Targets Recognition Research In Remote Sensing Images Based On PCNN And Moment Features

Recently,the target recognition technology based on satellite remote sensing images get development rapidly,and has been widely applied in the fields of military and civilian.At present,the domestic and foreign scholars have achieved a breakthrough on aircraft target recognition technology in remote sensing images,but in fact,all of them have a big impact on the subsequent recognition,because remote sensing images have interfering factors,such as noise,complex background and so on.There are still many shortcomings of existing algorithms,such as low recognition accuracy and time-consuming of the algorithms.At the same time,they don't have a kind of general algorithms.Therefore,the emphasis of this paper is that how to recognize aircraft target efficiently in the complex environment.The steps about aircraft target recognition in remote sensing image including preprocessing,remote sensing image segmentation,feature extraction and matching.However,as the main research contents of this paper,the achievements of the middle two steps are as follows:1.In order to improve the accuracy of the target segmentation of remote sensing image,a kind of algorithm of remote sensing image segmentation is proposed in this paper,which is based on parameter optimization of gravitational search algorithm.The first step is to reform the classical PCNN model,we should redescribe the relationship of excitation and inhibition between neurons,and improve the linking item and the dynamic threshold.Then we need use the new PCNN model to ignite the neurons of input image,and extract the ratio of image entropy and energy from the output of PCNN.At the same time,we treat it as the fitness function of gravitational search algorithm and the convergence criteria of gravitational search algorithm is entropy variation.According to the global searching ability of gravitational search alogrithm,we could find out the optimal parameter of PCNN model,which has an influence on segmetation effect.Finally,compared with OTSU method and maximum entropy segmetation method and PCNN segmetation method,the result of matlab simulation proved that the proposed method is much better than the others,it's adapt to the remote sensing image segmetation.2.According to the shortage of invariant moment in deformation object description,in order to improve the recognition accuracy of aircraft type,this paper proposes a new aircraft type identification method.First of all,according to the normalized binary images of aircraft,the features of wavelet moment and affine invariant moment are extracted respectively;Then,the features with good robustness and high stability are selected by calculating the ratio of mean and standard deviation of the sample features,and two different features are fused by normalization method in order to eliminate the differences between two features;Finally,five types of sample set of the aircrafts are constructed through Matlab program,the support vector machine(SVM)is used as classifier to identify the aircraft type of test sample set which consists of the whole sample set except the training set,and the differences among wavelet moment,affine invariant moment and the proposed method are compared in recognition accuracy and the effect of the training sample set size and the parameters of SVM on the identification accuracy.Experimental results show that the proposed algorithm improves the recognition accuracy,and the recognition accuracy is still great when the training sample set is small.3.we complete the whole recognition process based on the above two important steps and SVM.According to a large number of experiments,they proved that the proposed recogintion algorithm not only overcome various kinds and different levels of noise,but also be applied to aircraft target recognition from different background image.Meanwhile,it still can obtain higher recogintion accuracy and less time-consuming.