Compact Representations of Hyperspectral Images Based on the Fusion of Neural Network and Spectral Features

In this paper, we propose a method for generating compact representations of hyperspectral images. The method is based on the distance-based fusion procedure, which merges spatial features extracted using convolutional neural networks and spectral features, followed by dimensionality reduction. The paper shows that the compact representations formed in this way allow solving the classification problem with a quality comparable to modern hyperspectral image classification techniques. The study was carried out using the Wuhan hyperspectral image dataset.