Coded aperture snapshot spectral imaging (CASSI) enables single-shot acquisition of spatial–spectral information, showing strong potential for rapid spectral sensing on underwater dynamic platforms. In underwater environments, due to spatial non-uniformity changes in water’s absorption and refractive index, the focal-plane is more sensitive perturbations compared to terrestrial conditions. However, existing focusing approaches—based on visual inspection or reconstruction-dependent post evaluation—lack real-time and quantitative criteria. In this work, we investigate the effect of coded measurements at different focal distances on reconstruction performance through simulation experiments., and propose a Noise-Aware Dual-Adaptive Frequency-domain Sharpness (NADFS) metric, which is computed directly from raw coded measurements to quantify focus status and facilitate real-time selection of optimal measurement frames. Guided by the simulation, we developed a CASSI prototype with a high-precision focusing mechanism operating in the visible range (520–670 nm) with 2048 × 2048 spatial resolution, 65 spectral channels. Water-tank experiments indicate that the performance of the CASSI system is highly sensitive to focus accuracy, where defocus errors can directly degrade the spatial structure and frequency-domain characteristics of the reconstructed images. The proposed NADFS, in conjunction with the integrated prototype, enables more stable and accurate focus estimation, reliable selection of optimal-focus measurements, enhances the precision of spatial–spectral reconstruction, and significantly improves the overall performance and robustnes of the CASSI system in underwater environments.
Qi et al. (Sun,) studied this question.