High-throughput pharmaceutical classification via short-range interferometric hyperspectral imaging

Short-wave infrared hyperspectral imaging (SWIR-HSI) enables rapid, non-destructive material characterization with high potential for process analytical technology and pharmaceutical screening. Here, we demonstrate high-throughput, full-field SWIR-HSI using Fourier transform spectroscopy (FTS) coupled with deep learning. Conventional FTS typically requires thousands of sampling points over a wide optical path difference (OPD) range to resolve spectral features. By directly processing raw interferograms with deep learning, bypassing conventional Fourier transformation, we achieved 98.9% pixel-level and 100% whole-tablet classification accuracy across 11 over-the-counter pharmaceutical products using only 200 data points acquired over a 10 μm mirror displacement (20 μm OPD). Under challenging independent-lot, multi-day testing conditions, the framework maintained 96.7% pixel-level and 100% whole-tablet accuracy. Latent-space visualizations confirmed that short-range interferograms retain sufficient chemical and physical information for rapid, robust pharmaceutical classification. These findings indicate that deep-learned short-range interferometry offers a promising, vibration-resilient approach for high-throughput, non-destructive quality control of solid dosage pharmaceuticals.