Based On The Ls - Svm Radar Target Recognition Algorithm Of The Fpga Design

Radar target identification (RTI) technology which is widely used in military andcivilian life is one of the important development direction of modern radar systems.Compared to the other RTI technology, radar range profile is more readily available andeasily implemented on FPGA. Recently, with the rapid development of programmablelogic devices, high-density, high bandwidth, high performance programmable logicdevices provide a good platform for radar target recognition. The newly improvementand innovation of the target recognition algorithm provides a variety of feasiblesolutions to the implementation of radar target recognition technology based on FPGA.In this thesis, the algorithm based on radar range profile is simulated and verifiedon StratixIII device family. The algorithm we chose is called least square support vectormachine（LS-SVM）which is based on statistical learning theory. The main content is asfollows:1. An introduce of the theories related to the design including the LS-SVMalgorithm, the algorithm of support vector machines (SVM) and the algorithm ofunclassifiable sample sets condition(C-SVM).2. With the concept of Renyi entropy, the samples are clustered by a new kind ofclustering algorithm and trained to be Standard Template Library (STL).3. Radar data used in this thesis is applied to the IEEE754single precisionfloating point standard to meet the requirements of the real-time and accuracy of RTI.,and completed the design of the single precision floating point multiplier, adder, divider,and square root module.4. Using the multipliers, adders designed in this thesis and the memories calledfrom FPGA core, we realized a256-point FFT with very high precision. And with theFFT module and vector norm module we implemented the pretreatment process of radarrange profile.5. According to the LS-SVM algorithm, a classifier is implemented in this thesis.The tests show that the recognition rate can achieve89.49%. And when the operatingfrequency set on100MHz, completing the target indentify arithmetic needs64.2s.