Supervised Band Selection with a Concrete Layer for Hyperspectral Imagery in Remote Sensing and Autonomous Driving
Keywords: hyperspectral imagery, band selection, Gumbel-Softmax, concrete layer, remote sensing, autonomous driving, deep learning, semantic segmentation, plug-and-play models
TL;DR: We present a novel supervised band selection method for hyperspectral imagery that utilizes a concrete layer with the Gumbel-Softmax trick, achieving superior performance on four datasets in remote sensing and autonomous driving applications
Abstract: Hyperspectral imagery captures rich spectral information, which is valuable for a wide range of applications but poses challenges due to high data dimensionality. Current band selection methods are often computationally intensive, non-embedded, or lack adaptability for specific tasks. We address this gap by introducing a novel plug-and-play embedded method for supervised band selection in hyperspectral imagery, utilizing a concrete selector layer based on the Gumbel-Softmax re-parameterization trick. Our approach allows for dynamic and task-specific selection of optimal bands, eliminating the need for pre-processing and enabling seamless integration with downstream models. We evaluated the method on four hyperspectral datasets, covering three remote sensing benchmarks and an autonomous driving task, demonstrating consistent improvements over state-of-the-art methods. This is the first work to perform comprehensive band-selection research on an autonomous driving dataset of this type, and the first to employ a concrete layer for supervised band selection. Our findings highlight the potential of this approach for real-time hyperspectral analysis in applications such as autonomous driving and environmental monitoring, laying the groundwork for further exploration of efficient, domain-specific band selection.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9744
Loading