Spatial classification method for hyperspectral camouflage targets image based on local Gabor binary patterns

A spatial classification method based on local Gabor binary patterns (LGBP) is proposed to improve the accuracy of hyperspectral camouflage targets image classification. The LGBP, a multiple scale algorithm, is employed to extract both local and global texture features of hyperspectral image (HSI). The extracted texture features have properties of gray scale invariance and rotation invariance. Each pixel is characterized by both spectral and spatial features. In this way, diversity of inter-class is enhanced. A multi-kernel support vector machine (SVM) is employed as the classifier to integrate spectral and spatial information for classification. Experiments are conducted to demonstrate the efficiency of the proposed method. The overall accuracy and Kappa coefficient of the classification reach 95.6% and 0.937 respectively. The proposed method is helpful to improve the accuracy and robustness of hyperspectral image classification.