Hyperspectral Image Unmixing Based On DSP And Backtracking Search

The development of remote sensing technology promotes the development of hyperspectral image in the fields of forestry and farming,marine monitoring etc.However,due to the complexity of the surface space environment and the limited spatial resolution,hyperspectral images contain mixed pixels.The presence of mixed pixels affects the accuracy of surface recognition and the development of quantitative remote sensing.Therefore,it is an urgent problem to unmix hyperspectral images,extract endmember and obtain its abundance.The paper studys algorithm of blind source separation and hyperspectral image unmixing theory based on linear spectral mixture model,concludes that the problem of the hyperspectral images unmixing can learn from the blind source separation technology,and builds hyperspectral image unmixing model based on blind source separation.The unmixing model transforms the problem of unmixing into a numerical optimization problem,and uses correlation optimization algorithm to optimize the objective function in the unmixing model.In the existing optimization algorithm,the traditional gradient method is easy to fall into local extreme value,but swarm intelligence optimization algorithm has good robustness and high convergence performance.Therefore,the paper proposes hyperspectral image unmixing algorithm based on backtracking search,and the experimental results show that the algorithm has better unmixing effect.Digital Signal Processing(DSP)which is known for its high performance,real-time performance and easy programming,provides an ideal hardware platform for the theory to practical engineering application.Therefore,it is research and practical value for that the hyperspectral image unmixing algorithm achieves on the DSP platform.Aiming at the DSP chip C6748,which is producted by TI company,this paper deeply studies the DSP implementation scheme of hyperspectral image unmixing algorithm based on backtracking search.Compared with the development of simulation environment,the development of DSP platform is the development difficulty of the two dimensions of language and development process,together with high computational complexity of the hyperspectral image unmixing algorithm,therefore this paper firstly studies the realization of typical blind source separation algorithm-FastICA algorithm in DSP.According to the FastICA algorithm DSP implementation process,through skilled development of the DSP platform to further improve the DSP algorithm,which is based on backtracking search hyperspectral image unmixing algorithm.Finally,combined with the actual hardware resource of C6748 and hyperspectral image unmixing algorithm based on backtracking search,source files and CMD files are compiled,algorithm transplantation is completed,the algorithm is debugged and optimized and finaly hyperspectral image unmixing can be realized on the TMS320C6748 platform.The experimental results show that the efficiency of the algorithm is improved by optimizing the DSP algorithm in both language and structure.