A two stage method for hyperspectral image classification

Hyperspectral imagery typically provides a wealth of information captured in a wide range of the electromagnetic spectrum for each pixel in the image; however, when used in statistical pattern-classification tasks, the resulting high-dimensional feature spaces often tend to result in ill-conditioned formulations. Popular dimensionality selection techniques such as filtering and wrapper methods fail to select features automatically and how to determine the number of the selected features is still open. On the other hand, although embedding methods such as L₀ - SVM or L₁ - SVM have the advantage that they include the interaction with the classification model and being far less computationally intensive, they can only be used for solving linear classification problems. To solve this problem, this paper proposes an novel two stage method for hyperspectral image classification: firstly the L₁ - SVM is used for feature selection and then L₂ - SVM is used for final classification based on selected features. So the performance of hyperspectral imagery could be improved because more information of the data is used. The experimental results on two real datasets prove the performance of the proposed method.