Research On Spatial-spectral Information Incorporated Classification For Hyperspectral Imagery

Owing to the rapid and steady development of remote sensing technology,hyperspectral imagery can achieve more accurate classification information than ordinary imagery,and its image classification technology has become an important application in the field of remote sensing research.However,hyperspectral data has a high dimensionality and the small-size labeled training set,which brings great difficulties to hyperspectral image processing and data analysis.At the same time,the traditional hyperspectral image classification methods often use spectral information only but ignores the important effect of spatial information.And how to select valuable sample information and accurately predict its feature categories in a large number of unlabeled sample sets.In view of the above problems,this paper makes a thorough study of semi-supervised classification algorithm and the auxiliary role of spatial information on the basis of previous studies,and proposes two new semi-supervised learning algorithms.The details are as follows:1.A novel semi-supervised learning algorithm based on active learning and label propagation method is proposed that addressing the above problems.The proposed algorithm first reduces dimensionality to obtain spectral information by principal component analysis,and then Gabor filtering method is used to extract spatial texture information.The spatial-spectral feature will be cascaded,and then BT(Breaking Ties,BT)technique is used to select the most informative samples.After that,the categories of the selected unlabeled samples are predicted by the Label Propagation(LP)algorithm.The obtained samples are added to training set to initialize SVM classifier.The experiments on two classical hyperspectral datasets show that the proposed method can make full use of unlabeled samples and improve classification accuracy under very few labeled samples.2.Semi-supervised classification of hyperspectral images based on spatial information and particle swarm optimization is proposed.This method makes full use of the spectral and spatial information of the ground objects to improve the classification effect of hyperspectral images in three aspects.Firstly,the spatial texture information image is obtained by Gabor filtering of the three principal components after PCA dimensionality reduction and combined with the first principal component spectral information image.Then,the labeled training sample set is further expanded by using particle swarm optimization(PSO)algorithm and spatial neighborhood sample information.Finally,the spatial-spectral features are combined kernel function.The spatial-spectral kernel is used for support vector machine classification.The experimental results show that the algorithm makes full use of spatial information to improve the classification effect of classifiers from multiple angles.