Fuzzy inspired image classification algorithm for hyperspectral data using three-dimensional log-Gabor features

In hyperspectral image classification problems, the discriminative efficiency of the classifier depends on the features. To classify the heterogeneous classes present in hyperspectral imagery, biologically inspired models such as log-Gabor features are useful as they exhibit joint spatial-spectral characteristics of each pixel. Log-Gabor features occupy the state-of-the-art hyperspectral research domain for extracting the features at different scales and orientations. In this proposed work, three-dimensional log-Gabor wavelets with different scales and orientations are designed to obtain the complete spatial, spectral and joint spatial-spectral characteristics of individual pixels in the hyperspectral data. Aiming to improve the accuracy, a simple fuzzy inspired algorithm is also proposed. The performance of the proposed algorithm is evaluated and is compared with other existing methods and supremacy is observed. The proposed methods are experimented on airborne visible infrared imaging sensor (AVIRIS) data of Indian Pine Site. The results witness the accuracy of 92.13% even while only 5% of the samples in each class were used for training for 3D log-Gabor features. Fuzzy inspired 3D log-Gabor features produce the accuracy of 93.11% for 5% training samples.