The Application of Autoencoders for Hyperspectral Data Compression
Alexander S. Minkin · 2021 International Conference on Information Technology and Nanotechnology (ITNT) · 2021
In this paper, a neural networks based algorithm for hyperspectral data compression is proposed. Three different types of autoencoders along with principal component method were implemented and compared. Kennedy Space Center dataset is used for evaluation of data compression algorithms and similarity metrics between source and reconstructed signals. For low value of the compression ratio, the training error of neural networks tends to oscillate and the use of the principal component method shows better metric score. For high value of the compression ratio, the multilayer and convolutional autoencoders demonstrate a significant advantage of compression efficiency over the principal component method.