A New Classification Method of High Spatial Resolution Remote Sensing Image

Along with the spatial resolution of remote sensing images getting higher and higher, the complex structure in the simple objects becomes obvious, which makes the classification algorithms based on pixels begin losing their advantages. The huge number of clastic plaques contained in the classification results not only have no clear meanings, but also disturb the post-process. The phenomenon is much more pronounced in the classification studies of high spatial resolution remote sensing images. In order to solve the problem, this paper proposes a new method that integrates image segmentation method and pixel based classification method. In the proposed method, we segment the image and classify the image at the same time, link the two results with geographic location, and make the segmentation regions get the class label, which the mode position-relative pixels have. Results of Classification experiments by SPOT-5 & QuickBird images manifest that the new method performs well in classifying high spatial resolution images.