High-precision and efficient forward models for near-field diffraction are essential for computational imaging systems. However, achieving a balance among physical accuracy, computational efficiency, and geometric flexibility remains a fundamental challenge. This paper presents a systematic review of numerical simulation methods for near-field diffraction, covering their evolution from classical physical formulations to modern efficient sampling approaches. We begin with the Rayleigh–Sommerfeld (RS) integral as a rigorous physical benchmark, highlighting its high physical fidelity and prohibitive O(N⁴) computational cost. We then review fast Fourier transform (FFT)-based acceleration frameworks, focusing on the angular spectrum method (ASM) and its band-limited variant (BL-ASM), the latter of which effectively suppresses sampling-induced aliasing artifacts. To address the limited geometric flexibility of fixed-grid methods, we further examine adaptive approaches, including Fresnel diffraction, the fractional Fourier transform (FrFT), and the scalable angular spectrum (SAS) method, which enables zoom propagation via sampling decoupling. Finally, we discuss the limitations of current flexible-geometry methods, particularly the degradation of physical accuracy in extreme near-field and wide-angle regimes, which restricts their applicability in high numerical aperture (high NA) scenarios. We conclude by outlining future directions toward high NA vectorial modeling and intelligent operator integration.

Unmanned aerial vehicle (UAV)-based infrared target tracking has become a research hotspot with wide application prospects. However, issues such as similar target appearance features and insufficient real-time performance limit its practical deployment. To address these problems, this paper proposes an integrated framework for infrared target detection, multi-target tracking, and edge deployment. For detection, we present an anti-interference and multi-scale-aware detector. A vision Transformer (VIT)-based backbone is optimized via interference-aware and edge-aware pre-training, and a multi-scale feature encoder is designed to enhance detection robustness. For tracking, an end-to-end graph attention network-based tracker is developed. By optimizing edge feature modeling, message-passing, and graph pruning, and adopting a query decoupling strategy, the framework alleviates task coupling and enables joint optimization. Furthermore, a lightweight convolution structure is proposed to improve depthwise separable convolution, compressing the model for efficient deployment on the RK3588 platform. Extensive experiments validate the effectiveness of the proposed method, which achieves strong tracking performance and meets real-time requirements on edge devices.

Recent advances in distributed computing enable privacy-preserving aggregation and artificial intelligence (AI)-driven analysis of multimodal medical data, empowering real-time distributed healthcare applications and large-scale disease detection systems. Inspired by its remarkable performance, we proposed to discover medical knowledge embedded in big data with a large multimodal model, offering intelligent disease analytics via this promising AI route. In this paper, we proposed a distributed multimodal representation learning structure for zero-shot medical image classification. Specifically, we simultaneously adopted the implicit knowledge extracted from a large multimodal model built on images, and the explicit knowledge extracted from the medical knowledge graph built on textual records. Facing the inconsistent alignment in latent space constructed by multimodal data, a cross-modal alignment strategy was proposed to adjust intra- and inter-modal representations for convinced learning. Experiments on several public datasets proved that the proposed framework could improve the accuracy of zero-shot medical image classification, achieving robust and accurate disease analytical results.

Optical computing is increasingly regarded as a promising alternative to conventional electronic computing due to its inherent advantages in speed and energy efficiency. Diffractive deep neural networks (D²NNs), as a typical optical computing architecture, suffer from insufficient feature extraction capability and poor flexibility in nonlinear classification. To tackle these issues, a hybrid D²NN architecture is proposed. In this hybrid D²NN architecture, feature extraction is accomplished by cascaded optical diffraction layers, whose material parameters (relative permittivity ε and permeability µ) are optimized using particle swarm optimization (PSO) to enhance light field modulation. Specifically, three diffractive layers are retained and a linear gradient distribution of ε and µ is applied to simplify the fabrication process, thereby enabling the refractive index n to be modified for improved feature modulation. Next, nonlinear classification is facilitated by fully connected electronic layers appended after the diffractive stages, which compensate for the linear constraints inherent to purely optical systems. Consequently, the proposed hybrid architecture combines efficient optical feature extraction with flexible electronic classification. Experimental results on the MNIST, Fashion-MNIST and grayscale CIFAR-10 datasets demonstrated that classification accuracies of 96.8%, 89.73% and 53.32% were achieved by the hybrid architecture of D²NN, respectively, which significantly outperforms the original D²NN.

Reverberation and clutter suppression is critical for active sonar systems, especially when detecting weak underwater targets in shallow water environments. We propose an online dereverberation algorithm that models reverberation as a low-rank background and targets as sparse outliers. The method utilizes the alternating direction method of multipliers (ADMM) for sparse decomposition, alternating it with subspace tracking on the Grassmann manifold to enable frame-wise adaptation with low latency. Compared with traditional batch-based methods and representative online algorithms, the proposed algorithm achieves comparable suppression performance while demonstrating lower latency and higher computational efficiency. Experiments on real-world sonar datasets demonstrate its effectiveness, robustness, and suitability for deployment on resource-constrained embedded platforms.

Clutter suppression is a critical challenge in airborne passive radar systems, where platform motion induces significant Doppler spreading that renders conventional suppression algorithms ineffective. To address this limitation, this paper proposes a clutter suppression algorithm based on joint sparse recovery. Similar to the CLEAN iterative procedure, the proposed algorithm utilizes the least-squares method to estimate the clutter component in each iteration. During each iteration, it leverages both the joint sparse characteristics of multi-channel clutter and the cluster structure within channels to select multiple clutter components. This design enables efficient joint clutter estimation across multiple channels. Furthermore, the proposed algorithm utilizes the estimated direct-path signal to derive correction factors, ensuring stable performance even in the presence of channel amplitude and phase errors. Finally, a series of simulation experiments are conducted to evaluate the performance of the proposed algorithm.

Research in quadrotor control has yielded promising results, with most algorithms being deployed onboard. However, the substantial mass of the onboard computer limits the quadrotor’s performance potential. In this work, we propose a remote autonomous control (REAC) system that shifts computational loads to a remote computer, eliminating the onboard computing payload and enabling multi-quadrotor control. REAC consists of an explicit trajectory planner that generates reference trajectories at low frequency and a robust controller that follows them by generating real-time actions. However, remote control inherently introduces persistent and time-varying transmission delays, leading to inaccurate state estimation and a temporal mismatch between control actions and system dynamics. To address this, we incorporate a temporal buffering mechanism into the controller. In parallel, we propose a two-phase training strategy: initial policy shaping through imitation learning, followed by reinforcement learning refinement with randomized disturbance injections. Simulation experiments demonstrate the effectiveness of the REAC in remote quadrotor control and exhibit robustness against environmental uncertainties brought by transmission delays, while preliminary real-flight tests indicate practical feasibility.

In distributed speech front-end frameworks, multi-channel enhancement typically relies on inter-channel time synchronization and is therefore susceptible to delay mismatch, which may lead to performance degradation. Based on this, this work regards independent single-channel enhancement as a complementary solution: when time synchronization information becomes unreliable, it can still maintain relatively stable performance. In single-channel research, noisy scenarios hinder accurate retrieval of non-structured, envelope-bound phase; without explicit constraints, amplitude-phase imbalance disrupts harmonics and continuity, making explicit modeling pivotal. MP-SENet (mask phase aware speech enhancement network) significantly advanced parallel amplitude-phase modeling for simultaneous optimization but struggles with residual noise and phase adaptation in low-SNR environments. To address this, an enhanced MP-SENet is proposed: FRI (filter recycle interguide) in TF-Transformer disentangles global-local speech-noise features; a lightweight SR post-processing module refines the amplitude–phase estimation, curbing noise and artifacts; parallel phase and metric discriminators enable joint optimization to enhance quality and intelligibility. It achieves state-of-the-art results on VoiceBank+DEMAND (PESQ: 3.67, CSIG: 4.96, STOI: 0.97) and robust generalization on real-world WHISPER_SET_1 at 0, –5, –10 dB, which makes it promising for distributed front-end systems.