Hyperspectral Image Classification Based On Sparse And Low-Rank Representation Theory

As a frontier field in the remote sensing area,hyperspectral remote sensing achieves a significant increase in spectral resolution while maintaining high spatial resolution,so that it is able to recognize more ground features.Nowadays,hyperspectral remote sensing has been widely used for geology,agriculture,military,medical care,water resources management,etc.However,the high-dimensional data of hyperspectral image also brings new challenges,and it's urgent to develop fast and accurate methods to exploit the required information from a large amount of hyperspectral data and efficiently classify the ground objects.In this thesis,by considering the intrinsic characteristics of hyperspectral image,classification algorithms based on sparse and low-rank representation theory are studied.And the main research work includes the following two aspects:Inspired by the sparse and low-rank representation model,this thesis presents a hyperspectral image classification method based on spectral consistency constraint via kernel sparse and low-rank representation.First,the sparse and low-rank constraints in the proposed method can simultaneously capture the local and global structural information of the hyperspectral image.Secondly,by introducing the spectral consistency constraint,the proposed method advances both the sparse term and the low-rank term,and makes full use of the strong correlation between samples.Finally,taking the linear inseparability of hyperspectral data into account,kernel method is used to map samples into a high-dimensional space to make them linearly seperable.In addition,since neighboring pixels of the hyperspectral image usually are strongly correlated with each other and probably belong to the same material,the contextual information of the hyperspectral image is incorporated into the model to spatially extend the proposed method.Experimental results show that the proposed method has better performance than some state-of-the art classification methods.To address the problem of redundant information and large computation complexity in hyperspectral image,this thesis introduces the idea of dimensionality reduction and proposes a hyperspectral image classification method based on sparse and low-rank embedding.The algorithm first establishes a sparse and low-rank embedding model by combining the sparse and low-rank representation which cannot reduce the dimensionality directly with subspace learning.Then,the projection matrix obtained by the constructed model is used to reduce the dimensionality of the original hyperspectral data to obtain robust low-dimensional features.Finally,the low-dimensional features are utilized for classification and the proposed algorithm is verified by experiments on real hyperspectral images.