Water Bridge Target Recognition Algorithm

Automatic target recognition is quite significant in computer vision and artificial intelligence, however still a challenge nowadays. In recent years, the automatic recognition of artificial objects in natural scene has attracted extensive interests, which has potential application in civil field as well as military field. As a typical representative, bridge recognition has become a hot field, but there has been no effective and adaptive method in the literature to our best acknowledge yet.This paper mainly discusses the recognition algorithm of bridges over river in aerial or satellite optic remote-sensing picture, and basic bridge recognition algorithm and multi-resolution analysis method are proposed.In basic method, the knowledge model about bridge and its scene is built first. Then preprocess picture by selective line stretch aimed at enhancing the gray range of interest' contrast. Picture is segmented by the combine of gray selection & iterative threshold & user-defined variance, and the river is connected by morphological close based on chain. By the context of bridge and river, whether there are rivers containing bridges can be detected and the searching range is reduced using river traverses the edge of picture and the number of traversing the edge or the area increase after connection. After that, by the second segmentation and the knowledge of bridge passing by two riverside lines of one river(including the combination of riverside lines)and so on, hypothesis and verification method is used to recognize the bridges and quantificationally describe them in the end. Every step is guided by knowledge, and the basic operations include fast tracing boundary contours approach, contour filling algorithm based on chain codes description and morphological open and close based on chain. Owing to the restrict of width in basic method and aimed at reducing time complexity, multi-resolution method is presented, which reduces the resolution by linear interpolation algorithm, detects the bridge in low-resolution, then finds the regions of interest in original picture, rotates them and recognizes bridges by horizontal and vertical projection.Experiments prove the algorithm proposed in this paper can instantly detects whether the picture contain bridges, avoiding blindfold operation and locate the bridges accurately. In addition, it is adaptive, accurate, and has low false-alarming and missing-detected ration and high computational efficiency, satisfying the real-time need.