The rapid evolution of the Internet of Things and edge computing has drawn significant attention to WiFi-based non-intrusive human pose estimation due to its ubiquity and privacy-preserving nature. However, translating this technology from laboratory settings to practical edge deployment presents two key challenges: firstly, mainstream models, characterized by their massive parameters and high computational complexity, demand substantial hardware resources, which stands in stark contrasts to the limited storage, computing power, and power budget of edge devices; secondly, most existing methods are limited to single-person scenarios, lacking effective estimation for multi-person poses in real-world settings. To address these issues, this paper proposes a lightweight WiFi-based multi-person pose estimation model specifically optimized for edge environments. By introducing an efficient backbone network built upon depthwise separable convolutions and selective convolutional modules, we significantly compress the model′s complexity. Experiments on a self-built dataset demonstrate that our solution achieves comprehensive optimization in model size, computational overhead, and memory consumption while maintaining leading accuracy. These three key metrics were reduced by 70.1%, 85.5%, and 70.3%, respectively, compared to the baseline method, alongside notable improvements in key metrics like PCK@10 and PCK@20. This study confirms the feasibility of deploying lightweight models for real-time, multi-person WiFi pose estimation on edge devices, providing an effective pathway for real-world application.

As a revolutionary breakthrough intelligent construction, 3D printing construction technology (3DCPT) is driving the transformation of traditional civil engineering materials toward high performance and functionalization. However, construction material for 3D printing still face limitations in terms of rheological properties, printability, mechanical performance, and durability. This article briefly outlines the technical principles of 3DCPT and the performance requirements of 3D printed concrete (3DPC), with a focus on reviewing the modification effects and mechanisms of nanomaterials such as nano-calcium carbonate, nano-alumina, nano-silica, and carbon nanomaterials in 3DPC, as well as the synergistic effects of nanomaterials. Finally, the current challenges of 3DCPT are summarized, and future prospects are proposed.

Critical points with emergent symmetry have become an important topic in studies of phase transitions and critical phenomena in recent years. The existence of emergent symmetry is often accompanied by the appearance of a dangerously irrelevant scaling variable (DISV), which induces rich scaling behaviors. In this work, we consider a three-dimensional Z₅-clock model that hosts an emergent U(1) symmetry at the critical point, and investigate its scaling behaviors during nonequilibrium relaxation processes under different initial conditions. Using Monte Carlo simulations, we show that the relaxation scaling behaviors of the amplitude order parameter, Binder cumulant, and the correlation length is governed by the three-dimensional XY universality class and are not affected by the DISV. In contrast, the angular order parameter exhibits a two-stage scaling behavior. In the early stage, its dynamic scaling also obeys the three-dimensional XY universality class, while in the long-time regime, its relaxation scaling behavior should be characterized by an additional dynamic exponent z′ induced by the DISV. Our results demonstrate the irrelevant and dangerous aspects of the DISV in nonequilibrium relaxation dynamics.

Differential Evolution (DE) has been widely applied to complex optimization problems due to its simple structure and strong global search capability. However, its performance highly depends on parameter settings and mutation strategies, which severely limits its generalization ability and convergence efficiency. To address these issues, this paper proposes a Reinforcement Learning and Surrogate-based Adaptive Differential Evolution (RSADE) algorithm. The proposed RSADE algorithm integrates three cooperative mechanisms, namely reinforcement learning, surrogate modeling, and hybrid mutation strategy. First, an Actor-Critic reinforcement learning module is introduced to adaptively adjust the control parameters F and CR according to the current search state, thereby achieving dynamic control of the search behavior. Second, a surrogate model based on Gaussian Process Regression (GPR) with a Radial Basis Function (RBF) kernel is embedded to replace real fitness evaluations with predicted values when the uncertainty is low, which significantly reduces computational cost. Finally, a hybrid mutation strategy combining linear scheduling and adaptive probability switching is designed to emphasize global exploration in the early stage and local exploitation in the later stage, thus maintaining a dynamic balance between exploration and exploitation. Numerical experiments on truss structure optimization problems demonstrate that RSADE outperforms conventional algorithms in terms of convergence speed, optimization accuracy, and solution stability, verifying the effectiveness and superiority of the proposed method.

Perovskite solar cells (PSCs) have emerged as a leading candidate for next-generation photovoltaics due to their high power conversion efficiency and low-cost processability. This review focuses on all-inorganic perovskites, providing a systematic overview of recent advances in their material properties, film morphology control, and interfacial passivation strategies. We first summarize the intrinsic advantages of all-inorganic systems in terms of thermal stability, moisture resistance, and photo-oxidation durability, while elucidating key performance-limiting factors such as surface and bulk defects, halide migration, and phase instability. We then highlight the mechanisms and effects of mainstream passivation strategies—including surface termination engineering, low-dimensional phase incorporation, and interfacial reconstruction—with representative examples such as buried interlayers, photo-switchable passivation molecules, and low-dimensional reconfiguration that enable synergistic improvements in efficiency and long-term stability. Finally, we outline future research directions: coupled multiscale in-situ characterization and theory, sustainable and dynamic molecular passivation design, lead-free and oxide-based alternatives, dimensional engineering of quasi-2D structures, and scalable fabrication with standardized stability evaluation protocols. This review aims to provide a comprehensive perspective and roadmap for the material innovation and device engineering of all-inorganic perovskite photovoltaics.

In this study, a novel poly(p-phenylene benzodioxazole) nanocarbon fiber paper self-supporting platinum-ruthenium alloy catalyst layer is innovatively prepared, presenting a significant departure from the traditional physical ultrasonic mixing of commercial platinum-carbon (Pt/C) and ruthenium dioxide (RuO₂) mixed catalyst layers. It possesses uniform active sites for both oxygen reduction and evolution reactions and hydrogen evolution reactions. The PtRu alloy was synthesized via CO₂ activation and reduction of the precursors of Pt and Ru, demonstrating superior electrochemical performance. In 0.1 M KOH electrolyte, the overpotential of CO₂ PtRu350 for the oxygen evolution reaction (OER) can reach 266 mV, and the half-wave potential of the oxygen reduction reaction (ORR) is 0.857 V, with a potential difference (ΔE) of 0.663 V, which is better than that of EG PtRu150 (ΔE=0.91 V) and CO₂-Ru@EG-Pt150 (ΔE=0.735 V). The test results of zinc-air batteries show better battery performance than the commercial Pt/C+RuO₂ catalyst layer, with a peak power density reaching 100.7 mW·cm^-2, significantly higher than the 93.1 mW·cm^-2 of the commercial Pt/C+RuO₂-zinc-air battery. It exhibits excellent rate performance and good reversibility. After 1440 cycles of charge and discharge, the voltage gap only increases to 0.741 V from its initial value, demonstrating great stability. This study provides a new and promising approach for the design of bifunctional catalysts in self-supported catalyst layers for zinc-air batteries.

Mosquito-borne diseases pose a serious threat to human health, and controlling mosquito population is the key to preventing and controlling these diseases. Population replacement with Wolbachia can inhibit the replication and transmission of pathogens in mosquito bodies. In this article, we first consider the population replacement of a wild mosquito population in a single region dependent on fitness cost, and study the dynamic trend of Wolbachia infection proportion by constructing a discrete model. Then consider the migration of mosquito vectors between two different sub-regions. By analyzing the different fitness costs of female mosquitoes infected with Wolbachia in different subregions, the effects of initial concentration and adaptation costs on population replacement were presented. We obtain the fitness cost thresholds, denoted by s^*_f1 and s^*_f2, and find that Wolbachia-infected mosquitoes can persist only when 0<s_f≤s^*_f1 or s^*_f2≤s_f<1. The theoretical results are verified through numerical simulations. The study also explores the migration effect between two regions in a heterogeneous environment, demonstrating that the infection frequency is higher in regions with lower fitness costs. Further, by introducing the threshold of breeding site ratio β^*, it is shown that the threshold of Wolbachia infection frequency decreases when the breeding site ratio p₀<p^*, β<β^*, and further reduces to 0 when p₀<p^*, β≥β^*.

This paper studies a predator-prey system with a strong Allee effect in the prey and intraspecific competition in the predator. Using small perturbation analysis and normal form theory, we comprehensively and systematically characterize the local dynamical behaviors of the system near the Bogdanov-Takens (BT) bifurcation point. We present the saddle-node bifurcation curve, Hopf bifurcation curve, and homoclinic bifurcation curve that bifurcate from the BT point, thereby enriching the previous research findings on this model. The research results indicate that intraspecific competition in predators has a profound impact on the system′s dynamical behaviors, causing the system to display a wide variety of dynamic characteristics, encompassing complex phenomena such as bistability, limit cycles, and homoclinic orbits.

The sustainable development of the banana (Musa spp.) industry, a key economic and food crop, is vital for regional economies and global food security. Remote sensing has become essential for large-scale, dynamic, and precise monitoring. This paper reviews current research in four areas: plantation identification, growth parameter retrieval, stress monitoring, and yield estimation.Significant progress has been made. Methods are shifting from traditional statistics to machine and deep learning, data sources from single-sensor to multi-source fusion, and monitoring scales from regional to individual plants, improving accuracy across applications.Yet three major challenges persist. First, asynchronous growth-caused by extended planting seasons and differences between parent and ratoon crops-creates spectral heterogeneity, reducing model accuracy. Second, multi-source data fusion lacks a clear understanding of synergistic mechanisms and effective integration methods. Third, reliance on empirical models limits robustness and generalization in complex conditions such as intercropping and fragmented plots.Future research should mitigate asynchronous growth effects, integrate remote sensing with growth process models to develop more mechanistic approaches, and broaden stress monitoring to include events like lodging and drought. Combining advanced sensors (UAVs, hyperspectral, LiDAR) with lightweight real-time algorithms will be key to enabling single-plant precision management.