3d Aircraft Target Recognition Using Neural Network

3D object recognition is one of the important parts of computer vision, and the recognition of airplane is one of the applications of 3D object recognition. Today the researchers have made great progress in this field. It plays an important role in chain of command of air traffic to timely and exact identify airplane, which is important in fields of not only military aviation but also civil aviation.In this paper, firstly the development course of 3D object recognition is introduced. And we also compare several methods of 3D object recognition and analyze in-depth all kinds of theories and methods involved. Finally we propose a new method of 3D object recognition, which is to combine moment invariants and neural network. The main work is listed as follows:1) In this part, we study common steps of digital image processing and some algorithm of threshold. Finally the most threshold is chosen to segment airplane image, which to separate objects from backgrounds.2) In the stage of feature selection, we discuss mostly some conception of feature and steps of feature selection etc, which include moment invariants and Fourier descriptors etc that are in common used presently. Image shape is determined by image kurtosis of fourth order moments which have no translation, rotating and scale invariance, so we process airplane image using fourth order moments.3) In the stage of recognition, firstly we compare several methods that are used abroad in near year, and introduce neural network mostly. We analyze the building of models, learn rule and network training of BP algorithm. At same time, several methods are taken to improve BP algorithm, and are used in recognition of 3D aircraft targets.Experiments of two different airplanes have proved that this recognition systemperform swell.