Hierarchical Sparse Learning And Collaborative Representation For Hyperspectral Imagery Restoration And Classification

With the development of hyperspectral imaging techniques and the increasing of spatial resolution,hyperspectral imagery(HSI)can provide more accurate and detailed information about the observed objects.For each image pixel,the dedicated spectral information provides the ability to exactly identify and distinguish different materials of interest.Therefore,hyperspectral images have been broadly applied in various fields,such as classification,target detection and recognition.But,hyperspectral images are inevitably polluted by various noises during acquisition process,which greatly influence their visual impression and subsequent applications.Also,the spatial-spectral information becomes more and more important in the HSI analysis.After analyzing the recent development on denoising and classification techniques in HSI analysis,a deeper study on denoising and classification of HSI was made in this thesis by combining the hierarchical sparse learning and collaborative representation.The spatial-spectral information was also effectively exploited to improve the denoising or classification performance.Specifically,the major topics of this thesis include the following four methods:(1)Beta process priors is introduced into hierarchical sparse Bayesian learning for recovering underlying degraded hyperspectral images,including suppressing the various noises and inferring the missing data.The proposed method decomposes the HSI into the weighted summation of the dictionary elements,Gaussian noise term and sparse noise term.With these,the latent information and the noise characteristics of HSI can be well learned and represented.Solved by Gibbs sampler,the underlying dictionary and the noise can be efficiently predicted with no tuning of any parameters.The performance of the proposed method is compared with state-of-the-art ones and validated on two hyperspectral datasets,which are contaminated with the Gaussian noises,impulse noises,stripes and dead pixel lines,or with a large number of data missing uniformly at random.The visual and quantitative results demonstrate the superiority of the proposed method for HSI restoration.(2)A novel Bayesian approach integrating hierarchical sparse learning and spectral-spatial information is proposed for HSI denoising.Based on the structure correlations,spectral bands with similar and continuous features are segmented into the same band-subset.To exploit local similarity,each subset is then divided into overlapping cubic patches.All patches can be regarded as consisting of clean image component,Gaussian noise component and sparse noise component.The first term is depicted by a linear combination of dictionary elements,where Gaussian process with Gamma distribution is applied to impose spatial consistency on dictionary.The last two terms are utilized to fully depict the noise characteristics.Furthermore,the sparseness of the model is adaptively manifested through beta-bernoulli process.Calculated by Gibbs sampler,the proposed model can directly predict the noise and dictionary without priori information of the noisy HSI.The experimental results on both synthetic and real HSI demonstrate that the proposed approach can better suppress the existing various noises and preserve the structure/spectral-spatial information than the compared state-of-art approaches.(3)Collaborative patch learning is introduced into Bayesian low-rank matrix factorization to denoise the HSI(BLRMFwCPL).Based on the spatial consistence and nonlocal self-similarities,the HSI is partitioned into overlapping cubic patches;the similarity metric using fusing features is applied to select the most similar patches with the testing patch and construct the corresponding collaborative patch.Bayesian low-rank matrix factorization model is utilized to learn each collaborative patch.By using the low-rank properties of the collaborative patch in spatial and spectral domain,the noise-free data can be learned;also,the characteristics of noises can be well depicted by the mixture of Gaussians.The widely used hyperspectral data sets with different environmental settings are adopted to validate the proposed methods.The experimental results indicate the effectiveness and superiority of BLRMFwCPL method in HSI denoising.(4)By combining the nonlocal weighted joint collaborative representation(NLWJCR)and extreme learning machine(ELM)in a fusion manner,a novel classifier is constructed for hyperspectral image classification.To fit the spatial unequal contributions between samples,the weighted matrix is firstly calculated by exploiting both correlation and Euclidean distance between each testing sample and its surroundings.By using this weighted matrix,the joint spectral pixels are constructed and the NLWJCR method is applied to classify the joint spectral pixels,in which different samples contribute different roles in the decision boundary construction.Then mid-features are learned by utilizing these coefficients to conduct an ELM model.Finally,the NLWJCR and ELM models are combined in a multiplicative fusion manner,and then the fused probability is used to achieve the final classification.The proposed approach is evaluated with different hyperspectral images by comparing with other popular classifiers.The visual and numerical results indicate the efficacy of the proposed method on HSI classification.