Study Of Feature Extraction And Recognition Method Of Space Radar Target

This paper is combined with the project of defence technology and the background of detection and recognition of space target using milliwave radar, a systematic study is performed on the method of space target recognition.In chapter 1, the background and significance of space target recognition are set forth. Then the current status of the study and the technical means of space target recognition are expounded. At the end of this part, the main work of this paper is introduced.In chapter 2, the property of space target is analyzed in detail. Firstly, the property of orbit and the property of dynamics of space target are clarified. Secondly, the property of electromagnetic scattering and the property of RCS of space target are discussed in detail. Finally, the calculational method of RCS of complex target and the property of 1D range profile of space target are set forth.In chapter 3, space target recognition of low resolution radar(LRR) based on RCS sequence is studied. Space target recognition algorithm of LRR based on RCS sequence is put forward. Firstly, discrete wavelet transform is performed on RCS sequence of LRR of space target, and consequently local property of signal is manifested in time-scale plane. Secondly, ten effective statistical features are extracted from time-scale plane. Finally, space targets are classified through classification method of fuzzy logic. Simulation experiment using real data of four kinds of space target is performed, and gains good recognition effect.In chapter 4, space target recognition of high resolution radar(HRR) based on 1D range profile is studied. Space target recognition algorithm based on local surrounding-line integral bispectrum is put forward. In this algorithm, the feature of local surrounding-line integral bispectrum is extracted from bispectrum of 1D range profile of space target firstly, this kind of feature is shift-invariant and retains scale information and partial phase information of signal,and it can avoid losing important bispectrum and using bad bispectrum,it can restrain Gaussian noise effectively. Then BP neural network is used to classify targets,in order that network weights aren't local extremum, initial value of network weight is obtained through genetic algorithm. Space target recognition algorithm based on sub-block integral bispectrum is put forward. Firstly, bispectrum plane of range profile of space target is divided into many sub-blocks, and the feature of sub-block integral bispectrum is extracted. This kind of feature is shift-invariant, it can restrain Gaussian noise effectively, and redundant information of bispectrum can be reduced. Secondly, KL transform is used to reduce feature numbers. Finally, the method of template match is used to classify targets. Simulation experiment of the two algorithms above gains good recognition effect ,even though the SNR is very low.Space target recognition algorithm based on the sequence of range profile is put forward. Firstly, denoise of the sequence of range profile is performed using wavelet transform. Secondly, location and amplitude of strong scatter centers of target are extracted using RELAX algorithm, then radial length is extracted. Finally, three-axis stabilization satellite,spin stabilization satellite and debris are recognized according to mean and standard deviation of the sequence of radial length. Simulation experiment of this algorithm gains good recognition effect.In chapter 5, space target recognition based on high resolution monopulse radar(HRMR) is studied. Space target recognition algorithm based on the feature of volume of 3D image is put forward. Firstly, 3D image of space target is obtained using HRMW. Secondly, the feature of volume of 3D image is extracted. Finally, satellite and debris are recognized through the feature of volume. Space target recognition algorithm based on the feature of outline area is put forward. Firstly, 2D image of azimuth-elevation is obtained using HRMW. Secondly, the feature of outline area of 2D image of azimuth-elevation is extracted. Finally, satellite and debris are recognized through the feature of outline area. Simulation experiment of the two algorithms above gains good recognition effect.In conclusion, all the work in this paper is summarized, the defects are analized, and the next work to be researched is set forth.