Gabor Filter-Based Multi-Scale Dense Network Hyperspectral Remote Sensing Image Classification Technique

Abstract

Since hyperspectral remote sensing images are three-dimensional data cubes with spatial and spectral information, with many wavebands and high inter-band correlation, the number of training samples required for classification is greatly increased. In order to achieve better classification of hyperspectral remote sensing images with small samples, this paper proposes a hyperspectral remote sensing image classification method based on Multi-Scale Dense Network (MSDN) with 3D Gabor filter. The method extracts the texture features of hyperspectral remote sensing images by using 3D Gabor filter; then extracts the spatial spectral features of hyperspectral remote sensing images at different scales in both horizontal and vertical directions by using Multiscale Dense Network; and finally achieves the classification of hyperspectral remote sensing images by using Softmax classifier. The introduction of 3D Gabor filter in this method can improve the extraction effect of the features of hyperspectral remote sensing images, and at the same time reduce the dependence of the multiscale dense network on the labeled samples in the classification of hyperspectral remote sensing images. Experiments are conducted on three publicly available hyperspectral remote sensing datasets, and the experimental results are compared with other classification methods to prove that the method has better classification performance.INDEX TERMS hyperspectral remote sensing image classification, multiscale dense network, small sample, three-dimensional.