High-fidelity compressed high-speed imaging for resolving rapid micro-dynamics

High-speed optical imaging is essential for capturing rapid microscopic phenomena, but often faces a fundamental trade-off between spatiotemporal resolution and data throughput. In this work, we propose a high-fidelity single-shot compressed high-speed imaging system, termed HF-CHI, which is specifically designed for resolving non-repetitive and anisotropic micro-dynamics. HF-CHI enhances the galvanometer-based scanning architecture by integrating a high-transmission static encoding mask to maximize optical throughput, supported by a multi-prior physics-enhanced network (mPEN) reconstruction framework designed to solve the associated ill-posed inverse problem. By enforcing dual physical constraints derived from both the dynamic sheared measurement and an unsheared spatial prior, the network effectively suppresses reconstruction artifacts and hallucinations common in end-to-end deep learning. We experimentally demonstrate the system's performance, achieving a high spatial resolution of 181 lp/mm and an equivalent frame rate of 50 Kfps with a single exposure by using a 40× objective. The system's versatility is validated through the precise quantitative tracking of both single and multiple microsphere trajectories in microfluidics, as well as the observation of power-dependent photothermal ablation dynamics in onion epidermal cells. Specifically, by quantifying the temporal evolution of anisotropy ratios, we reveal that the ablation morphology is inherently guided by the organized tissue microstructure across varying laser intensities. The proposed HF-CHI offers a robust and high-fidelity tool for investigating rapid micro-dynamics in biomedical microscopy and laser processing monitoring.