Design And Realization Of Automatic Target Recognition System Based On Optical Remote Sensing Images

A lot of image data are supplied by the reconnaissance equipments such assatellites， high-altitude scouting plane， and UAV(Unmanned Aircraft Vehicle) etc.，which contain a lot of information of the military targets. As a kind of importantmilitary target， aircraft is one of main targets that are scouted and attacked in thebattlefields. It is key point to find out the aircraft targets and recognize themautomatically， correctly and quickly from satellite or aviation image database of hugeamount that the accurate attacks can be carried out successfully. If the recognizingalgorithms are used to recognize the aircraft targets from the image data of hugeamount， a lot of time and system resource will be wasted， and the timeliness ofinformation processing will be influenced. Considering that the aircraft usually stays atthe airport and as one of the important image categories the airport images have specialfeatures， in the thesis， firstly， the airport images are searched from the imagedatabase; then the aircraft targets are recognized from the airport images; at last， aretrieval and recognizing system for aircraft targets staying at the airport in the databaseof satellite images is designed.The main work of the thesis includes the following:①On the basis of the systematic studying the CBIR， a retrieval method forairport images based on the SVM classification is presented. The method realizes theretrieval for the airport images by means of extracting the images’ general features ofcolor， texture and shape， and combining with the SVM classifier. The experimentresult shows that the method raises the effectiveness of airport image retrieval from thesatellite image database to a certain degree.②In the light of the weak points of the traditional algorithms of target segment，a segment algorithm for aircraft targets based on the extracting of airport area is putforward. By means of using the Hough transform to locate the runway of airport， thealgorithm realizes to extract the runway of airport and parking apron accurately andcomplete segmenting aircraft targets staying at the airport. The experiment result showsthat the algorithm improves the weak points of the algorithms of airport extractionbefore that the subsidiary parts such as terminal buildings are retained and the extractingresults contain areas out of the airport area.③A method of recognizing the aircraft targets staying at the airport based on the SVM classification is presented. The method comprehensively utilizes fuzzy affineinvariant moments and the method of the SVM multi-classification to realize theclassifying and recognizing of the aircraft targets. The experiment result shows that themethod has higher recognition rate than the Bayes classifier and the BP neural net.④A system of retrieval and recognizing for aircraft targets staying at the airport inthe satellite image database is designed. The system completes all the stages of theretrieval and recognizing for aircraft targets staying at the airport in the satellite imagedatabase， including image input， image pre-processing， airport image retrieval，aircraft target segment， feature extracting and aircraft target classifying andrecognizing etc.The research work in this thesis provides reference frame for the further research onthe topic of military target recognizing from image database.