Compressive optical architectures for data acquisition: transmissive and reflective cases

Abstract

The growth of optics and electronics, together with more robust learning algorithms, has driven the implementation of compressive sensing in more economical architectures for data acquisition, such as single-pixel architectures. Taking advantage of these advances, a novel solution applicable to multiple sectors is presented. An example of this is compact hyperspectral sensors, which have low adoption in industry and research due to their high costs. This work is presented as a practical and theoretical tutorial on the design and implementation of single-pixel hyperspectral architectures with compressive sensing. The fundamental theory is detailed, and guidance is provided on the process of constructing two systems: one based on a reflective modulator and the other on a transmissive modulator. Step-by-step instructions are provided on how to integrate these elements to acquire hyperspectral data and reconstruct images. The efficiency of both assemblies was evaluated, obtaining Mean Absolute Error (MAE) values of less than 0.15 when compared to a commercial hyperspectral camera, demonstrating an acceptable average level of accuracy; a Spectral Angle Mapper (SAM) value of less than 0.3, metrics considered appropriate for practical applications in different contexts, such as the analysis of food, materials, or structures. It is confirmed that, following the optical design guidelines and compressive sensing tools presented in this work, it is feasible to develop functional hyperspectral sensors that are low cost and highly accessible.