The incredible increase in the volume of remote sensing data has made the concept of Remote Sensing as Big Data reality with recent technological developments. Remote sensing image processing is characterized with features of massive data processing and intensive computation, which makes the processes difficult. To optimize the remote sensing image processing for GPU, compute unified device architecture (CUDA) is widely used to implement remote sensing algorithms. However, the usage of GPU in remote sensing image processing has been constrained by the complexity of its implementation and configuration. Therefore, how to take full advantage of the parallel organization of GPU architecture is awfully challenging. In this paper, a dynamic adaptive acceleration (DAA) method is proposed to determine calculation parameters of GPU adaptively and preprocess the input remote sensing images on host dynamically. By this method, we determine calculation parameters according to the hardware parameters of GPU firstly. And then, the input remote sensing images are reconstructed based on the calculation parameters. Finally, the preprocessed image blocks are arranged to stream tasks and executed on GPU respectively. The effectiveness of the proposed DAA method in accelerating remote sensing algorithm with point operations was verified by experiments in this paper, and the experimental results indicated that the DAA method can obtain better performance than traditional methods.

随着最近的技术发展，遥感数据量的惊人增长使遥感作为大数据的概念成为现实。遥感图像处理具有数据量大、计算量大的特点，处理难度大。为了优化 GPU 的遥感图像处理，计算统一设备架构 (CUDA) 被广泛用于实现遥感算法。然而，GPU在遥感图像处理中的使用受到其实现和配置的复杂性的限制。因此，如何充分利用GPU架构的并行组织是极具挑战性的。在本文中，提出了一种动态自适应加速(DAA)方法来自适应地确定GPU的计算参数并在主机上动态地对输入的遥感图像进行预处理。通过这种方法，我们首先根据GPU的硬件参数确定计算参数。然后，根据计算参数重建输入的遥感图像。最后，预处理后的图像块被安排为流式任务并分别在 GPU 上执行。本文通过实验验证了所提出的DAA方法在点操作加速遥感算法中的有效性，实验结果表明DAA方法可以获得比传统方法更好的性能。我们首先根据GPU的硬件参数确定计算参数。然后，根据计算参数重建输入的遥感图像。最后，预处理后的图像块被安排为流式任务并分别在 GPU 上执行。本文通过实验验证了所提出的DAA方法在点操作加速遥感算法中的有效性，实验结果表明DAA方法可以获得比传统方法更好的性能。我们首先根据GPU的硬件参数确定计算参数。然后，根据计算参数重建输入的遥感图像。最后，预处理后的图像块被安排为流式任务并分别在 GPU 上执行。本文通过实验验证了所提出的DAA方法在点操作加速遥感算法中的有效性，实验结果表明DAA方法可以获得比传统方法更好的性能。