Research On Detection And Recognition Algorithms Of Aircraft Targets In Remote Sensing Images

With the development of science and technology, the aircraft target detection and recognition in remote sensing images has become a hot and important research in the field of computer vision, and has been widely applied in the fields of military and civilian. Aircraft target as an important military target and transport tool, gaining its information is not only the key to win the war, but also very important to manage aircraft in the airport too.In recent years, aircraft target detection and recognition technology in remote sensing images has made some progress at home and abroad, but in fact, because remote sensing images have a large number of interfering factors, such as noise, complex background and illumination and so on, there are serious impact on the recognition technology and there are still many shortcomings of existing algorithms, such as low recognition accuracy and time-consuming of the algorithms, the algorithms are very purposeful and do not have a kind of general algorithm and so on. Therefore, in order to improve the recognition accuracy of the algorithms, reduce the time-consuming, and make the algorithm universal, on the basis of the existing algorithms, this thesis begins to study the research. The research contents and achievements of this paper are as follows:1. This thesis analyzes the key technology of aircraft target detection and recognition in remote sensing image,this includes feature extraction, classification methods and the evaluation performance of the algorithms. In this part, which focuses on analyzing the advantages and disadvantages of shape characteristics and recognition method, besides, this paper defines a new evaluation index based on the rate of detection and recognition.2. Aiming at specific to target and time-consuming of the algorithms, this paper propose aircraft recognition algorithm based on the saliency map and multi-feature combination. The recognition algorithm is a hierarchical approach to detect and identify aircraft images, at first, it roughly locate the candidate aircraft target contained false alarm in the use of saliency map; and uses the region increasing and line marked searching algorithm to find the connected regions to determine the number and location of the candidate targets, then extracts MSA,Pseudo-Zernike moment and Harris-Laplace feature descriptor and uses the ratio of the standard deviation and mean to measure the feature and combines the feature to feature vector. Finally it recognizes the candidate targets using the combined moments and Support Vector Machine(SVM). The experiment result shows that the algorithm can overcome illumination, time-consuming less,has good robustness and is universal.3. Proposing a aircraft detection algorithm based on image entropy and multiple shape feature fusion. Image entropy can generate naturally clumps area, it is different from edge detection, which only has the edge of targets, and it also differs from the saliency map, which can not directly reflect the shape of the aircraft target. Image entropy is suited for coarse locating target aircraft coarsely. This part uses image entropy to gain suspected target area; then extracts affine invariant moments, normalized moment of inertia and singular values; analyzes the stable performance of the three shape feature of the aircraft and weighted fusion into a new feature vector; at last use SVM to finish aircraft target detection and recognition. The experiment result shows the algorithm is effective, can overcome aircraft target detection depending on edge and contour, and gain higher rate of detection and low rate of false alarm.