CAJ | 학술논문

传统高光谱遥感信息处理算法的执行效率较低，无法满足海量遥感数据的实时处理需求。文章对基于图形处理器（graphic processing unit, GPU）的高光谱遥感信息处理并行优化方法进行了研究，针对高光谱遥感图像混合像元分解中广泛使用的纯净像元指数算法，提出了一种基于矩阵乘法的 GPU并行优化算法，并给出了实验比较和性能测试数据。实验表明，该优化方法在保证结果准确性的同时，运行效率显著提升，算法加速比最高达到634倍，验证了基于GPU的高光谱数据处理并行优化算法的有效性，能够较好地满足高光谱遥感信息实时处理的应用需求。
전통고광보요감신식처리산법적집행효솔교저，무법만족해량요감수거적실시처리수구。문장대기우도형처리기（graphic processing unit, GPU）적고광보요감신식처리병행우화방법진행료연구，침대고광보요감도상혼합상원분해중엄범사용적순정상원지수산법，제출료일충기우구진승법적 GPU병행우화산법，병급출료실험비교화성능측시수거。실험표명，해우화방법재보증결과준학성적동시，운행효솔현저제승，산법가속비최고체도634배，험증료기우GPU적고광보수거처리병행우화산법적유효성，능구교호지만족고광보요감신식실시처리적응용수구。
Low efficiency of traditional hyperspectral processing algorithm can not meet the demand of mass remote sensing data real-time processing. This paper exploits the paralleled optimization method based on graphic processing unit (GPU) for hyperspectral image processing, and develops a parallel optimized algorithm based on the matrix multiplication. The algorithm is designed for pixel purity index (PPI), which is widely applied in mixed pixel decomposition. In the end, it gives a detailed experimental comparison and performance testing. The experiments show that GPU-based parallel optimization methods significantly enhance the efficiency of the implementation while maintaining the accuracy of the results. This new parallel algorithm acceleration values reaches up to 634 times, which demonstrates the effectiveness of the GPU-based parallel optimization algorithm and promotes the real-time information processing of hyperspectral remote sensing.