Remotely Sensed Percent Tree Cover Mapping Using Support Vector Machine Combined with Autonomous Endmember Extraction
Remotely sensed forest mapping has become an important way to meet the increasing needs for forest-cover-associated data. However, accuracy for such products varies with the condition of forest ecosystem. In this paper, a support vector machine (SVM) classifier combined with autonomous endmember extraction technique was performed to improve the performance of machine learning in satellite land cover classification and percent tree cover mapping. For the study area, Pingnan County, Guangxi Zhuang Autonomous Region, China, that featured as a complex and fragmented subtropical forest habitat, the TM imagery was first processed with SMACC endmember extraction to find spectral endmembers of expected land cover classes. Secondly, the refined endmembers were input into SVM instead of conventional visual selection of training ROIs. The percent tree cover for the county is 53.6％, underestimated by 1.3％ when compared with the National Continuous Forest Inventory 2004 statistics, suggesting a fair agreement with ground truth. The approach also shows a robust performance with an overall RMSE of 10.1.
percent tree cover TM SMACC SVM classification
Liming Bai Hui Lin Hua Sun Zhuo Zang Dengkui Mo
Center of Forestry Remote Sensing Information Engineering Central South University of Forestry and Technology Changsha, Hunan Province 410004, China
国际会议
2010 Second Asia-Pacific Conference on Information Processing(2010年第二届亚太地区信息处理国际会议 APCIP 2010)
南昌
英文
514-517
2010-09-17（万方平台首次上网日期，不代表论文的发表时间）