林树种的识别和分布是森林资源监测的重要内容,是森林生态规划的基础。以塞罕坝机械林场为研究区域,利用14个时相的哨兵2号遥感影像组成时间序列数据,构建红、绿、蓝、近红外四基础波段特征时间序列和包括归一化植被指数(Normalized differential vegetation index,NDVI)在内的五特征时间序列,对这2个多特征时间序列分别采用动态时间规整(Dynamic time warping,DTW)算法计算的距离作为相似性度量标准,然后利用K均值算法完成研究区域树种聚类,并对树种识别精度分别进行评价。研究结果显示,NDVI特征时间序列非常有效体现植被的物候信息,加入NDVI特征时间序列的五特征时间序列运用DTW-K均值方法可以达到现实中树种分类调查精度的需求,总体分类的精度为88.58%,Kappa系数为0.84;落叶松分布最为广泛,面积达到35 892.23 hm~2,分类精度为94.14%,多集中分布于研究区域北部和东南部;桦树次之,面积为18 376.24 hm~2,其分类精度为76.30%,东部地区桦树分布明显高于西部地区;樟子松、云杉和柞树的分布面积较少,分别为8 749.30 hm~2、1 217.08 hm~2和3 814.40 hm~2,分类精度分别为83.82%、69.88%、72.22%。
Forest tree species identification and distribution is the basis of forest resource monitoring and forest ecological planning. Taking Saihanba Machinery Forest Farm as the study area, the four multi-characteristic time series of red, green, blue, near infrared and the five multi-characteristic time series of red, green, blue, near infrared and NDVI were constructed by the time series data consisting of 14 time-phase Sentinel 2 remote sensing images, and for these two multi-feature time series,the distance calculated by Dynamic time warping algorithm was used as the similarity measurement standard.Then the K-means algorithm was used to complete tree species clustering in the study areas,and the accuracy of tree species identification was evaluated.The results show that the five characteristic time series with NDVI feature time series can effectively reflect the phenological information of vegetation,and the DTW-K means method can meet the demand of tree species classification survey precision in reality.The overall classification accuracy is 88.58%,and the Kappa coefficient is0.84;Larch is the most widely distributed,with an area of 35 892.23 hm~2 and a classification accuracy of 94.14%,most of which are distributed in the north and southeast of the study area.The area of birch is 18 376.24 hm~2,and the precision of classification is 76.30%.The distribution of birch in the eastern region is obviously higher than that in the western area.The distribution areas of scotch pine,spruce and oak were 8 749.30 hm~2,1 217.08 hm~2 and3 814.40 hm~2,respectively.The classification accuracy of scotch pine,spruce and oak was83.82%,69.88%and 72.22%,respectively.
Forest tree species identification and distribution is the basis of forest resource monitoring and forest ecological planning. Taking Saihanba Machinery Forest Farm as the study area, the four multi-characteristic time series of red, green, blue, near infrared and the five multi-characteristic time series of red, green, blue, near infrared and NDVI were constructed by the time series data consisting of 14 time-phase Sentinel 2 remote sensing images, and for these two multi-feature time series,the distance calculated by Dynamic time warping algorithm was used as the similarity measurement standard.Then the K-means algorithm was used to complete tree species clustering in the study areas,and the accuracy of tree species identification was evaluated.The results show that the five characteristic time series with NDVI feature time series can effectively reflect the phenological information of vegetation,and the DTW-K means method can meet the demand of tree species classification survey precision in reality.The overall classification accuracy is 88.58%,and the Kappa coefficient is0.84;Larch is the most widely distributed,with an area of 35 892.23 hm~2 and a classification accuracy of 94.14%,most of which are distributed in the north and southeast of the study area.The area of birch is 18 376.24 hm~2,and the precision of classification is 76.30%.The distribution of birch in the eastern region is obviously higher than that in the western area.The distribution areas of scotch pine,spruce and oak were 8 749.30 hm~2,1 217.08 hm~2 and3 814.40 hm~2,respectively.The classification accuracy of scotch pine,spruce and oak was83.82%,69.88%and 72.22%,respectively.
引文
[1] 森林遥感分类技术研究——以浙西北山区为例[D].杭州:浙江大学,2003.
[2] 沈文娟,李明诗,黄成全.长时间序列多源遥感数据的森林干扰监测算法研究进展[J].遥感学报,2017,22(6):1005-1022.
[3] 沈文娟.南方人工林森林干扰和恢复遥感监测研究——以广东省佛冈县为例[D].南京:南京林业大学,2014.
[4] 张继贤,杨贵军.单一时相遥感数据土地利用与覆盖变化自动检测方法[J].遥感学报,2005,9(3):294-299.
[5] 谭炳香,李增元,李秉柏,等.单时相双极化ENVISAT ASAR数据水稻识别[J].农业工程学报,2006,22(12):121-127.
[6] 韩涛,徐晓桃,颉耀文.基于单时相MODIS数据的决策树自动构建及分类研究[J].武汉大学学报:信息科学版,2009,34(2):191-194.
[7] 竞霞,王锦地,王纪华,等.基于分区和多时相遥感数据的山区植被分类研究[J].遥感技术与应用,2008,23(4):394-397.
[8] 温一博,范文义.多时相遥感数据森林类型识别技术研究[J].森林工程,2013,29(2):14-20.
[9] 王二丽,李存军,周静平,等.基于多时相遥感影像的北京平原人工林树种分类[J].北京工业大学学报,2017,43(5):710-718.
[10] 吴晓平,崔光照,路康.基于DTW算法的语音识别系统实现[J].电子工程师,2004,30(7):17-19.
[11] 张备伟,吴琦,刘光徽.基于DTW的交警指挥手势识别方法[J].计算机应用研究,2017,34(11):3494-3499.
[12] 孟瑜,林蕾,袁媛,等.一种基于CD-DTW距离的时间序列遥感影像土地覆盖分类方法:CN 105005784 A[P].2015-10-28.
[13] 韩晓勇,韩玲,陈鲁皖.基于DTW距离的时序NDVI数据植被信息提取——以秦巴山区为例[J].测绘工程,2016,25(3):11-16.
[14] 姚丹阳,董雪婷.塞罕坝国家级自然保护区不同森林类型植物物种多样性比较[J].林业与生态科学,2018,33(1):44-49.
[15] 常伟强,宋彦文.塞罕坝林业功能分区及经营措施探讨[J].河北林果研究,2016(4):357-360.
[2] 沈文娟,李明诗,黄成全.长时间序列多源遥感数据的森林干扰监测算法研究进展[J].遥感学报,2017,22(6):1005-1022.
[3] 沈文娟.南方人工林森林干扰和恢复遥感监测研究——以广东省佛冈县为例[D].南京:南京林业大学,2014.
[4] 张继贤,杨贵军.单一时相遥感数据土地利用与覆盖变化自动检测方法[J].遥感学报,2005,9(3):294-299.
[5] 谭炳香,李增元,李秉柏,等.单时相双极化ENVISAT ASAR数据水稻识别[J].农业工程学报,2006,22(12):121-127.
[6] 韩涛,徐晓桃,颉耀文.基于单时相MODIS数据的决策树自动构建及分类研究[J].武汉大学学报:信息科学版,2009,34(2):191-194.
[7] 竞霞,王锦地,王纪华,等.基于分区和多时相遥感数据的山区植被分类研究[J].遥感技术与应用,2008,23(4):394-397.
[8] 温一博,范文义.多时相遥感数据森林类型识别技术研究[J].森林工程,2013,29(2):14-20.
[9] 王二丽,李存军,周静平,等.基于多时相遥感影像的北京平原人工林树种分类[J].北京工业大学学报,2017,43(5):710-718.
[10] 吴晓平,崔光照,路康.基于DTW算法的语音识别系统实现[J].电子工程师,2004,30(7):17-19.
[11] 张备伟,吴琦,刘光徽.基于DTW的交警指挥手势识别方法[J].计算机应用研究,2017,34(11):3494-3499.
[12] 孟瑜,林蕾,袁媛,等.一种基于CD-DTW距离的时间序列遥感影像土地覆盖分类方法:CN 105005784 A[P].2015-10-28.
[13] 韩晓勇,韩玲,陈鲁皖.基于DTW距离的时序NDVI数据植被信息提取——以秦巴山区为例[J].测绘工程,2016,25(3):11-16.
[14] 姚丹阳,董雪婷.塞罕坝国家级自然保护区不同森林类型植物物种多样性比较[J].林业与生态科学,2018,33(1):44-49.
[15] 常伟强,宋彦文.塞罕坝林业功能分区及经营措施探讨[J].河北林果研究,2016(4):357-360.
© 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 |