Improving Autoencoder Training Performance for Hyperspectral Unmixing with Network Reinitialisation

TitleImproving Autoencoder Training Performance for Hyperspectral Unmixing with Network Reinitialisation
Publication TypeConference Proceedings
Year of Publication2022
AuthorsKsiążek K, Głomb P, Romaszewski M, Cholewa M, Grabowski B, Buza K
Conference Name21st International Conference on Image Analysis and Processing
Volume13231
Date Published05/2022
PublisherSpringer, Cham
Conference LocationLecce, Italy
ISBN978-3-031-06427-2
KeywordsAutoencoders, Hyperspectral unmixing, Network reinitialisation, Training stability
Abstract

Neural networks, in particular autoencoders, are one of the most promising solutions for unmixing hyperspectral data, i.e. reconstructing the spectra of observed substances (endmembers) and their relative mixing fractions (abundances), which is needed for effective hyperspectral analysis and classification. However, as we show in this paper, the training of autoencoders for unmixing is highly dependent on weights initialisation; some sets of weights lead to degenerate or low-performance solutions, introducing negative bias in the expected performance. In this work, we experimentally investigate autoencoders stability as well as network reinitialisation methods based on coefficients of neurons’ dead activations. We demonstrate that the proposed techniques have a positive effect on autoencoder training in terms of reconstruction, abundances and endmembers errors.

URLhttps://arxiv.org/abs/2109.13748
DOI10.1007/978-3-031-06427-2_33