Event-guided microfluidic hologram deblurring via spatial-frequency learning

Microfluidic holographic imaging plays a vital role in particle analysis and biomedical applications, yet suffers from motion blur in high-speed moving targets that degrades reconstruction quality. However, conventional deblurring methods designed for macro-scale imaging fail to handle the unique high-frequency patterns of fringe-rich holograms. To address this, here we present an event-guided spatial-frequency learning approach, simultaneously exploiting the event sensor's high temporal resolution and the spectral similarity between event frames and clear holograms. Specifically, we develop an event-guided dual-domain adaptive fusion network (EDAF-Net) that integrates a frequency-domain branch to recover lost fringe patterns and a spatial-domain branch to preserve structural integrity. The framework further incorporates a dual-domain fusion module to preserve spectral fidelity and spatial accuracy. Quantitative evaluations demonstrate EDAF-Net's superior performance in high-frequency fringe recovery over state-of-the-art methods, while maintaining the lowest overall model complexity and computational burden in the comparison. Experimental validation using standardized microspheres and human red blood cells confirms the method's effectiveness for quantitative particle analysis and flow cytometry applications.