Dimensionality Reduction for the Feature System in Classification of Hyperspectral Earth Remote Sensing Data by Use of Neural Networks

The hyperspectral method of analyzing the Earth's surface is very effective in solving classification problems as applied to both objects located on it and the state of these objects (e.g., agricultural crops). However, full-scale hyperspectral analysis is a very expensive procedure and looking for ways of cheapening it are quite explicable. The most consistent way is to reduce the number of spectral components (classification features) by selection (or formation from them) of the most informative ones. It is proposed to implement it by using neural network technologies. By an example of processing a 200-channel hyperspectral image, it is shown that the high accuracy achieved in classification due to reducing the dimensionality of the feature space by these technologies exceeds the accuracy obtained by other known methods.