Search Results (1,144)

EFL vocabulary learning is an emotionally rich process, particularly in video-based instruction where learners encounter multiple affective cues. This mixed-methods study examined whether instructors’ emotional expressions and idiom valence influence EFL learners’ idiom acquisition and learning experience. A 3 × 3 within-subjects experimental design was adopted, manipulating instructor emotion (positive, neutral, and negative) and idiom valence (positive, neutral, and negative). Forty-five undergraduate EFL learners from a university in China participated in the study. After viewing each video, participants completed an immediate idiom recall test, a perceived learning experience scale, the self-assessment Manikin scale, a cognitive load scale and an open-ended questionnaire. The results showed that neither instructor emotion nor idiom valence had a significant main effect on immediate idiom recall. However, a significant interaction effect was observed, although follow-up comparisons did not identify a robust advantage for any specific congruent or incongruent condition. These findings provide implications for the emotional design of instructional videos in EFL vocabulary learning. Full article

To investigate the effects of different lactic acid bacteria additives on the bacterial community structure of silage prepared from fresh corn processing by-products, three treatment groups, including a non-inoculated control group (C), a Lactiplantibacillus plantarum-supplemented group (P), and a Lentilactobacillus buchneri-supplemented group (B), were set up in this study. Fermentation quality and bacterial community dynamics were measured on days 1, 3, 7, 15, and 45 of ensiling, while nutritional components, in vitro digestibility, and aerobic stability of the silage were systematically analyzed at the end of the 45-day ensiling period. The results showed that pH and ammonia nitrogen (NH₃-N) contents in all groups gradually decreased with the prolongation of ensiling duration. Lactic acid (LA) and acetic acid (AA) contents increased progressively over the ensiling period, and their responses were significantly influenced by additives × day interactions (p < 0.05), with group P showing higher levels than group B on days 15 and 45. Propionic acid (PA) and butyric acid (BA) were not detected in any group. After 45 days of ensiling, the organic matter (OM) content in group B was significantly higher than that in the other two groups (p < 0.05). Compared with group C, the acid detergent lignin (ADL) content was significantly lower in both groups P and B (p < 0.05). Group P presented significantly higher in vitro organic matter digestibility (IVOMD), in vitro crude protein digestibility (IVCPD), and in vitro neutral detergent fiber digestibility (IVNDFD) than groups C and B (p < 0.05). In contrast, group B showed greater in vitro dry matter digestibility (IVDMD) than group P (p < 0.05). All experimental groups maintained silage aerobic stability exceeding 72 h, with no significant differences detected among treatments. The bacterial community of sweet corn by-product silage changed substantially after ensiling. Bacterial community diversity gradually decreased as ensiling proceeded. At 45 days of ensiling, the dominant bacterial genera shifted from Weissella, Levilactobacillus, and Pediococcus to Levilactobacillus and Lentilactobacillus. On days 1, 7, and 15 of ensiling, differences in the relative abundances of specific bacterial taxa were observed among the three groups. At specific ensiling time points, inoculation with L. plantarum was associated with higher relative abundances of most dominant bacterial taxa in sweet corn by-product silage than inoculation with L. buchneri. Full article

Pedestrian safety at urban intersections remains a major challenge in Intelligent Transportation Systems (ITSs). This study investigates whether crossing intention can be reliably inferred directly from temporal body-pose dynamics to drive real-time collision warnings on embedded edge platforms. Existing vision-based approaches that rely primarily on bounding-box proximity or scene-level spatial grids are often prone to false alarms in complex urban environments with motorcycles, stationary pedestrians, and background clutter. To overcome these limitations, we propose an end-to-end framework consisting of four sequential processing stages: (1) a perception layer integrating YOLOv8s, ByteTrack, a displacement filter, and rider suppression to generate reliable pedestrian trajectories; (2) a skeleton extraction layer utilizing YOLOv8s-pose to construct temporal sequences of 17 anatomical keypoints; (3) an ultra-lightweight Skeleton Graph Convolutional Network (SkeletonGCN, comprising 33.8 K parameters, <0.2 MB) that models body-joint kinematics and temporal motion dynamics; and (4) an image-space Time-to-Collision (TTC) risk-fusion module. While this fusion approach avoids explicit geometric camera calibration, it still relies on predefined scene-profile parameters and image-space motion assumptions. Furthermore, while the intention classifier is quantitatively evaluated, the risk-fusion module is procedurally defined, and its resulting four-level collision warnings are demonstrated operationally rather than validated against ground-truth hazard annotations. Evaluated on 49,948 valid sequences from the JAAD and PIE benchmark datasets under a strict video-level partitioning protocol, the unified SkeletonGCN achieves a macro-F1 score of 0.717 (with per-scene subset macro-F1 scores of 0.761 on JAAD/PIE urban and 0.895 on intersections), significantly outperforming baseline models. When deployed on an NVIDIA Jetson Orin NX edge device using TensorRT FP16, the full pipeline achieves an instrumented latency of 70.7 ms per frame (~14 fps) and a sustained wall-clock throughput of 7.4 fps on real-world urban dashcam video. System limitations include sensitivity to 2D printed human imagery and reduced prediction reliability under low-light nighttime conditions. Full article

Conventional vertical-axis wind turbines (VAWTs) rely on rigid blades that are heavy and require expensive materials and manufacturing processes to withstand aerodynamic loads. This study explores flexible sail-type blades as a low-cost alternative capable of achieving comparable performance. Small-scale sail-type VAWTs employing blades made out of fabric similar to maritime sails were designed, fabricated, and tested in a wind tunnel. One edge of each sail was hemmed to a vertical rod, while the opposite edge was secured at its end to top and bottom end disks. The turbine radius and height were fixed at 19 cm and 21 cm, respectively. Experiments were conducted in a wind tunnel and measurements included wind speed, turbine rotational speed, and the generator current and voltage. A total of 36 configurations were evaluated by varying four design parameters: number of sails (2, 3, and 6), chord length (10 cm and 15 cm), pitch angle (0°, 15°, 30°, and 45°), and blade shape (rectangular or trapezoidal). Performance was assessed using power output (P), power coefficient (Cp), and tip-speed ratio (TSR). Experimental results were validated by comparison with test data for a conventional rigid-bladed VAWT. The best-performing configuration was a three-sail turbine with rectangular blades, 15 cm chord length, and 0° pitch angle, producing 547 mW at 177 RPM, corresponding to a Cp of 0.053 at a TSR of 0.59. Numerical predictions using state-of-the-art codes were used to further assess sail-type VAWT performance. Under the test conditions considered, the sail-type VAWT not only showed better performance but also comparable self-starting capability compared to a rigid-bladed VAWT. However, due to the maximum wind speed limitation of the test facility, the complete Cp-TSR characteristic curve could not be determined. Moreover, preliminary estimates indicate a potential 24% reduction in total turbine capital cost and an overall 30% reduction in turbine mass of sail-type turbines over conventional HAWTs, demonstrating that flexible sail blades are a promising low-cost option that needs further investigation to realize their full potential. Full article

This paper presents the design and experimental characterization of a distributed ESP32/MicroPython edge–cloud sensing system with packet-level latency decomposition. Sensor nodes transmit periodic telemetry to an ESP32 gateway over ESP-NOW; the gateway appends reception and MQTT-publication timestamps and forwards records through a local Mosquitto bridge v2.1.2, EMQX Cloud v5, Telegraf v1.36.0, and InfluxDB Cloud Serverless (Storage Engine Version 3). A three-probe two-way gateway-referenced synchronization procedure provides corrected sender timestamps while exposing an interval-based synchronization-uncertainty diagnostic. The bridge-assisted campaign comprised three independent 30 min repetitions with one, three, and five active nodes. Across runs, mean gateway-referenced node-to-gateway latency was 23.17 ± 0.13 ms, 24.13 ± 0.12 ms, and 24.84 ± 0.47 ms, respectively; the corresponding p95 values were 28 ms, 33 ms, and 37–38 ms. Mean gateway-processing latency remained nearly unchanged at 13.31–13.46 ms. Exact full-run database-visible PDR was 100% in all one-node runs, 99.28–99.88% in the three-node runs, and 96.75–97.05% in the five-node runs. Independent GPIO/oscilloscope validation showed a reproducible positive software-to-hardware difference of 12.132 ± 1.819 ms across run means, so the local metric is interpreted as a gateway-referenced application-level delivery metric rather than unbiased physical one-way radio latency. Relative to aggregate end-to-end reporting, the instrumentation separates local, gateway, and downstream ingestion contributions rather than claiming a universally faster transport method. Quantitative performance and scaling claims are confined to the evaluated bridge-assisted configuration and controlled indoor periodic workload. Full article

Rolling element bearings are critical components in rotating machinery, yet assessing and predicting their reliability under heavy industrial noise remains challenging. Existing methods suffer from three major limitations: (1) single-channel signal processing and single-scale indicators lack robustness against non-stationary noise; (2) classical multi-channel decomposition methods, such as multivariate variational mode decomposition (MVMD), rely on empirical parameter tuning, which frequently leads to over- or under-decomposition; and (3) monolithic deep architectures and homogeneous ensemble models suffer from prediction drift and generalization bottlenecks during long-term temporal extrapolation. To address these issues, this paper introduces an automated framework that combines adaptive multi-channel signal purification with a heterogeneous stacking ensemble (HeteroStack-LR). Unlike conventional MVMD pipelines that fix $[K,\alpha ]$ empirically, the Sequoia Optimization Algorithm (SOA) autonomously determines the globally optimal configuration, achieving a mean SNR of $-2.08\mathrm{dB}$—a $1.88$ to $2.50\mathrm{dB}$ improvement over standard VMD/MVMD baselines—along with up to a $37.1\%$ reduction in computation time. Rather than relying on conventional single-metric intrinsic mode function (IMF) selection, we construct a multi-domain hybrid index integrating the Fault Correlation Factor, energy ratio, and refined composite multiscale dispersion entropy (RCMDE) to robustly identify noise-resistant components, thereby enhancing denoising quality by $22.4\%$ to $32.2\%$ over single-scale criteria. Furthermore, contrasting with linear PCA-based reduction, Diffusive Topology Neighbor Embedding (D-TNE) effectively preserves the nonlinear manifold structure of degradation trajectories in a low-dimensional space. Finally, a heterogeneous stacked ensemble featuring an out-of-fold (OOF) leakage-prevention strategy and a logistic regression meta-learner is designed to suppress prediction drift while avoiding the over-parameterization typical of deep architectures. Experimental results across four bearing datasets demonstrate that HeteroStack-LR achieves a minimal MAE of $0.063$ with a variance of ≤±0.002, outperforming state-of-the-art deep architectures (such as TCN, CNN-LSTM, BiLSTM-Attention, and Transformer) as well as classical baselines (Bi-LSTM, CNN, and LSSVM). Ablation studies confirm that removing SOA and MVMD degrades MAE by $12.7\%$ and $19.0\%$, respectively, validating that the framework’s strength stems not from any isolated module, but from the end-to-end synergistic integration of signal purification and reliability prediction. Full article

The textile, apparel, and fashion (TAF) industry generates significant environmental burdens across its entire supply chain. As the global textile recycling market expands, a systematic framework for classifying exhibitors and assessing their sustainability-related characteristics is increasingly needed. This study analyzes exhibitors at the 2026 Textiles Recycling Expo USA, the first specialized textile recycling exhibition in North America, to develop an exploratory sustainability assessment framework. Using qualitative content analysis, 78 exhibiting companies were categorized into four functional groups: (1) Hard-tech Infrastructure, (2) Chemical & Material Innovation, (3) Logistics & Waste Management, and (4) Knowledge & Support Services. Based on a review of sustainability assessment literature in the TAF industry, a four-dimensional framework was developed, encompassing Material Renewability, Process Sustainability, End-of-Life Options, and Technical & Digital Attributes. For a preliminary pilot application, 11 companies were purposively selected and evaluated by eight experts using defined key performance indicators (KPIs). The total scores ranged from 10.3 to 16.3 out of 20, with ESO RECYCLING Società Benefit receiving the highest overall score (16.3), followed by MARGASA (15.8). Across the evaluated companies, Technical & Digital Attributes generally showed relatively lower scores than the other dimensions, indicating comparatively limited publicly evidenced digital traceability capabilities. These findings demonstrate the preliminary applicability of the proposed framework for characterizing heterogeneous exhibitors while highlighting the need for further refinement and validation using larger and more diverse samples. Full article

Cemented paste backfill (CPB) is widely used in mining, but its high brittleness, low toughness, and limited ductility can cause it to crack and spall, or even damage the overall structure, thereby limiting its application in deep underground mine excavations. To this end, this study investigates the damage and failure mechanisms of cemented gangue–rubber paste backfill (CGRPB) and analyses its energy evolution characteristics. The aims are to: (i) assess the CGRPB mechanical properties, i.e., toughness, ductility, and brittleness due to incorporating rubber; (ii) analyze the fracture propagation process of CGRPB from an energy evolution perspective. Therefore, monotonic and cyclic compression tests were conducted on CGRPB samples containing 0%, 5%, and 10% recycled rubber powder. This study focuses on analyzing compressive strength, failure modes, stress–strain responses, energy evolution and the damage evolution process. Key findings include: (1)the effect of rubber incorporation on strength is dosage- and curing-age-dependent; a moderate rubber content (5%) improves early-age strength, whereas excessive rubber addition reduces strength due to increased porosity and weakened load-bearing capacity; (2) samples with rubber significantly reduce the length, number, and width of cracks, achieving better structural integrity; (3) introducing rubber improves the pre-peak deformation capacity of the samples; (4) the strain growth magnitude is positively correlated with the rubber content, enhancing their toughness and ductility; (5) adding rubber effectively reduces the damage propagation rate within the sample; (6) under loading, rubber elastic deformation in samples dissipates energy, which describes the approximately linear energy storage and dissipation trend; (7) among the investigated rubber contents, 5% rubber incorporation achieved a favorable balance between mechanical strength, toughness, and ductility. Full article

Virtual reality (VR) applications are often designed as episodic experiences, with limited support for persistence, longitudinal revisitation, and structured integration of interaction data across sessions. This paper presents a persistent multi-user VR garden architecture that combines snapshot-based state restoration, structured event, movement, and transcript records, an asymmetric owner–visitor workflow, cloud-mediated speech transcription, and deferred AI-supported synthesis. The architecture separates the current spatial configuration of the environment from the interaction traces through which it evolves, supporting repeated access, state restoration, historical consultation, and post-hoc processing. A controlled technical validation using synthetic or researcher-generated inputs was conducted through Unity Editor/backend tests and on Meta Quest 3 hardware. Persistence was evaluated at 20, 100, and 250 objects in the Editor and at 1, 50, and 100 objects on Quest, with two Quest replicas per load. Application-level visitor restrictions were verified across nine prohibited write operations. A frozen production speech-to-text corpus completed 40/40 requests with a micro-averaged word error rate of 9.50% and a median end-to-end latency of 1.619 s. The deferred AI pipeline was additionally verified as a functioning technical integration, while limitations in semantic-detail preservation were observed. The results support the implementation-level feasibility of the proposed persistent and traceable VR architecture. The present evaluation does not establish backend-level authorization guarantees, general AI or NPC-grounding performance, user outcomes, or clinical effectiveness. Full article

Due to system interactions, the formation and maintenance of dual squall lines are more complex than for single squall lines. In this study, we use upper-air soundings, ERA5 reanalysis data, high-density surface automatic weather station observations, and Doppler radar data to analyze a dual squall-line system that occurred over North China on 13 June 2022 under the Northeast China Cold Vortex. We focus on the differences between the two squall lines in mesoscale environments, convective triggering mechanisms, and maintenance processes. The main results are as follows: (1) The dual squall-line event occurred in different sectors of the Northeast China Cold Vortex, with both lines exhibiting a “dry-cold aloft, warm-moist below” stratification. However, significant spatiotemporal differences in mesoscale thermodynamic and dynamic conditions across Hebei and Shandong provinces led to distinct evolutionary pathways between the two squall lines. (2) Squall Line 1 (SL1) was triggered by the superposition of cold-pool outflow from convective cells over the Bohai Bay and convergence lines associated with surface cyclonic circulations. Squall Line 2 (SL2) was triggered by the thermal instability in the overlapping region of the temperature and dew-point fronts on the eastern slope of the Taihang Mountains, in conjunction with topographic uplift driven by the easterly flow. (3) This case study shows that squall-line maintenance depends not only on environmental CAPE and vertical wind shear but may also be closely related to the coordinated interplay between local thermal conditions and low-level shear. SL1, situated in a high-CAPE, low-LCL warm-moist environment, experienced relatively weak low-level shear; however, the ratio of cold-pool propagation speed to low-level shear remained near the RKW optimum, favoring persistence. Additionally, cold-pool spreading on the southern flank triggered new convection that merged into the southern end of the squall line, enhancing the cold pool via evaporative cooling and further promoting longevity. By contrast, SL2 displayed a pronounced north–south disparity: the northern segment failed to satisfy RKW balance due to insufficient cold-pool propagation relative to shear, leading to rapid echo dissipation; the southern segment, featuring an overly strong cold pool and low-CAPE, high-LCL conditions, inhibited deep convection. As a result, SL2 gradually split due to the spatial mismatch of thermodynamic and dynamic conditions along its north–south extent. Full article

The heterogeneous occurrence of coal-seam thickness represents a common geological characteristic in underground mining. Variations in coal thickness can directly alter the instability-failure behavior of coal–rock systems, thereby triggering various dynamic disasters. Therefore, revealing failure and disaster-inducing mechanisms of coal–rock systems dominated by coal-thickness effects is critical for deep mining engineering design as well as dynamic disaster prevention and control. To this end, uniaxial compression tests combined with acoustic emission (AE) monitoring were performed on coal–rock combinations with different coal thicknesses. The evolution laws of characteristic strengths (uniaxial compressive strength, initiation strength, and damage strength) versus coal thickness were systematically analyzed. Using full-process spatial localization of internal damage derived from absolute AE energy, an instability evolution model for coal–rock combinations was established. Furthermore, intrinsic disaster-inducing mechanisms governing coal–rock system instability under coal-thickness regulation were summarized, with corresponding engineering prevention-control suggestions put forward. The results show that: (1) UCS, initiation strength, and damage strength of specimens exhibit a nonlinear negative correlation with coal thickness. Initiation strength and damage strength account for approximately 50% and 75% of UCS, respectively; (2) Increasing coal thickness weakens the confinement effect of upper- and lower-sandstone, which shifts the dominant failure zone gradually from coal–rock interfaces to coal interiors. Meanwhile, internal energy accumulation-release processes of combinations present staged evolution characteristics; (3) Different coal thicknesses produce distinct disaster-evolution paths for coal–rock systems. Larger coal thickness corresponds to higher risks of high-energy dynamic disasters. Accordingly, a differentiated hierarchical prevention strategy of “thin protection, medium pressure relief, and thick control” was proposed. These findings provide a theoretical basis for mine engineering design and dynamic disaster prevention-control under dominant coal-thickness effects. Full article

Background/Objectives: Temporomandibular joint (TMJ) abnormalities are a common finding in the general adult population, but they have been rarely investigated in people receiving chronic dialysis. The TMJ is a unique synovial joint that can be altered by either inflammatory or degenerative processes, both of which are amenable to ultrasound (US) assessment. The aim of this study was to describe the pattern of TMJ involvement in a cohort of hemodialysis patients using TMJ ultrasound (TMJ US). Methods: This cross-sectional, single-center study evaluated the clinical utility of TMJ US to detect inflammatory and degenerative changes in patients undergoing chronic hemodialysis and was carried out between June and December 2025. Demographic data, dialysis vintage, the Controlling Nutritional Status (CONUT) score and the Charlson Comorbidity Index (CCI) were collected and related to TMJ US findings and to bone-metabolism biomarkers (calcium, phosphate, parathyroid hormone). Results: 100 hemodialysis patients were included (65 male, 65%; mean age 71.6 ± 13.1 years). Comorbidity and undernutrition were frequent findings (CCI ≥ 4 in 63%; CONUT ≥ 3 in 68%). TMJ US revealed that degenerative indicators were more common than inflammatory ones: calcifications (36%), condylar irregularities (33%) and enthesophytes (25%) were the most prevalent degenerative findings, whereas joint effusion (16%), synovial hypertrophy (15%) and cartilage changes (15%) were the leading inflammatory findings; a positive power Doppler signal was rare (1%). Cortical/condylar irregularity was significantly more prevalent in patients with a dialysis vintage of less than 30 months (43.8% vs. 23.1%, p = 0.034), who also had lower serum calcium levels. No other TMJ US finding showed a significant association with age, sex, comorbidity burden, nutritional status, or bone-metabolism parameters on univariate. Conclusions: TMJ US demonstrated a substantial burden of subclinical degenerative and, to a lesser extent, inflammatory TMJ findings in this cohort of hemodialysis patients; none of these abnormalities were spontaneously reported as symptomatic, and their prevalence was independent of age, dialysis vintage, comorbidity, nutritional status and classical bone-metabolism parameters. These findings cannot be directly attributed to the dialysis/uremic environment rather than to age-related degeneration; they support the concept that the TMJ warrants further investigation as a potentially under-recognized target within the broader systemic metabolic derangement of end-stage kidney disease, and suggest that TMJ US is a feasible, accessible technique for detecting subclinical TMJ involvement in this population, pending confirmation of clinical utility in controlled studies. Full article

Top-loading smart vending cabinets require precise transaction-level product detection under strict hardware and deployment constraints. In this paper, “sensorless” refers specifically to the absence of auxiliary product-level sensing hardware, such as RFID tags, weight sensors, shelf load cells, or product-slot instrumentation; the system still uses two camera sensors. Conventional snapshot-difference methods compare only a small number of frames at the beginning and end of a transaction and therefore cannot explicitly represent intermediate product motion, such as pickup, return, inspection, occlusion, and shelf resettling. This paper proposes ZAB-Fusion, a dual-camera edge sensing framework with zone-aware multi-object tracking for sensorless smart vending cabinets. The framework combines a YOLO11-seg and RT-DETR detection ensemble with ByteTrack temporal association, projects product tracks into a three-zone vertical cabinet model, interprets compressed zone sequences using a finite-state event classifier, and integrates camera-specific event streams through an evidence-gated cross-camera fusion rule. The proposed method was evaluated on 220 in-service vending transactions containing 227 ground-truth TAKEN events and 87 RETURNED events across seven product classes. Compared with the snapshot-difference baseline, ZAB-Fusion improved recall from 0.665 to 0.925 and F1-score from 0.780 to 0.944, while maintaining a high precision of 0.963. At the transaction level, exact receipt accuracy increased from 0.645 to 0.900. Runtime analysis on an Intel N100 CPU-only edge device showed an average processing latency of 562 ms per transaction under the selected-frame inference protocol. The zone classification, finite-state event interpretation, and evidence-gated cross-camera fusion stages required only 3 ms in total. These results demonstrate that explicit motion semantics and auditable cross-camera evidence gating can improve sensorless retail transaction level recognition in sensorless smart vending cabinets without adding auxiliary product-level sensing hardware. Full article

Ultraconserved elements (UCEs) are among the most evolutionarily conserved DNA sequences in vertebrate genomes, yet the biological mechanisms underlying their extraordinary conservation remain poorly understood. Using the recently developed dedUCE database comprising 12,813 human UCEs, we performed a comprehensive genome-wide analysis of their distribution relative to protein-coding genes, transcription factor (TF) genes, and long noncoding RNA (lncRNA) genes. UCEs showed a highly non-random genomic organization, with approximately 40% occurring in clusters within 20 kb genomic intervals. Non-KRAB transcription factor genes exhibited a striking sevenfold enrichment of UCEs compared with random expectation, whereas KRAB zinc-finger genes displayed an approximately tenfold depletion. Beyond TFs, UCE-rich genes were predominantly involved in developmental regulation, chromatin remodeling, RNA processing, and embryonic neurogenesis, whereas similarly large UCE-poor genes primarily encoded membrane proteins, ion channels, and synaptic components required for mature neuronal function. UCEs also demonstrated strong positional bias, with approximately fourfold enrichment near the 3′ ends of protein-coding genes but no comparable distribution pattern in lncRNAs. Although lncRNA genes showed only modest overall UCE enrichment, a small subset contained numerous UCEs. These findings demonstrate that UCEs preferentially associate with master developmental regulators rather than downstream neuronal effector genes, providing new insights into the functional organization and evolutionary conservation of the human genome. Full article

Synthetic-aperture radar (SAR) ship detection is a fundamental task in maritime remote sensing, supporting wide-area surveillance, traffic monitoring, and emergency response under all-weather imaging conditions. Existing deep detectors mainly rely on spatial cues such as intensity, shape and context, but structured sea clutter and near-shore interference can still produce ship-like responses, while fine scattering details are weakened by deep downsampling. We address two practical representation limitations: incomplete preservation of shallow high-resolution details, and limited explicit modeling of local directional variation. To this end, we propose HMF-RTMDet, a shallow-neck spatial–frequency fusion detector. A P2 high-resolution path combines $C_2$ features with upsampled $P_3$ semantics. HybridMFBlock then processes the fused feature through a morphology branch and a trainable depthwise branch initialized by fractional Gabor templates, followed by channel-wise fusion. In the reported main HRSID run, HMF-RTMDet improves RTMDet-s from 67.9% to 72.6% in AP_50:95, from 90.2% to 94.2% in AP₅₀, and from 68.2% to 73.4% in AP_s. Across three runs, however, its AP_50:95 is 72.17 ± 0.38%, comparable to the SFS-Conv and MCU-only controls. The evidence therefore identifies the P2 path as the main gain source but does not establish a stable advantage for HybridMFBlock over these controls. On SSDD, overall AP_50:95 remains nearly unchanged and large-target performance decreases, defining an important boundary of the current design. Full article