Multi-Feature Learning For Hyperspectral Remote Sensing Image Classification

Hyperspectral imaging(HSI)offers new opportunities for accurate interpretation of the image in the remote sensing community through its improved discrimination in the spectral domain.However,such advanced image processing also brings new challenges due to the high data dimensionality in both the spatial and spectral domains.The rich spectral information provided by HSI brings the challenge of the Hughes phenomenon to the classification task.In addition,using the spectral feature alone can't effectively distinguish different objects whose spectral curve are very similar.It is also easy to divide the pixels which belong to the same kind of objects but have different spectral properties into different categories.Therefore,it is necessary to introduce the spatial features such as texture and morphology profiles to assist the classification of hyperspectral images.But how to effectively integrate these features is still a difficult problem,which requires us to find an effective fusion method to make full use of the useful information provided by various features.To alleviate the problems caused by Hughes phenomenon and effectively utilize the existing features in multiple domains,in this paper,we present a novel multidomain subspace(MDS)feature representation and classification method for hyperspectral images.The proposed method is based on a patch alignment framework.In order to optimally combine the feature representations from the various domains and simultaneously enhance the subspace discriminability,we incorporate the supervised label information into each domain and further generalize the framework to a multidomain version.MDS has the ability of efficiently combining multiple informative features into a latent discriminate subspace,thereby benefiting the subsequent HSI classification.Furthermore,we develop an iterative approach to alternately optimize the MDS objective function by considering it as multiple subconvex optimizations.The classification performance on three standard hyperspectral remote sensing images confirms the superiority of the proposed MDS algorithm over the state-of-the-art subspace learning methods.