关键词: 遥感图像, 森林类型分类, 纹理特征, 小波变换, 支持向量机法

Abstract: Remote sensing forest classification is fast and economical method to survey, inspect and analyze vast  forest resource. However, it is very difficult to classify forest species accurately because of their similar spectrums. Therefore, with the assistance of high spatial resolution images and GIS, broadleaf, conifer and bamboo images are segmented with window on Landsat TM image of riverhead forest. Wavelet coefficient l1 norm of texture features are got as classification feature vectors after the sample images are decomposed with wavelet transform. Finally, TM image is classified with Support Vector Machine（SVM）. Results show that the average classification accuracies for broadleaf, conifer and bamboo in TM image are over 80%. Forest species can be distinguished correctly with SVM. Total classification accuracy for the whole image is 90.2% and Kappa coefficient is 0.77, which are higher than those with neural network and maximum likelihood algorithms. Errors mainly result from mixed forests and intersection between two forest types. The forest species classification accuracy in remote sensing images can be enhanced with the algorithm effectively.

Key words: remote sensing image, forest species classification, texture, wavelet transform, Support Vector Machine（SVM）