Research On Bridge Recognition For Remote Sensing Image

The recognition and detection of bridge, which is one kind of typical man-made objects, has become a hot topic of research. It has very important applicable values both in military and civilian areas. In this thesis, with the recognition of bridge in remote sensing image as the main research background, problems such as feature extraction, river detection, river region correction and bridge recognition were researched, and with the above as base, a relatively complete software system of bridge recognition and localization was constructed. The main contents of this thesis are as follows:(1) For the problem of serious interference caused by similar geographic regions in remote sensing image, based on a large number of experiments and observations of geographic targets, and analysis of differences between river region and background region, corner information feature is proposed. The local entropy, texture and corner information were employed as the feature vectors to train the support vector machine classifiers in order to perform the coarse segmentation of rivers. The experimental results have demonstrated that the proposed algorithm has superior performance in terms of accuracy, efficiency and robustness.(2) Affected by various factors, the result of river coarse segmentation is usually composed of multiple isolated discontinuous areas. For this problem, a new river regional correction algorithm based on the improved geodesic active contour (GAC) model is proposed. This algorithm takes the results of the coarse segmentation as the initial curves to capture the desirable shapes of the rivers. The experimental results show that the methods can remove the impact of image mosaic and waves in river region and phony targets, such as ponds and farmlands, and successfully achieved the accurate extraction of the river region.(3) After analyzing the existing bridge detection algorithms, a new bridge recognition algorithm based on orientation information measure, fuzzy c-means clustering and prior knowledge is proposed. The algorithm first enhance the river region with orientation information measure, in order to sharpen the bridge target, which has strong edges and direction character, and inhibit other interference. Second, segment bridge target with fuzzy c-means clustering, and determine its approximate location. At last, remove phony bridges according to the a priori knowledge of remote sensing images. The experimental results show that the method can achieve good results in bridge recognition.(4) Based on the bridge detection algorithm of this thesis, a bridge recognition and location system was designed using Matlab and Google Earth development platform, and through this system verified the feasibility and practicability of the algorithm. The system can automatically detect bridge target within specified area of the Google Earth. The experimental results show that the proposed system is able to overcome a variety of adverse factors, detecte bridge with high accuracy and good generalization ability.