Abstract
Spectral data provide material-specific information across a broad electromagnetic wavelength range by acquiring numerous spectral bands. However, acquiring such a significant volume of data introduces challenges such as data redundancy, long acquisition times, and substantial storage capacity. To address these challenges, band selection is introduced as a strategy that focuses on only using the most significant bands to preserve spectral information for a specific task. State-of-the-art methods focus on searching for the most significant bands from previously acquired data, regardless of the optical system and the classification model. Nevertheless, some deep-learning methods, such as end-to-end frameworks, allow the design of optical systems and the learning of the classification network parameters. In this paper, we model the optical band selection as a trainable layer that is coupled with a classification network, where the parameters are learned in an end-to-end framework. To guarantee a physically implementable system, we proposed two regularization terms in the training step to promote binarization and also the number of the selected bands, as we need to provide the conditions to design the physical element where the light passes through. The proposed method provides better performance than state-of-the-art band selection methods for three different spectral datasets under the same conditions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.