Novel Approach for Recognizing Bridges over Water in Large Remote Sensing Images

There have been many approaches to recognize bridges, but most of them can not be effectively used on real-time occasions. In this paper, a novel approach is presented to rapidly recognize bridges over water in large remote sensing imagery. First, the rough river regions are quickly detected based on fuzzy pattern classification and clustering analysis. Next, with the knowledge of relationship between a bridge and the river under it, regions of interest are obtained. Then, in a ROI sub-image, several features of a candidate bridge and its relative water regions are extracted. At last, three rules are applied to judge whether the candidate bridge is a real one or not. Experiments on real images were conducted, and the results demonstrate that the approach is feasible and effective.