Overcoming the Nyquist Limit in Molecular Hyperspectral Imaging by Reinforcement Learning

Recent advances in vibrational spectroscopic imaging have revolutionized molecular characterization across biomedical applications. However, the inherent Nyquist limit results in prolonged acquisition times and, more critically, may lead to decreased analytical accuracy, a phenomenon known as the Hughes phenomenon. To address this challenge, an explorative spectral acquisition guide (ESAG) has been developed to directly optimize downstream analytical tasks while drastically reducing data acquisition. ESAG uniquely enables on‐the‐fly wavenumber optimization without prior knowledge. In hyperspectral stimulated Raman scattering imaging experiments, ESAG demonstrates remarkable improvements in segmentation accuracy while requiring only 70% spectral frames, and notably, achieves higher accuracy using only 60% spectral frames and maintains exceptional segmentation performance (>93% accuracy) in complex polymer mixtures using merely half of the conventional spectral measurements. These results establish ESAG as a transformative approach for efficient and accurate hyperspectral imaging across chemistry, biomedicine, and remote sensing.