Fast Recognition Of Bridge And Dam In Complex Background Through Masses Of Data

This thesis is mainly about research on the bridge and dam recognition, which helps on geological disaster monitoring in the civil field. In the military it can quickly find temporary bridges, assist in accurate guidance and hitting effect evaluation. The research on how to recognize these targets automatically in remote sensing image and aerial image is of great practical significance. This thesis is based on the image features of the bridges, dams and rivers, to carry out research on recognition and feature extraction on optical remote sensing image in the bridges dams.First of all, this thesis realizes a preprocessing for a large scale and high resolution image. The image with large scale and high resolution image needs long time to process, while the target will be too small in the image. This increases the difficulty of detection. To resolve this problem, this thesis proposes a method to reduce the amount of follow-up operation in image preprocessing stage, with the detection of gray distribution in each sub regions of the image, and selects ones which have river.Secondly, this thesis proposes an improved segmentation algorithm for water area in images. Remote sensing image information is complex and difficult to completely and accurately segment the target, which is the difficulty of the entire recognition process. To solve this problem, based on the research of remote sensing image gray distribution feature, this thesis propose that the statistics histogram conforms to the Gauss mixture model, and a segmentation method is proposed for hypothesis testing and parameter estimation based on the threshold. Based on research probability of boundary algorithm, this thesis proposes the improvement of corner detection by change the geometrical structure of the operator, puts forward by mean distance function to calculate the contour strength more reasonably, and proposes weighted histogram to improve the detection of uneven brightness images. This several improvements can not only be applied in remote sensing image, which is conducive to the target recognition, and can also be used as a general method to improve the ordinary contour detection in visible light image.Base on the precise segmentation for water area, this thesis realizes detection for target’s contour. Hough transform to the whole image will get high false alarm rate, and the calculation amount will be large. This thesis proposes a fast location algorithm using the geometric and connectivity characteristics. Aim at the problem that the remote sensing images have complex background and exist a large number of similar targets, this thesis proposes using the relationship between the targets and the bank lines to detect the influence of similar targets with river bank lines and remove the false target. Experiments show that the detection accuracy and processing speed are improved.Based on the recognition of the target, this thesis further studies the damage analysis for bridges and dams, and proposes a standard which can measure the assessment capability of the bridge more suitably. Compared to the common standard using damage area proportion, the proposed standards have more practical significance.