Dropout Concrete Autoencoder for Band Selection on HSI Scenes

• DOI: 10.48550/arxiv.2401.16522
• Publication Date: 2024-01-29
• Authors: Lei Xu; Mete Ahishali; Moncef Gabbouj
• Keywords: Autoencoder; Dropout (neural networks)

Deep learning-based informative band selection methods on hyperspectral images (HSI) recently have gained intense attention to eliminate spectral correlation and redundancies. However, the existing deep either need additional post-processing strategies select descriptive bands or optimize model indirectly, due parameterization inability of discrete variables for procedure. To overcome these limitations, this work proposes a novel end-to-end network selection. The proposed is inspired by advances in concrete autoencoder (CAE) dropout feature ranking strategy. Different from traditional methods, trained directly given required subset eliminating further post-processing. Experimental results four HSI scenes show that CAE achieves substantial effective performance levels outperforming competing methods.