Single-frame lens-free on-chip microscopy (LFOM) offers a compact and cost-effective solution for biomedical diagnostics and micro/nanoscale inspection but suffers from severe twin-image artifacts and frequency-dependent degradation due to numerical aperture (NA) limitations. Existing reconstruction methods often neglect the heterogeneous frequency sensitivity of LFOM and fail to explicitly suppress artifacts, leading to suboptimal complex-field measurement quality. To address these challenges, we propose SEMNet, a spectral-expert measurement network that introduces frequency-aware modeling into single-frame LFOM reconstruction. Specifically, SEMNet transforms captured holograms into polar Fourier space, partitions them into directional subbands, and adaptively routes each subband via a soft top-k gating mechanism to lightweight expert networks specialized for low-, mid-, and high-frequency restoration. A physics-consistent loss jointly enforces complex-field fidelity, Fresnel-likelihood consistency, twin-image suppression, and spectral load balancing, thereby aligning the reconstruction objective with physical imaging principles. Extensive experiments on 404 high-resolution hologram–complex field pairs from seven types of biological tissues demonstrate that SEMNet improves peak signal-to-noise ratio by up to 1.7 dB over strong single-frame baselines, with statistically significant gains in structural similarity indices (p < 0.001). SEMNet restores high-frequency textures while suppressing NA-boundary artifacts. It delivers 47.1 FPS end-to-end inference with 100× fewer parameters than a representative CNN baseline and 5× fewer MACs than a representative Transformer baseline, highlighting a frequency-structured, physics-aligned design for efficient LFOM.
18 December 2025 in Optics and Lasers in Engineering
Quantitative phase imaging via the transport-of-intensity equation (TIE) enables label-free phase measurement, but its accuracy degrades under large defocus, where the underlying nonlinearity becomes pronounced. We propose SA-RAdam, a symmetric-anchored, rectified-Adam-refined framework for high-fidelity TIE phase retrieval. A symmetric two-plane architecture yields a second-order accurate derivative prior at a virtual mid-plane, anchored to the reference plane via angular-spectrum propagation. An adjoint gradient of the forward model, combined with rectified adaptive momentum and line search, then recovers nonlinear high-frequency phase details typically lost in linear solvers. Simulations and experiments on binary and biological phase targets show consistent convergence, elimination of boundary overshoot, and RMSE reductions of over 60% relative to existing iterative solvers, substantially extending the practical defocus range of TIE-based phase retrieval.