DMD-based reflective compressive spectral imaging system coupled with transformer-based reconstruction method.

Abstract

Compressive spectral imaging (CSI) enables rapid acquisition of encoded measurements, followed by spectral image reconstruction using compressive sensing algorithms. However, existing CSI systems lack compact, effective encoding designs coupled with fast, high-quality decoding methods. This paper presents a CSI system with a compact reflective optical path design based on a digital micromirror device (DMD), which facilitates additional independent compressive measurements by switching DMD patterns, thereby improving reconstruction accuracy. The system also establishes a direct one-to-one mapping between object points and image points, simplifying the design of the patch-based reconstruction algorithm. Leveraging this feature, a transformer-based reconstruction method is proposed, which divides measurements into patches and employs a transformer network to capture long-range dependencies and inter-patch similarities, reconstructing coefficients under a spatial-spectral dictionary for each patch. The proposed system and method achieve efficient acquisition and reconstruction of 256 × 256 × 22 spectral image data cubes across the 465-675 nm wavelength range.