Abstract: In order to achieve automatic classification for high resolution remote sensing image and further improve the accuracy and efficiency of unsupervised classification, an object-oriented automatic classification method with automatic chosen train samples is presented in this paper. First, the image is segmented into homogeneous segments by using mean shift algorithm. Second, the multi-features of object segmentation are extracted （including spectral feature, texture feature and shape feature）. Based on the geometric distance of feature vector, the train samples are automatically chosen. Finally, the classification results are gotten through Support Vector Machine （SVM） classifier. The experiment result shows that the proposed algorithm of object-oriented automatic classification can make use of the rich feature of image object and better avoid the spiced salt phenomenon, so that both accuracy and efficiency of automatic classification have been much improved.

Key words: remote sensing image, feature vector, Support Vector Machine（SVM）, object-oriented, automatic classification

摘要： 为了实现高分辨率遥感影像自动分类及进一步提高非监督分类的精度和效率，提出了一种训练样本自动选取的面向对象自动分类方法。首先利用均值漂移算法对遥感影像进行分割，获取同质性分割单元；然后对分割对象进行多特征（光谱特征、纹理特征和形状特征）提取，基于特征向量的几何距离进行训练样本自动选择，进而利用支持向量机分类器得到分类结果。实验研究表明，提出的面向对象自动分类算法不但可以利用影像对象丰富的特征信息，而且较好地避免了“椒盐现象”，使自动分类的精度和效率得到较大提升。

关键词: 遥感影像, 特征向量, 支持向量机, 面向对象, 自动分类