The Research Of Nonlinear Unmixing Of Hyperspectral Data With Sparsity Constraint And Manifold Regularization

The emergence of hyperspectral remote sensing technology has brought tremendous change to the field of remote sensing, it can obtain more precise data, more bands, and provide more detailed and rich feature information for the study of remote sensing technology. Over the last few decades, hyperspectral remote sensing has been playing important role in environmental resource detection, the evaluation of loss in natural disaster, region analysis and construction, biological modeling, weather forecast and military reconnaissance.Mixed pixels are widespread in hyperspectral images due to insufficient spatial resolution and mixing effects of the ground surface, which bring about great difficulties for conventional pixel-level applications. Therefore, spectral unmixing is an essential step for deep exploitation of hyperspectral image. Spectral unmixing which is based on linear mixing model( LMM) is relatively simple, but neglecting the nonlinear scattering effect. The nonlinear mixing models( NLMM) provide an alternative to overcoming the inherent limitations of the LMM.The modified generalized bilinear model( MGBM) which combines the advantages of generalized bilinear model( GBM) and polynomial postnonlinear mixture model( PPNMM) has achieved great performance in spectral unmixing. In this paper, we proposed a nonlinear unmixing method via sparse and manifold regularization based on MGBM model, named RNSM. we employ the joint sparse and manifold regularization to capture the sparseness and the intrinsic manifold structure of hyerspectral data. And L21-norm loss function is adopt because it is more robust for noises. Besides, we provide a simple yet efficient updating algorithm with rigorous convergence analysis for RNSM. Finally, experiments on both synthetic and real hyperspectral data demonstrate that RNSM can get better performance in unmixing problem compared with other algorithms.