On the basis of the GPU hardware structure and CUDA,a parallel implementation of the Bayes algorithm was given.By analyzing the computational complexity of Bayes algorithm,a mapping model was established between the parallel computing units and the image pixels,where the CPU focuses on procedure control and the GPU focuses on data level parallel computing.It explored the optimization in data transmission and kernel design efficiency for further performance improvement. Compared with the original sequential algorithm,the parallel implementation of the Bayes algorithm provides 25~54 fold speedups.%本文提出了基于 GPU 的高光谱图像贝叶斯并行技术优化算法,通过对高光谱图像分类流程计算复杂度的分析,基于 GPU 的硬件特性和 CUDA 编程模型将待分类图像像元映射到计算线程,GPU 控制流程逻辑, GPU 执行数据级并行计算,并从数据传输和核函数设计两方面进行了优化设计。实验结果表明,该并行分类算法在保证分类精度的同时能大大提高算法的计算效率,获得25倍~54倍的计算加速比。
展开▼