Time-resolved fluorescence imaging is a powerful tool in life sciences and materials research. However, conventional point-scanning approaches suffer from high photodamage risk and complex optic

Time-resolved fluorescence imaging is a powerful tool in life sciences and materials research. However, conventional point-scanning approaches suffer from high photodamage risk and complex optical setups. Moreover, as temporal resolution increases, Poisson noise intensifies and may significantly obscure the fluorescence signal. Here, we propose a frame-tuning U-Net for fluorescence imaging, a U-Net based single-pixel imaging framework that incorporates frame-wise fine-tuning for enhanced time-resolved fluorescence signal recovery, particularly under photon-limited conditions. By exploiting the strong temporal correlation between adjacent frames and applying a frame-wise fine-tuning strategy, the proposed method effectively suppresses Poisson noise while preserving image details and improving reconstruction efficiency. We performed simulation experiments based on a photon-counting statistical model derived from a Poisson random process, and the results indicate that the proposed method achieves superior image reconstruction quality compared to the Hadamard inverse transform. Optical experiments on samples containing two fluorophores further validate the approach, enabling clear fluorescence imaging. This work addresses the challenge of high temporal resolution under low signal-to-noise ratios and offers a new route for dynamic biological imaging and optoelectronic material characterization.show less

Visible-light communication suffers from signal-to-noise ratio loss and blurred modulation features due to ambient light, underwater scattering and absorption, and device nonlinearity, degrading

Visible-light communication suffers from signal-to-noise ratio loss and blurred modulation features due to ambient light, underwater scattering and absorption, and device nonlinearity, degrading modulation-format recognition (MFR). We propose TF-former, a lightweight time-frequency Transformer that jointly models time and frequency domain features to enhance robustness to complex impairments. To improve computational efficiency, we introduce linear attention to MFR, reducing attention complexity from quadratic to linear. On four real-link datasets covering 19 modulation formats, TF-former achieves average accuracies of 97.25%, 98.11%, 98.68% and 66.94%. It notably retains 62.74% under extremely low SNR conditions (corridor link, transmission distance 26 m, AWG peak-to-peak voltage 0.05 V, and corresponding BER of all modulation formats between 0.34 and 0.49), gains up to 47.85% over baseline method TLDNN, and outperforms on RML2016.10a.show less

We propose and experimentally validate a high-sensitivity fiber-optic sensing framework based on the theory of preselection-free weak measurement. Unlike conventional weak measurement schemes re

We propose and experimentally validate a high-sensitivity fiber-optic sensing framework based on the theory of preselection-free weak measurement. Unlike conventional weak measurement schemes requiring precise pre-selection of input polarization states, this approach achieves weak value amplification without pre-selection through post-selection optimization alone. Consequently, it enables high-sensitivity fiber-optic sensing without stringent polarization control. Experimental results demonstrate that this sensing framework can independently achieve phase sensing at 62 dB/rad (resolution of 1.6×10^(-5) rad), pressure sensing at 2.348 dB/N (resolution of 4.2×10^(-4) N), and temperature sensing at 12.695 dB/℃ (resolution of 7.8×10^(-5) ℃). This experimental setup is simple and highly sensitive, demonstrating the immense application potential of high-sensitivity sensing based on preselection-free weak measurement.show less

We present an experimental study of proton acceleration driven by femtosecond multi-PWs laser of three different prepulse parameters with the peak laser intensity of 1.2×1021 W/cm2 irradiat

We present an experimental study of proton acceleration driven by femtosecond multi-PWs laser of three different prepulse parameters with the peak laser intensity of 1.2×1021 W/cm2 irradiating micrometer-thick metal foils. For the 4-μm-thick copper foils, the highest-energy proton beam of 58.9 MeV is generated with the moderate-contrast laser, while the low-contrast or high-contrast lasers result in the lower proton cutoff energies. The 1D hydrodynamic and 2D particle-in-cell simulations indicate that the front preplasma of foils induced by laser prepulse can enhance electron acceleration and in turn improve proton acceleration, while the rear preplasma will weaken the sheath field and be unfavourable for accelerating ions. For the case of the moderate contrast, the scale length of front preplasma is long enough to generate high-temperature electrons and the scale length of rear preplasma is so short that the sheath field still keeps strong, which is advantageous for generating high-energy protons.show less