Ensemble Learning for Hyperspectral Remote Sensing Image Classification

摘要：

Hyperspectral remote sensing imagery is often used for precise classification due to its detailed spectral and spatial information. The performance of traditional hyperspectral classification method such as SAM is limited to achieve high accuracy. In this paper, the performance of Spectral Information Divergence (SID), Support Vector Machine (SVM) and ensemble learning (random forest algorithm) are investigated and compared with the tradition approaches. An Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) dataset acquired from Indian Pines, USA is applied to the above classifiers. The results demonstrate that the advantages of the three algorithms can obtain higher accuracies than SAM, especially SVM and RF outperforms other methods. The impact of number of trees in random forest is also studied. Although there is some diversity between SVM and RF, the two methods can be considered as the attractive aspects for classifying hyperspectral remote sensing images.

关键词： Hyperspectral Remote Sensing Classification Ensemble Learning Random Forest SVM

作者: Shaojie Chen Junshi Xia Peijun Du Wang Tingting

作者单位: School of Resource Engineering Longyan University LongYan, Fujiang Province 364012,China Department of Remote Sensing and GIScience China University of Mining and Technology Xuzhou, Jiangsu Geomatic College Shandong University of Science and Technology Qingdao, Shandong Province 266510,Chi

会议类型: 国际会议

会议名称: 2010 International Conference on Circuit and Signal Processing(2010年电路与信号处理国际会议 ICCSP 2010)

会议地点: 上海

会议语种:英文

页码: 286-289

在线出版日期: 2010-12-25（万方平台首次上网日期，不代表论文的发表时间）