A Method to Rapidly Detect Great Rivers in High-Resolution Satellite Images

An algorithm for rapidly detecting great rivers in high-resolution remote sensing imagery is presented. Based on the fact that rivers in high-resolution satellite imagery have the characteristic that their gray value is smaller than that of theirs backgrounds, the original large image is firstly divided into small sub-images with the same size and all sub-images are classified as river area or non-river area by fuzzy pattern classification. According to another characteristic of rivers in high resolution remote sensing imagery that their shapes are general long-rectangles, all classified river area sub-images are gathered to obtain several sets by a nearest neighbor clustering method and wrongly-classified sub-images are deleted in terms of the ratio of length-to-width of the minimum enclosing rectangle formed by pixels in each set, meanwhile, the real rivers are detcted. Experimental results on real satellite images with high resolution demonstrate the efficiency of the proposed method.