Research On Airplane Target Detection Algorithms In Optical Remote Sensing Imagery

With the development of remote sensing technology, Remote sensing images are paid widespread attention and research on account of the wide coverage, large amount of information and short observation period, the target detection in remote sensing images as an important research direction of remote sensing image analysis, it has vital significance on resource survey, disaster monitoring, resource exploration, military target recognition and so on. Because of the great amount of remote sensing data, depending on traditional artificial interpretation to extract specific targets information is difficult to adapt to the development trend of remote sensing technology. How to automatically extract the required information fastly and accurately from these remote sensing data has became the important and difficult points nowadays. This thesis focuses on target detection algorithms in optical remote sensing images, then how to improve the efficiency of target detection algorithms and the prediction accuracy of target center were studied.The target detection algorithms of remote sensing image were summarized in this thesis, and the basic principles of these algorithms and the detection processes were introduced, then the simulation experiments of these algorithms were made and the classical target detection algorithms of each class were analyzed. Firstly, the importance of the feature extraction to target detection is described, Then several commonly used features were reviewed and aircraft target detection algorithms based on these characteristics were described. Then, through analyzes of simulation experiments, the existing problems in the existing methods had been summarized.Secondly, owing to the saliency map lack of objective evaluation criteria in the application of remote sensing images, it was introduced to the application of remote sensing images in this thesis that the quantitative evaluation standard of saliency map. Three kinds of saliency map algorithms were simulated and analyzed, combined with remote sensing images and the characters of the airplane target, the performance of saliency map was analyzed from the aspects of subjective and objective evaluation. In comprehensive consideration of the accuracy and integrity, Itti algorithm is superior to GBVS and SR algorithm to highlights the small airplane target in remote sensing image. Then the saliency map algorithm and template matching algorithm were combined to simulate, the conclusion was verified that the target detection algorithm based on saliency map can discard more than seventy percent of the background area, by this means, the searching space is reduced and the efficiency of detection algorithm is improved.Finally, as to the target detection algorithm based on random forest, the average weight in hough voting phase was improved, the average weight of original algorithm based on leaf node information of random forest was replaced by index distribution weights based on the dictionary of target features. In the training phase, the feature vectors of target pixels were stored as dictionary, as to the predicted target pixels by random forest in test phase, the euclidean distance between the features of predicted pixels and the features of target pixels stored in the dictionary was calculated, the corresponding weights satisfied with exponential function distribution were calculated according to the distance, then voting for the target center by the offset vectors stored in the dictionary, characteristics of the hough image about two algorithms were analyzed through the simulation experiments, it was verified that the improved algorithm could improve the performance of target detection.