Infrared Color-coded Aperture Optimization for Object Tracking and Spectral Classification
Abstract
This report presents the outcomes of the project Infrared color-coded aperture optimization for object tracking and spectral classification conducted from September 15th, 2021, to September 14th, 2023. The report outlines the proposed approach for designing near-infrared coded apertures employing an end-to-end method that links the sensing with inference tasks. This methodology consists of a fully differentiable sensing model coupled with deep learning models to perform either spectral reconstruction or spectral classification, directly from the encoded measurements. Specifically, the proposed approach was studied for two different compressive spectral imaging systems: the single-pixel camera and the color-coded filter array sensor. In addition, an infrared spectral image dataset was acquired during this project. This dataset was employed to train the deep learning model. Simulation results show that the designed color-coded apertures can significantly enhance classification and reconstruction performance compared to random or analytical designs, what indicates a promising technology to be applied in the sensing and processing of near-infrared images. Further, test-bed implementations of the optical systems were built to evaluate the effectiveness of the proposed design in controlled laboratory scenarios. The results show that the proposed design improves the spatial-spectral resolution and reduces the number of measurements needed for a suitable reconstruction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 06, 2023
- Accession Number
- AD1224896
Entities
People
- Henry Argüello
Organizations
- Industrial University of Santander