GPU Parallel Optimization of Hyperspectral Image Kernel Sparse Representation Classification Based on Spatial-Spectral Graph Regularization

With the development of hyperspectral remote sensing information processing, hyperspectral image classification becomes a hot topic. The algorithm of kernel sparse representation classification based on spatial-spectral graph regularization and sparsity concentration index (SSGSCI-KSRC) gains a good result. Due to the big scale of hyperspectral image data, time-critical requirement in the practical application makes it impossible to use the original SSGSCI-KSRC algorithm. This paper proposes a parallelization method for SSGSCI-KSRC algorithm. The optimization method achieves the efficient calculation operations of hyperspectral image matrix data, coalesces memory accesses to reduce the time of transferring data to the GPU devices, and designs proper kernel functions in the classification algorithm. The experimental results demonstrate that the parallel SSGSCI-KSRC algorithm obtains a better result in terms of computational performance when the accuracy stays the same.

[1]  Nagiza F. Samatova,et al.  Enabling Fast, Noncontiguous GPU Data Movement in Hybrid MPI+GPU Environments , 2012, 2012 IEEE International Conference on Cluster Computing.

[2]  Mahmut T. Kandemir,et al.  Managing GPU Concurrency in Heterogeneous Architectures , 2014, 2014 47th Annual IEEE/ACM International Symposium on Microarchitecture.

[3]  Antonio J. Plaza,et al.  Parallel unmixing of remotely sensed hyperspectral images on commodity graphics processing units , 2011, Concurr. Comput. Pract. Exp..

[4]  Sartaj Sahni,et al.  GPU-to-GPU and Host-to-Host Multipattern String Matching on a GPU , 2013, IEEE Transactions on Computers.

[5]  Qian Du,et al.  Unsupervised Hyperspectral Band Selection Using Graphics Processing Units , 2011, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

[6]  Prakash S. Raghavendra,et al.  A GPU Framework for Sparse Matrix Vector Multiplication , 2014, 2014 IEEE 13th International Symposium on Parallel and Distributed Computing.

[7]  Jaewon Lee,et al.  ScaleGPU: GPU Architecture for Memory-Unaware GPU Programming , 2014, IEEE Computer Architecture Letters.

[8]  Lorenzo Bruzzone,et al.  Kernel-based methods for hyperspectral image classification , 2005, IEEE Transactions on Geoscience and Remote Sensing.

[9]  Julien Michel,et al.  Remote Sensing Processing: From Multicore to GPU , 2011, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

[10]  Liang Xiao,et al.  Hyperspectral Image Classification Using Kernel Sparse Representation and Semilocal Spatial Graph Regularization , 2014, IEEE Geoscience and Remote Sensing Letters.

[11]  Dhabaleswar K. Panda,et al.  Designing efficient small message transfer mechanism for inter-node MPI communication on InfiniBand GPU clusters , 2014, 2014 21st International Conference on High Performance Computing (HiPC).

[12]  Jie Wei,et al.  Parallel optimization of hyperspectral unmixing based on sparsity constrained nonnegative matrix factorization , 2013, 2013 IEEE International Geoscience and Remote Sensing Symposium - IGARSS.

[13]  Holger Fröning,et al.  Energy-Efficient Stencil Computations on Distributed GPUs Using Dynamic Parallelism and GPU-Controlled Communication , 2014, 2014 Energy Efficient Supercomputing Workshop.

[14]  Feng Ji,et al.  RSVM: A Region-based Software Virtual Memory for GPU , 2013, Proceedings of the 22nd International Conference on Parallel Architectures and Compilation Techniques.

[15]  Jaewon Lee,et al.  GPUdmm: A high-performance and memory-oblivious GPU architecture using dynamic memory management , 2014, 2014 IEEE 20th International Symposium on High Performance Computer Architecture (HPCA).

[16]  Enrico Magli,et al.  Highly-Parallel GPU Architecture for Lossy Hyperspectral Image Compression , 2013, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.