The “curse of dimensionality” phenomenon in machine learning tasks results in poor generalization when dealing with high-dimensional hyperspectral images (HSIs).To solve this issue, Band selection is a frequently used dimension reduction technique for hyperspectral images that identifies and selects a subset of the most crucial bands from the original ones to remove redundancy and noisy bands while maintaining optimal generalization ability.Band selection methods can be categorized as supervised or unsupervised techniques based on whether labels are used.Unsupervised approaches have the potential for extensive applications since they do not require labeled data, which can be difficult to obtain.Several techniques have been proposed for unsupervised band selection, including Autoencoders (AE), which aim to represent datasets from the original data space to a reduced and more informative feature space. In this study, we propose an innovative framework for unsupervised band selection and feature extraction that trains a sub-neural network to identify the most important and informative bands.The classification performance of the selected band combination on the Indian Pines, Pavia University, and Salinas hyperspectral datasets have been verified using machine learning techniques. Our proposed method not justenhances the classification resultsof HSIs, butalso reduce the computational time when compared to other state-of-the-art band selection approaches.
Previous Article in event
Analysis of Mode Reliability Factors among Off-Campus Students Using Structural Equation Modeling in Dhaka CityPrevious Article in session
Next Article in event Next Article in session
Autoencoder Based Unsupervised Hyperspectral Bands Selection for Remote Sensing Land-Cover Classification
Published: 07 November 2023 by MDPI in The 4th International Electronic Conference on Applied Sciences session Computing and Artificial Intelligence
https://doi.org/10.3390/ASEC2023-15879 (registering DOI)
Keywords: Keyword: Hyperspectral Images, unsupervised band selection, Deep Learning, Autoencoders, classification, clustering