Development of a cost-effective dual-mode hyperspectral imaging system with machine learning for enhanced bacterial classification.

This work presents the development of a cost-effective dual-mode hyperspectral imaging (HSI) system integrated with machine-learning models to detect and classify bacteria with enhanced accuracy. The HSI system was constructed using commercial off-the-shelf components and 3D-printed parts, with detailed optical simulations performed to aid in the design and validate the system's performance. A compound prism-grating-prism was implemented in an on-axis spectrograph configuration to simplify the optical assembly and minimize field-dependent aberrations. The system supports wide-field HSI in both reflectance and fluorescence modes, illuminated by chip-on-board LED sources with a visible-to-near-infrared spectrum and a narrow-band UV. The spectral resolution of the custom HSI system was determined to be 1.55 nm, with spatial resolutions of approximately 0.81 mm in the in-track direction and 0.49 mm in the cross-track direction. These resolutions are sufficient for effective spatio-spectral imaging of bacterial colonies. Furthermore, a model-training framework leveraging spectral feature fusion from both modes was developed to classify bacterial species, Staphylococcus aureus and Pseudomonas aeruginosa. The classification accuracies achieved using reflectance, fluorescence, and dual modes were 92.55%, 93.48%, and 97.11%, respectively. This dual-mode optical-computational platform not only demonstrates enhanced classification accuracy but also represents a scalable and economical solution for high-throughput bacterial identification.