Remote Sensing Water-land Scenery Classification Based On Texture Analysis

With the resolution’s improvement of the satellite and aerial image, remote sensingimages contain more abundant information. As a result, remote sensing technology iswidely applied in many fields. Remote sensing image scenery classification is an importbranch in remote sensing image processing. Theories and methods are developedcontinuously recent years. Many new feature extraction and classification methods havebeen generated. Valid feature extraction and classification methods play key roles onclassification effect.This paper proposed a new method for remote sensing image scenery classification,which can efficiently classify the regions of the image as water areas and land areas. In theproposed method, we added in the region merging processing after coarse segmentationand classify the achieved regions to obtain the classification result that can describe theterrain under a large scale.Firstly, as remote sensing images show strong texture feature, we adopt texture as thedescription of images. Through the analysis of several texture features, we choose thefeature with higher accuracy and efficiency as the final classification feature.Secondly, do clustering segmentation on the original image with the extracted texturefeatures and obtain the initial result. Then, merge the regions of the former result andobtain several regions which are large-scaled and more semantics while the small regionsare removed.In the final step, a measurement called edge density is defined as a feature for regions.By statistically analyzing plenty of samples, we can determine an empirical value as theclassification threshold to distinguish the water and land. According to the threshold, wecan label each region as related class (water/land) and obtain the classification result oforiginal input image.