Development of a dual-modal microscopic algae detection system integrating hyperspectral imaging and a U-Net convolutional neural network.

This study presents an integrated dual-modal microscopic algae detection system that combines hyperspectral imaging and a U-Net convolutional neural network. The system achieved a spatial resolution of and a spectral resolution of 10 nm. The fluorescence imaging mode exploits chlorophyll autofluorescence under 488 nm excitation for rapid preliminary screening and precise algae impurity differentiation, achieving 98.4% accuracy. Subsequently, the transmission imaging mode performs spectral scanning on the identified regions, constructing characteristic absorption profiles through Beer-Lambert modeling. The integration of dual-modal data with U-Net analysis enabled the precise classification of three algae species with 96.5% accuracy. The system employs a centralized microcontroller-based architecture integrating a precision filter wheel mechanism, a USB 3.0 high-speed interface, and DMA double buffering for real-time data transmission. Combined with microfluidic technology, the system achieves millisecond-level mode switching and stable image acquisition, providing efficient real-time monitoring for the early detection of harmful algae.