Parallel Optimization Of Hyperspectral Target Detection Algorithm Based On GPU

Hyperspectral remote sensing is the frontier of remote sensing technology.The use of hyperspectral remote sensing images for target detection is a hot topic in the research and application of hyperspectral remote sensing technology.Hyperspectral target detection has strong practicability and can be applied to many aspects such as public safety,environmental detection,urban planning and food hygiene.However,due to its resolution,it causes huge data volume and complicated algorithm calculation process,which brings inconvenience to application and analysis.In recent years,the development of GPU has provided new ideas for studying the accelerated processing of hyperspectral image algorithms.In this paper,parallel optimization algorithms based on GPU platform is proposed for hyperspectral target detection algorithm.On the basis of ensuring the detection accuracy,the efficiency of the original algorithm is improved.The detailed work is as follows:On the one hand,the parallel optimization methods are proposed for the hyperspectral target detection algorithm based on subspace model(ROVP).The optimization algorithms based on GPU parallel platform and CPU serial platform are respectively completed.The GPU parallel algorithm is divided into two parallel modes based on CUDA architecture and CUBLAS-based library,then the experimental comparison and analysis of the ROVP algorithm.The time performance of ROVP algorithm under GPU parallel platform and CPU serial platform is compared horizontally.The time performance of ROVP algorithm and OSP and OVP algorithm on GPU platform are compared vertically.The experimental results show that the ROVP algorithm for hyperspectral target detection is suitable for parallel computing.In the GPU parallel architecture,the obvious acceleration effect is obtained,which effectively improves the execution efficiency of the algorithm.And the ROVP parallel algorithm works better than the OSP and OVP parallel algorithms.On the other hand,the parallel optimization methods are proposed for the hyperspectral target detection algorithm based on target constraint.The GPU-based single target detection CEM parallel optimization algorithm and the GPU-based multi.target detection LCMV parallel optimization algorithm are proposed respectively.Firstly,two parallel modes based on CUDA architecture and CUBLAS library are designed for two target detection algorithms;Secondly,combining the properties of the two algorithms,the corresponding parallel optimization algorithms are proposed,which are CUDA-based shared memory optimization algorithm and CUDA-based autocorrelation matrix calculation optimization algorithm;Finally,a lot of experiments are carried out on the two algorithms under different hyperspectral images.The experimental results show that the two GPU-based high-performance parallel optimization algorithms all get a higher speedup.The CUDA-based shared memory parallel optimization algorithm has the best acceleration effect.