Abstract: Aiming at the non-stationary and slowly varying stochastic nature of bearing degradation from normal operation to failure, this paper proposes a fratcional Generalized Cauchy (fGC) prediction model with 1/f process and dual parameters: fractal dimension and Hurst exponent. First, 1/f process sequences exhibit long-range dependence and power-law characteristics. Next the fGC degradation model is established, and the Hurst exponent and fractal dimension are calculated using the R/S method and box-counting dimension method, respectively. Then a dimensionless jump descriptor is employed as a Health Indicator to detect incipient faults and estimate degradation parameters. The maximum likelihood algorithm method is applied to parameter estimation. Finally, a experiment verifies the satisfactory prediction performance through compared with CNN and LSTM predicting model.

Abstract: Titanium-based dental implants have evolved significantly, with the development of binary alloys like Ti-15Zr (Roxolid™) representing a pivotal advancement in mechanical performance. Current research focuses on biomimetic surface engineering to further accelerate osseointegration and optimize bone regeneration, particularly in clinically compromised sites. This review constitutes a narrative synthesis of how these strategies replicate the bone extracellular matrix (ECM) through a holistic framework of architectural, mechanical, and biochemical integration. A structured literature search across PubMed, Scopus, and Web of Science (2010–2026) identified relevant studies focusing on the synergy between Ti-15Zr substrates and surface modifications. Evidence confirms that the high fatigue strength of Roxolid™ alloys provides an ideal foundation for advanced, hierarchical surface engineering without compromising structural integrity. This strategy utilizes macro-topography for primary stability, nano-topography for protein adsorption, and bio-functionalization (e.g., RGD peptides, osteogenic ions) to direct mesenchymal stem cell (MSC) differentiation. This synergy accelerates the transition from passive to active osseointegration, effectively bridging the "biological gap" during early healing. Biomimetic engineering transforms implants into instructive biological platforms, improving outcomes for patients with compromised bone quality and facilitating predictable immediate loading protocols.

Abstract: The k-nearest neighbor (KNN) algorithm remains one of the most fundamental and widely-used methods in machine learning. A common rule of thumb sets the number of neighbors as k = √n, where n is the size of the training set. Despite its widespread adoption, the theoretical justification for this choice has remained obscure. We provide a comprehensive rate-based analysis. First, we derive the minimax-optimal exponent β★ = 4/(d+4) for k = n^β under standard Hölder-smoothness assumptions, recovering as a special case Theorem 1: k = √n is minimax-optimal if and only if the feature space has dimensionality d = 4. Second, Theorem 2 quantifies the sub-optimality of any fixed β as R_n(n^β) = Θ(n^(−r(β,d))) with r(β,d) = min{β, 4(1−β)/d}, yielding an asymmetric penalty for the classical rule when d ≠ 4 that we make precise in Corollary 1. The predicted rate is empirically verified across d ∈ {2,…,20} on controlled synthetic data. On 48 datasets from the OpenML-CC18 benchmark suite, the dimension-aware rule k = ⌊n^(4/(d+4))⌋ outperforms the classical √n rule in 32 of 48 head-to-head comparisons (paired Wilcoxon p = 4.6 × 10⁻⁴, mean accuracy gain +2.5 percentage points), demonstrating that the theoretical improvement translates to a practical one. We further test the Hassanat distance metric against Euclidean across all KNN variants on the same 48 datasets, finding that Hassanat outperforms Euclidean in five of six configurations (paired Wilcoxon p < 0.05), with the largest gains on unstandardized data. Cross-validation remains the strongest k-selection strategy when computationally feasible, and the theoretical results provide a principled non-cross-validated alternative.

Abstract: Automated diabetic retinopathy (DR) screening has achieved expert-level accuracy, yet clinical adoption remains limited by the opacity of deep neural networks. We address this gap with a DenseNet121-based binary classifier trained on 3,662 retinal fundus images from APTOS 2019, optimised through a two-phase transfer-learning and fine-tuning strategy with focal loss and class-balanced sampling. The model achieves 95.45% test accuracy, an AUC-ROC of 0.9881, sensitivity of 93.91%, and specificity of 95.94%. To make these predictions interpretable, we integrate and systematically benchmark six complementary explainable AI (XAI) techniques drawn from three theoretical families: perturbation (Occlusion Sensitivity, LIME, RISE), gradient (Integrated Gradients), and activation-based (Grad-CAM++, Score-CAM). Each method is evaluated on processing time, memory footprint, and agreement with expert-annotated anatomical structures. The six methods converge on clinically meaningful regions, including the optic disc (85% average agreement), major vessels (78%), and macula (66%), indicating that the network's decisions are grounded in established DR pathology rather than spurious correlations. Statistical tests (DeLong, McNemar, bootstrap with 10,000 resamples) confirm significant gains over standard CNN baselines. The framework demonstrates that strong screening performance and clinical interpretability can be jointly achieved, providing a deployment-ready template for DR decision support.

Abstract: Chronic Mycobacterium tuberculosis (M.tb) infection reflects failure of sterilizing immunity and persistent pulmonary bacterial burden. While CD4+ T cells and IFN-γ are central to protection, the role of CD8+ T cells in chronic disease remains unclear. This study examined whether CD8+ T cells contribute to immune dysregulation during chronic tuberculosis through IL-10 production. Susceptible CBA/J and resistant C57BL/6 mice were infected with a low-dose aerosol of M.tb Erdman and followed for 150 days. Lung bacterial burden, cytokine responses, and T-cell populations were assessed using high-purity CD8+ T-cell isolation (>97%), ELISA, ELISPOT, and in vivo CD8+ depletion. In susceptible CBA/J mice, chronic infection was associated with progressive pulmonary accumulation of CD8+ T cells, reduced CD4:CD8+ ratios, increased IL-10 levels, and impaired bacterial control. Antigen-experienced CD8+ T cells were a major source of IL-10, which correlated with reduced IFN-γ responses and higher bacterial burden. CD8+ depletion during chronic infection was associated with reduced bacterial burden and increased IFN-γ responses. Resistant C57BL/6 mice showed limited expansion of IL-10-associated CD8+ responses and better bacterial control. These findings support a model in which chronic M.tb infection is associated with expansion of IL-10-producing CD8+ T cells in susceptible hosts and altered immune control. CD8+ T-cell modulation during chronic disease is associated with changes in bacterial burden, suggesting a contributory role in disease outcome. These results highlight CD8+ T-cell functional polarization as a factor to consider in tuberculosis pathogenesis and vaccine design.

Abstract: Small and medium enterprises (SMEs) in the entertainment sector face significant challenges managing seat assignments through manual processes that are error-prone and time-consuming. This paper presents an intelligent agent-based system that automates seat assignment while providing natural language support for operational staff. The system integrates a large language model (Gemini 2.5 Flash) for conversational interaction with a constraint-based optimization algorithm that considers capacity, accessibility, revenue, and business priorities. A fuzzy matching engine combining spaCy with the fuzzy string matching library FuzzyWuzzy consolidates duplicate reservations from multiple channels. The cloud-based architecture leverages AWS serverless services (Lambda, Fargate) with PostgreSQL for data management. Technology Readiness Level 4 (TRL4) validation demonstrated 94% precision in duplicate detection, successful assignment of 87% of reservations with 82% average capacity utilization, and effective natural language query handling. The system reduces manual processing time by 65% while improving assignment quality through systematic enforcement of constraints. This work demonstrates the feasibility of AI-powered operations management for resource-constrained SMEs, offering a practical reference architecture combining conversational AI with algorithmic optimization.

Abstract: Traditional biomechanical research has focused on joint mechanics and the be-havior of the center of mass (CoM); however, the dynamics of the center of pressure (CoP) and the neuromuscular strategies that support stability under load remain un-derexplored. This study examined balance control during the back half squat by ana-lyzing the relationship between the CoM and the CoP in five experienced male weightlifters performing segmented squats at five load levels (20–80% 1RM). Vicon force platforms and motion capture systems were used to quantify the area of the 95% confidence ellipse. Due to the small sample size (n = 5), a dual inferential framework was used: frequentist repeated-measures ANOVA supplemented with unified adaptive Bayesian hierarchical modeling. The phase of the motion showed a strong effect on the stability of the CoP in both frames (frequentist: F(1.65, 6.59) = 19.44, p = 0.002, ηp² = .829; Bayesian: P(β_phase2 < 0) > 0.999). The external load was not significant in the frequentist analysis (p = 0.177), but the subsequent Bayesian analysis gave moderate evidence of a positive load effect (β_load = 0.059, 95% HDI [0.005, 0.115], P=0.981). The area of the CoM ellipse did not show significant effects in either frame. Limb asym-metries were substantial and stable (frequentist: 48.01 ± 30.13%; Bayesian: 69.48%, 95% HDI [55.86%, 81.44%], P(AI > 20%) = 1.000), with no condition-dependent modulation. CoP-CoM coupling was stronger in the mediolateral than the anteroposterior direction in both frames. These results suggest that phase is the main factor influencing pressure point stability, that loading may have a moderate positive effect detectable only with low-power probabilistic techniques, and that the dual framework improves the ro-bustness of conclusions in biomechanical research with small samples. Confirmatory studies with larger sample sizes are needed.

Abstract: Time-restricted feeding (TRF) has emerged as a promising intervention to improve metabolic health and promote healthy aging, yet the cellular mechanisms underlying its effects remain incompletely understood. Here, we applied imaging-based and quantitative cellular analyses to investigate how TRF modulates aging-associated and neurodegeneration-related phenotypes in vitro. Human fibroblasts and AC16 cardiomyocytes were used as models of cellular aging, alongside fibroblast-based models of neurodegeneration. TRF was simulated through cyclic nutrient availability, and cellular responses were evaluated using microscopy-based assessment of cellular morphology, senescence-associated features, metabolic state, and circadian rhythm-associated gene expression dynamics. Imaging analyses demonstrated that TRF modulated key hallmarks of cellular senescence, including changes in cell morphology and intracellular organization, consistent with enhanced cellular resilience and altered metabolic adaptation. In AC16 cardiomyocytes, TRF influenced aging-associated cellular phenotypes, indicating that its effects extend beyond proliferative cell systems to cardiac-relevant models. In neurodegeneration-associated fibroblast models, TRF altered disease-related cellular signatures and stress-associated phenotypes, supporting a potential protective role in neurodegenerative conditions. Quantitative analyses further revealed significant TRF-induced changes in circadian rhythm characteristics across all models, including altered oscillatory amplitude, supporting a mechanistic link between nutrient timing and cellular timekeeping. Together, these findings demonstrate that TRF induces measurable changes in cellular architecture and circadian regulation associated with improved aging- and neurodegeneration-related phenotypes. This work highlights the utility of imaging-based approaches for investigating the spatiotemporal cellular effects of metabolic interventions and supports TRF as a potential therapeutic strategy for age-associated diseases.

Abstract: The southern coast of Java, Indonesia, lies along the active Sunda subduction margin, where coastal landforms record the interaction between sea-level change, wave erosion, sedimentation, and tectonic uplift. Marine terraces and raised coastal surfaces are important geomorphic indicators of vertical deformation, but their interpretation remains difficult where chronological control is limited and where coastal surfaces have been modified by erosion, deposition, karstification, or human activity. This study presents new Real-Time Kinematic Global Navigation Satellite System (RTK-GNSS) topographic profiles from four coastal sites: Pantai Ajah, Kalijali, Kulon Progo, and Wingko. The profiles were measured from the beachward side toward the landward side and were used to identify terrace treads, risers, slope breaks, residual topographic highs, and possible raised coastal platforms. These field data are integrated with published information on Holocene sea-level change, marine terraces, coastal uplift, and forearc deformation along the southern Java margin. The RTK profiles show variable terrace morphology between sites. Pantai Ajah preserves a prominent riser and a probable terrace tread at approximately 7–8.5 m elevation. Kalijali records a lower terrace-like surface at approximately 4–5 m, an upper surface at approximately 7–9 m, and a higher local topographic high near 12–13 m. Kulon Progo shows a subdued low-elevation raised coastal surface, while Wingko contains a distinct slope break at approximately 1450–1500 m from the beachward end and a broad landward surface at approximately 5–6.5 m elevation. The profiles suggest two tentative morphostratigraphic terrace groups: a lower group at approximately 4–6.5 m and an upper group at approximately 7–9 m. Higher local peaks, including the 12–13 m high at Kalijali and comparable elevated points at other sites, may represent remnants of older or more strongly uplifted coastal features. One possible interpretation is that some of these higher surfaces originated near the mid-Holocene sea-level highstand, when relative sea level in parts of Indonesia and Sundaland was higher than present, and were subsequently uplifted to different elevations according to local uplift rates. However, this hypothesis requires direct chronological and sedimentological confirmation. The raised terrace ridges and topographic highs may also act as partial natural barriers that reduce tsunami flow penetration inland, although they should not be treated as complete protection. Overall, RTK profiling provides a useful field method for recognizing coastal terrace morphology and identifying priority sites for future dating, tsunami-inundation modelling, and coastal-hazard planning.

Abstract: Paediatric rhinosinusitis (RS), particularly chronic rhinosinusitis (CRS), is a common inflammatory condition with a significant impact on quality of life and a well-recognized association with asthma within the framework of united airway disease. This review aims to evaluate the impact of RS on asthma control in children and explore its role as a modifiable determinant. Mechanistically, RS and asthma share key pathophysiological features, including type 2 inflammation, epithelial barrier dysfunction, and airway microbiome dysbiosis, supporting the concept of a unified inflammatory process across the respiratory tract. Clinically, epidemiological data demonstrate a high prevalence of coexisting RS and asthma, with consistent associations with poorer asthma control, increased disease severity, and higher exacerbation burden, even in cases of subclinical sinonasal inflammation. Emerging evidence suggests that appropriate management of CRS, including medical therapy and, in selected cases, surgical intervention, may improve asthma outcomes such as symptom control and lung function. However, the current evidence base remains limited, with a predominance of small, heterogeneous, and observational studies. RS therefore represents a potentially treatable trait in paediatric asthma, warranting systematic evaluation in children with difficult-to-treat disease. Further prospective and interventional studies are needed to clarify causality and define its impact on long-term outcomes.

Abstract: In the healthcare sector, burnout has become a critical concern due to the combination of high job demands and sustained emotional strain. The present study aims to analyze whether emotional intelligence moderates the relationship between psychosocial risk factors, namely working values conflict and burnout among healthcare professionals. A cross-sectional online survey was conducted among 205 healthcare professionals. Measurement instruments included the Burnout Assessment Tool (BAT-23) to assess burnout dimensions; the Health and Work Survey (ERPS_INSAT) to evaluate psychosocial risk factors; and the Wong and Law Emotional Intelligence Scale (WLEIS-P) to assess emotional intelligence. A moderation analysis using the PROCESS macro (model 1) was conducted to examine whether emotional intelligence moderates the relationship between psychosocial risk, working values factor, and burnout among healthcare professionals. The results show that the psychosocial risk-work values dimension was a significant positive predictor of burnout (total scale: B = 0.27, p < 0.001; Exhaustion: B = 0.33, p < 0.001; Mental distance: B = 0.32, p < 0.001; cognitive Impairment: B = 0.14, p < 0.001; emotional Impairment: B = 0.30, p < 0.001), indicating that higher perceived risk was associated with higher burnout symptoms. Emotional intelligence did not significantly predict burnout on its own (total scale: B = 0.07, p > 0.05; Exhaustion: B = 0.09, p > 0.05; Mental Distance: B = 0.11, p > 0.05; Cognitive Impairment: B = 0.11, p > 0.05; Emotional Impairment: B = -0.04, p > 0.05). The interaction term (psychosocial risk: work values × emotional intelligence) was not significant, suggesting that emotional intelligence does not moderate the relationship between working values and burnout. These findings highlight the central role of psychosocial risk factors in the development of burnout among healthcare professionals, but emotional intelligence does not seem to have a protective effect against burnout. Such findings point to the need for organizational interventions that reduce workplace risks and demand a more in-depth analysis of organizational context determinants, with particular attention to the impact of working values conflict as a critical driver of burnout.

Abstract: Using liquid hydrogen (LH₂) in aviation creates a difficult design problem for cryogenic storage systems, especially in the early design stage. At that point, environmental, economic, and technical aspects need to be considered together. Here, a sustainability assessment framework is presented for LH₂ aircraft storage tank configurations from a life-cycle perspective. The study includes 24 design alternatives. These are obtained by changing the material combinations of the main structural components, while the overall tank architecture is kept unchanged. The environmental and economic dimensions are assessed through Life Cycle Assessment (LCA) and Life Cycle Costing (LCC), whereas the technical dimension is represented by system mass. Since the relative importance of the criteria is usually not fixed at this stage, Unweighted TOPSIS (UW-TOPSIS) is first used to examine the alternatives under different weighting scenarios. The most competitive solutions are then re-evaluated by a standardised TOPSIS variant (vector-normalised weights, Z-standardised distances) with objective weighting methods. The results show that the configurations based entirely on Al2219-T8 for the main structural components remain top-ranked and more stable under the examined scenarios, whereas mixed-material configurations are more sensitive to changes in weighting assumptions. In this way, the exploratory stage is kept separate from the later weighting stage, where the weights are computed from the decision matrix of the reduced set. This is suitable for early aerospace design, where subjective preferences are often not yet available.

Abstract: The aim of this paper is to show that Cantor's absolute infinite Ω can consistently be proven to exist by introducing a consistent class theory NBG_Ω that is sufficiently strong. Gödel's first incompleteness theorem asserts that an axiomatic theory cannot be simultaneously (1) arithmetical, (2) axiomatizable, (3) consistent, (4) complete, and (5) countable. When axiomatizability is generalized to non-countable theories, a self-sufficient theory can be defined as possessing the first four of these five properties. An Ω-expanded theory is a countable theory that is extended with Ω-many omni-independent axioms, where Ω is axiomatized as a class cardinality so large that it cannot be proven to exist as a set cardinality in any Ω-consistent first-order set theory. ZFC_Ω, an Ω-expansion of ZFC, is subsequently shown to be self-sufficient. NBG_Ω, which extends ZFC_Ω to the class level, can then consistently prove the existence of Ω.

Abstract: The increasing demand for higher efficiency and lower emissions in aircraft gas turbines motivates investigation of alternative thermodynamic cycle architectures. This study assesses the performance and NOx emission behavior of a triple-spool, separate-exhaust turbofan engine equipped with an interstage turbine burner (ITB). A baseline engine representative of the RB211 Trent 892 is first modeled and verified against publicly available takeoff reference data. The cycle is then modified by introducing an isobaric secondary combustion process between the high-pressure and intermediate-pressure turbines. The effects of fan pressure ratio, bypass ratio, overall pressure ratio, high-pressure turbine inlet temperature, and ITB exit temperature are examined using two-parameter response-surface sweeps. Main-combustor NOx is estimated using an RQL-type cycle correlation, while the ITB contribution is represented using an engineering source–sink model accounting for new NOx formation and partial reburning of upstream NOx. The baseline model predicts specific thrust, TSFC, and EINOx within ±8% of reference values. At a selected ITB operating point, specific thrust increases by 1.98%, TSFC increases by 9.84%, thermal efficiency decreases by 2.56%, and the adopted engineering source–sink model predicts a 20.03% reduction in fuel-flow-weighted EINOx.

Abstract: Objectives: To evaluate the impact of Point-of-Care Ultrasound (POCUS) performed by family physicians on the management of abdominal pain in the emergency department, assessing its effect on length of stay, performance of complementary diagnostic tests, diagnostic concordance, and patient satisfaction. Methods: Quasi-experimental pilot study with a control group conducted in a hospital emergency department. A total of 222 adult patients with abdominal pain were included and allocated according to the attending professional (with or without ultrasound training). Clinical, care-related, and patient-satisfaction variables (SERVPERF questionnaire) were analyzed. Non-parametric statistical tests were used, and multiple linear regression analyses were performed. Results: The POCUS group showed a shorter length of stay (3.46 vs. 4.41 hours; p=0.022) and a lower number of plain radiographies (16.8% vs. 69.9%; p<0.001) and CT scans (p=0.034). Diagnostic concordance was significantly higher in the experimental group (99.2% vs. 75.7%; p<0.001). Overall satisfaction with received care was also higher in the intervention group (p<0.001), with significant differences observed across all evaluated dimensions. The multivariate model explained 26.6% of the variability, with patient satisfaction emerging as a positive predictor. Conclusions: POCUS improves the quality of care in emergency departments by reducing length of stay and the use of complementary diagnostic tests while increasing diagnostic accuracy and patient satisfaction. Its implementation can be considered an effective and potentially cost-effective strategy; however, further studies with greater methodological robustness are required to validate the development of standardized composite indexes.

Abstract: Traditional reservoir flood control operations in China have long relied on a fixed Flood Limited Water Level (FLWL), which frequently results in the underutilization of water resources during flood seasons. Dynamic FLWL regulation and joint reservoir operation have emerged as core strategies to optimize floodwater resource utilization while ensuring flood control safety. However, these approaches typically treat the flood control storage capacity of individual reservoirs as fixed constraints, failing to consider the potential for reallocating this capacity within a cascade reservoir system. This study explores the concept of “equivalent utilization of flood control storage capacity” among cascade reservoirs. Focusing on the four major reservoirs (Wudongde, Baihetan, Xiluodu, and Xiangjiaba) in the lower reaches of the Jinsha River, a methodology for analyzing the equivalent index of their flood control storage capacity is established. The core of this methodology involves a two-round scheduling simulation under various design flood scenarios. The first round of simulation adheres to standard operating rules, while the second round allows upstream reservoirs to retain additional flood volume—with downstream reservoirs correspondingly reducing their outflow—on the premise that downstream safety targets are satisfied. The equivalent index is defined as the ratio of the reduced storage capacity utilized downstream to the additional storage capacity utilized upstream. Nine design flood scenarios (covering three typical years with 1%, 2%, and 5% exceedance probabilities) for flood control in the Sichuan-Chongqing reach were analyzed, with the tightly coupled Wudongde-Baihetan and Xiluodu-Xiangjiaba reservoir pairs treated as two integrated units. The results indicate that the equivalent indices between these two reservoir groups range from 0.96 to 0.999, demonstrating near-perfect functional interchangeability of their flood control storage capacities for the specified research objective. For practical engineering application, a value of 0.96 is recommended as the lower-bound equivalent index. This study provides a methodological framework and specific index to support the dynamic, coordinated, and more efficient utilization of flood control storage capacity in large-scale cascade reservoir systems.

Abstract: Natural lateral particle segregation commonly occurs during the hydraulic deposition of slurry and thickened tailings in surface tailings storage facilities (TSFs), producing spatial heterogeneity in physical, hydrogeotechnical, and mineralogical properties, as well as in the water table. In sulfide-rich tailings, such heterogeneity complicates the design of reclamation cover systems, which are themselves affected by it. This study investigates the impact of physical and rheological properties of hard-rock mine tailings slurries on their segregation under hydrodynamic conditions. It proposes a multiparametric equation for the segregation index (SI) based on Buckingham's π-theorem. For this purpose, six flume experiments were conducted using tailings with initial solids mass concentrations of 63%, 66%, and 69% at slopes of 0.5% and 1%. Results revealed strong exponential correlations (R² > 0.95) between SI and tailings' physical properties (solids concentration, bulk density) as well as rheological parameters (Herschel–Bulkley yield stress and flow index, Cross infinite dynamic viscosity). The SI equation was developed using MATLAB nonlinear least-squares optimization with a trust-region reflective algorithm. Using an SI threshold of 0.05 to define non-segregating behavior, the proposed model can predict segregation tendencies as a function of tailings properties and slope conditions. Further laboratory and field investigations are needed to validate and generalize the criterion.

Abstract: Herein, Diacetylcurcumin (DAC), a derivative of curcumin, was synthesised, and two new polymorphs (monoclinic and triclinic) are reported in addition to the previously known polymorph (P2₁). Solid-state NMR (CP-MAS) and X-ray studies allowed the unambiguous authentication of the elusive polymorph 2 (canoe-shaped, P2₁/n) and the concomitant polymorph 3 (elliptical-shaped, P-1). We demonstrate that morphological crystal analysis under a microscope, in conjunction with ATR-IR, is a rapid and inexpensive technique for exploring the polymorphic landscape of curcuminoids. This discovery highlights the ongoing progress in curcumin derivative research and should inspire fellow chemists and materials scientists to further explore it.

Abstract: This paper assumes that a thermodynamic system can be composed purely of coherence and information, and constructs a working model on that basis. We derive operational parameters for such systems using definitions of the Certainty Equation, semantic entropy, semantic temperature, and formulate five laws and three modes of coherence and information systems. This analysis is then compared to the features of black holes.

Abstract: To address the challenges of unauthorised drone flights in campus low‑altitude security—where traditional detection equipment is costly and ineffective at night—this paper designs a lightweight, all‑weather drone detection and early‑warning system based on a Raspberry Pi edge computing platform and visible/infrared dual‑spectrum fusion. The system uses an IMX219‑77IR infrared camera that automatically switches imaging modes according to ambient brightness, achieving day‑and‑night continuous perception. A YOLOv8n model is compressed to 10 MB via channel pruning and knowledge distillation, reaching an inference speed of 85.2 ms/frame on the Raspberry Pi. A self‑built campus drone dataset of 2,000 images (1,600 open‑source + 400 self‑collected) yields 97.16% precision, 93.79% recall, and 97.27% mAP50. A Flask backend and web map interface provide real‑time alerts by polling every 2 seconds. Total hardware cost is below 1,500 yuan, more than 70% lower than traditional systems. Field tests (5–50 m) show daytime confidence >0.9, nighttime infrared confidence ≈0.88, false negative rate <8%, false positive rate <5%, and stable continuous operation. The project has won a university‑level competition second prize and its software copyright application is under review, demonstrating strong practical value and promotion potential.