Parallel Acceleration of SAM Algorithm and Performance Analysis

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing Vol. 6; no. 3; pp. 1172 - 1178
• Main Authors: Qu, Haicheng, Zhang, Junping, Lin, Zhouhan, Chen, Hao
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.06.2013; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Acceleration; Algorithms; Architecture (computers); Central processing units; Computation; Computer architecture; Data transmission; Devices; distributed computing; GPU; Graphics processing units; High-performance computing; Instruction sets; MATLAB; Remote sensing; SAM; Studies; Training
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2013.2239261

Advances in sensor and computer technology are revolutionizing the way that remote sensing data with hundreds or even thousands of channels for the same area on the surface of the earth is collected, managed and analyzed. In this paper, the classical Spectral Angle Mapper (SAM) algorithm, which is fit for parallel and distributed computing, is implemented by using Graphic Processing Units (GPU) and distributed cluster respectively to accelerate the computations. A quantitative performance comparison between Compute Unified Device Architecture (CUDA) and MATLAB platform is given by analyzing result of different parallel architectures' implementation of the same SAM algorithm. Especially for the property of GPU, this paper studied the balance between resource acquirement of each thread and the number of active blocks, and the impact of computational complexity on speedup. In addition, page-locked memory and stream are also introduced to make CPU and GPU work collaboratively. Moreover, we improved the SAM algorithm, in which several training samples are instead of a single one. Experimental results on hyperspectral data have shown that recognition result of the improved SAM algorithm is better than that only using single spectrum. On the other hand, the GPU parallel implementation achieves a higher speedup comparing with the multithread CPU counterpart. And the asynchronous transfer function of CUDA covers the data transmission latency effectively, thus improves the devices' resource occupancy significantly.