Hyperspectral Image NRS Classification Algorithm GPU Acceleration Reseaech

The imaging principle of the hyperspectral apparatus determines that the imaging data can accommodate more features and rich information,which brings great convenience to the classification of the objects,but also causes the classification algorithm to run slowly and time consuming.The traditional remote sensing image classification method is difficult to adapt to this change,especially in some areas that need to respond quickly to emergencies such as sudden natural disaster monitoring,real-time warning of hostile targets in wartime.In recent years,with the development of image processing technology,based on Graphic Processing Unit(GPU),the hyperspectral telemetry image processing has been applied gradually.According to the above research,GPU-based hyperspectral image processing has been a number of related applications and processing,but based on GPU architecture,hyperspectral image processing still need to further optimize the content.Therefore,this paper proposes a parallel algorithm of NRS hyperspectral image classification based on GPU for NRS processing.The algorithm takes full advantage of the high-speed processing bandwidth of GPU and optimizes the data transmission between GPU and CPU and GPU core function design And so on,to achieve the completion of hyperspectral image classification processing speed,while the classification accuracy is similar or unchanged.In this paper,the relevant formula of NRS algorithm is deduced in detail,and the algorithm is given by means of flow chart.The advantages of NRS algorithm are given by correlation comparison.NRS algorithm is used as the basis of this paper from the theoretical and practical aspects.After that,combined with the previous section of the NRS classification algorithm flow related to the introduction of the general idea of the algorithm to achieve,as the GPU parallel classification algorithm to complete the NRS classification algorithm to achieve the serial CPU and detailed NRS algorithm CPU implementation of the main completion of the process of decomposition,Matrix solution,set variables and other related content.Finally,based on CUDA's hyperspectral image NRS classification algorithm parallel design,according to the hyperspectral image NRS classification CPU algorithm,complete hyperspectral image NRS classification GPU parallel algorithm design,and complete the relevant experiments.In this paper,the relevant formula of NRS algorithm is deduced in detail,and the algorithm is given by means of flow chart.The advantages of NRS algorithm are given by correlation comparison.NRS algorithm is used as the basis of this paper from the theoretical and practical aspects.Experiments show that the GPU-based parallel optimizer completes all operations for 6378.96s(1.772hours),which is much faster than the 169376.48s(5.268hours)of the CULA and the 139976.85s(38.88hours)of the CPU,Reached a 21.94 times the speedup.The relevant research results of this paper can be applied to some areas that need to make quick response,such as sudden natural disaster monitoring and real-time warning of hostile targets in wartime.