Parallel probabilistic relaxation labelling based on Markov random fields for spectral-spatial hyperspectral image classification

摘要: The large volume of data and computational complexity algorithms limit the application hyperspectral image classification to real-time operations. This work addresses use different parallel processing techniques speed up Markov random field MRF-based method perform spectral-spatial imagery. Metropolis relaxation labelling approach is modified take advantage multi-core central units CPUs adapt it massively systems like graphics GPUs. experiments on sets revealed that implementation has a huge impact execution time algorithm. results demonstrated MRF algorithm produced accuracy similar conventional methods with greatly improved performance. With modern CPUs, good speed-up can be achieved even without additional hardware support. CPU-GPU hybrid framework rendered otherwise computationally expensive suitable for time-constrained applications.