Research On Near Field Array Imaging Technology Of Millimeter Wave Radar Based On GPU

Millimeter-wave near-field array imaging system can realize detection and imaging of near-field targets with high frame rate and high resolution.It is suitable for security inspection,unmanned aerial vehicles,military reconnaissance and other fields which require high real-time performance.The disadvantage is that the large amount of echo data makes the original imaging method inefficient.In this paper,millimeter-wave near-field array imaging method and hardware graphics processing unit(GPU)parallelization acceleration and other key issues were studied.From the perspective of imaging system,this paper focuses on the planar millimeter wave near-field array imaging system and cylindrical millimeter wave near-field array imaging system.Firstly,the 3-D Range Migration Algorithm(RMA)of planar architecture is deduced.To solve the problem of time-consuming implementation of Python-based algorithm on CPU,a GPU-based method based on decorator in Numba compiler is proposed,which is 4.7013 times faster than that of a single CPU.To solve the time-consuming problem caused by the large amount of interpolation,a parallel linear interpolation algorithm based on the Unified Device Architecture(CUDA)is proposed,which is 1005.6 times faster than that under CPU.The fast Fourier algorithm is accelerated by using cufft library function,and the for loop iteration under CPU is accelerated by using GPU kernel function.The speed is 2.7277 times higher than that of MATLAB under CPU.Secondly,based on the principle of range migration,the fast 3D imaging algorithm in cylindrical system is deduced.To solve the time-consuming problem of the algorithm in CPU,the fast 3D imaging algorithm GPU in cylindrical system is implemented by using CUDA.For the required cumulative operation,we use the multithreading in the GPU,put the data in two threads,and use all streaming multiprocessors to parallel computing speed.Compared with the measured data speed,the MATLAB of CPU increased by 3.0713 times and 2.9891 times separately.