基于像素-空间上下文特征的高分辨率遥感影像分类

Classification of High Resolution Remote Sensing Image Based on Pixel and Spatial Context Features

针对高分辨率遥感影像分类中传统的像素级分类方法的“椒盐”现象和面向对象方法平滑地物细节的现象，提出了一种融合像素-空间上下文特征的分类方法。该方法结合了面向像素和面向对象的分类思想，具体为首选采用改进分水岭变换进行图像分割以获得像素的空间上下文，然后计算像素的上下文特征并结合像素本身的光谱特征构成多源分类特征向量，最后采用支持向量机对归一化特征进行分类。实验结果表明该方法既能有效减少传统方法的“椒盐”现象，又能保持地物的相对完整性和细节信息，提高地物分类精度。

A classification method based on the fusion of pixel and context features is pro- posed in this paper in view of the ＂salt and pepper＂ phenomenon of traditional pixel level methods and the excessive smoothing phenomenon of object oriented methods on the classifi- cation of high resolution remote sensing image. The method integrates the classification thoughts of pixel and object oriented methods. Specifically, image segmentation based on improved watershed transform is firstly used to obtain the spatial context of pixels. Then, the context features of pixels are calculated and the multiple feature vector is formed combi- ning the spectral features of the pixels themselves. Finally, support vector machine is used to classify the normalized features. Experimental results show that this method can not only effectively reduce the ＂salt and pepper＂ phenomenon of traditional methods but also can keep the relative integrity and detail information and improve the accuracy of the classification.