Yuchen He

Conventional target recognition approaches typically depend on large-scale datasets and computationally intensive electronic processing, which inherently limits processing speed and raises data privacy concerns. In contrast, optical computing supports ultra-high speed parallel processing, and the human visual system can directly perceive integrated light field information. Motivated by this, we propose a human-perceptible ghost recognition scheme with feedback optimization. Ghost recognition extends the principle of ghost imaging by directly recognizing targets from the bucket signal, thereby circumventing the conventional image reconstruction step. The system employs an all-optical diffractive deep neural network (D ² NN) to perform feature extraction, mapping the incident wavefront onto a predefined light field intensity distribution. By leveraging the visual persistence of the human eye, the time-integrated light signal can be directly recognized without electronic post-processing. The proposed system achieves 90.72% recognition accuracy, validating its feasibility in practical scenarios. This work highlights the potential of combining optical computing with visual perception to develop efficient recognition systems.

The Hanbury Brown–Twiss (HBT) interferometer, which is used to measure the correlation of two photons or two light beams, has played a crucial role in quantum optics and quantum information. We propose a generalized HBT interferometer, which is able to measure the correlation of photons or light beams up to eight in a single experimental run. With an improved time correlated single-photon counting (TCSPC) algorithm, two-, three-, and four-photon coincidence counts, and the degree of fifth-order coherence of pseudothermal light are measured with a data collection time of 60 s. The generalized HBT interferometer can be used to measure the correlation of two multi-photon detection events by simulating photon number resolving detectors with multiple single-photon detectors. The generalized HBT interferometer and improved TCSPC algorithm provide a versatile framework for studying multi-photon correlations and interference, which is helpful to understand the physics of multi-photon interference and may find applications in fluorescence lifetime imaging and other techniques using TCSPC.

We analyze correspondence ghost imaging (CGI) from a machine learning perspective, focusing on its intrinsic mathematical consistency with the perceptron model. Based on this analysis, linear classification algorithms such as the perceptron and logistic regression are applied to CGI data, yielding improved image quality compared to traditional methods. Experimental and simulation results demonstrate that logistic regression achieves the highest structural similarity index (SSIM) in reconstruction tasks. Furthermore, CGI is employed for solving standard classification tasks, achieving comparable accuracy with dramatically lower computation time—up to 10× faster than logistic regression. These findings uncover the potential of interpreting CGI as a lightweight linear classifier.

Scattering presents significant challenges for practical imaging applications. Single-pixel imaging offers advantages over conventional approaches, yet further advances are still required. In this work, we introduce sparse Hadamard speckle patterns with tunable sparsity and propose sparse Hadamard single-pixel imaging (SHSI), a new framework built on these patterns. By integrating orthogonality, sparsity, and pseudo-randomness, the proposed patterns provide clear benefits over traditional Hadamard speckle patterns and their variants. SHSI improves noise robustness across a wide range of sampling rates by encoding sparse but spatially correlated information within each illumination pattern. It also achieves high-resolution imaging with low-order Hadamard matrices, which substantially reduces memory requirements. Moreover, SHSI operates without prior knowledge, training, or post-processing, thereby simplifying implementation. To further enhance performance, we introduce two operational modes and develop a complete theoretical framework that is validated through simulations and experiments. SHSI has the potential to advance computational imaging platforms based on sequential correlation measurements and to enable real-time reconstruction of occluded or embedded targets within scattering media.

The fusion of light detection and ranging (LiDAR) and camera provides comprehensive information about the depth and color texture of objects, finding wide applications in autonomous driving, three-dimensional (3D) scene reconstruction, and other fields. However, point clouds acquired via LiDAR are susceptible to noise, impacting downstream tasks like surface reconstruction and object recognition demanding high-quality point cloud outputs from 3D imaging systems. This paper proposes a novel 3D fusion imaging technique that enhances the quality of the fused point clouds from three distinct aspects, improving the accuracy and precision of point clouds by employing matched filter method, designing a preprocessing workflow for point clouds and images to enable their effective fusion, and smoothing noise and accentuating texture features using a filter algorithm on fused information. The method of matched filtering has been employed to reconstruct object distance information from photon count-time histograms, resulting in an improvement of approximately 120% compared to the method of directly calculating the average photon flight time. The point cloud filter proposed utilizing fused information demonstrates superior denoising capabilities compared to the state-of-the-art and has the capability to cluster points with similar features to highlight the texture characteristics of the point cloud.