River Extraction Based on Knowledge and Fuzzy Classification

Aiming at the disadvantage rivers can not be rapidly detected in high resolution remote sensing imagery, a new speedy algorithm is proposed. The knowledge of the grayness and the shape of a river are used in extracting process. With the knowledge that the grayness of river areas is smaller than that of their backgrounds and it fluctuates little, all sub-images, which are obtained to speed up processing, are classified as river area or non-river area by fuzzy classification. With the knowledge that a river consists of several long-rectangular segments, the sub-images having been classified as river areas are gathered and are judged whether they belong to a real river or not by clustering analysis. Simultaneously, the real rivers are roughly extracted. Experiments on real images were conducted, and the results demonstrate that the proposed approach is feasible and effective.