Search Results (784)

This paper proposes a regional inertia estimation method in power systems using ambient data measured by phasor measurement units (PMUs). The proposed method employs band-pass filtering to suppress the low-frequency influence of mechanical power and to attenuate high-frequency noise and discrepancies between rotor speed and electrical frequency. By utilizing a simple first-order AutoRegressive Moving Average with eXogenous input (ARMAX) model, this process allows the inertia constant to be directly identified. This method requires no prior model order selection, rotor speed estimation, or computation of the rate of change of frequency (RoCoF). The proposed method was validated through simulation on three benchmark systems: the Kundur two-area system, the IEEE Australian simplified 14-generator system, and the IEEE 39-bus system. The method achieved area-level inertia estimates within approximately ±5% error across all test cases, exhibiting consistent performance despite variations in disturbance models and system configurations. The estimation also maintained stable performance with short data windows of a few minutes, demonstrating its suitability for near real-time monitoring applications. Full article

Background/Objectives: The vestibular labyrinth is classically viewed as a sensor of low-frequency head motion—linear acceleration for the otoliths and angular velocity/acceleration for the semicircular canals. However, there is now substantial evidence that air-conducted sound (ACS) can also activate vestibular receptors and afferents in mammals and other vertebrates. This sound sensitivity underlies sound-evoked vestibular-evoked myogenic potentials (VEMPs), sound-induced eye movements, and several clinical phenomena in third-window pathologies. The cellular and biophysical mechanisms by which a pressure wave in the cochlear fluids is transformed into a vestibular neural signal remain incompletely integrated into a single framework. This study aimed to provide a narrative synthesis of how ACS activates the vestibular labyrinth, with emphasis on (1) the anatomical and biophysical specializations of the maculae and cristae, (2) the dual-channel organization of vestibular hair cells and afferents, and (3) the encoding of fast, jerk-rich acoustic transients by irregular, striolar/central afferents. Methods: We integrate experimental evidence from single-unit recordings in animals, in vitro hair cell and calyx physiology, anatomical studies of macular structure, and human clinical data on sound-evoked VEMPs and sound-induced eye movements. Key concepts from vestibular cellular neurophysiology and from the physics of sinusoidal motion (displacement, velocity, acceleration, jerk) are combined into a unified interpretative scheme. Results: ACS transmitted through the middle ear generates pressure waves in the perilymph and endolymph not only in the cochlea but also in vestibular compartments. These waves produce local fluid particle motions and pressure gradients that can deflect hair bundles in selected regions of the otolith maculae and canal cristae. Irregular afferents innervating type I hair cells in the striola (maculae) and central zones (cristae) exhibit phase locking to ACS up to at least 1–2 kHz, with much lower thresholds than regular afferents. Cellular and synaptic specializations—transducer adaptation, low-voltage-activated K⁺ conductances (K_LV), fast quantal and non-quantal transmission, and afferent spike-generator properties—implement effective high-pass filtering and phase lead, making these pathways particularly sensitive to rapid changes in acceleration, i.e., mechanical jerk, rather than to slowly varying displacement or acceleration. Clinically, short-rise-time ACS stimuli (clicks and brief tone bursts) elicit robust cervical and ocular VEMPs with clear thresholds and input–output relationships, reflecting the recruitment of these jerk-sensitive utricular and saccular pathways. Sound-induced eye movements and nystagmus in third-window syndromes similarly reflect abnormally enhanced access of ACS-generated pressure waves to canal and otolith receptors. Conclusions: The vestibular labyrinth does not merely “tolerate” air-conducted sound as a spill-over from cochlear mechanics; it contains a dedicated high-frequency, transient-sensitive channel—dominated by type I hair cells and irregular afferents—that is well suited to encoding jerk-rich acoustic events. We propose that ACS-evoked vestibular responses, including VEMPs, are best interpreted within a dual-channel framework in which (1) regular, extrastriolar/peripheral pathways encode sustained head motion and low-frequency acceleration, while (2) irregular, striolar/central pathways encode fast, sound-driven transients distinguished by high jerk, steep onset, and precise spike timing. Full article

Modern Earth observation combines high spatial resolution, wide swath, and dense temporal sampling, producing image grids and sequences far beyond the regime of standard vision benchmarks. Convolutional networks remain strong baselines but struggle to aggregate kilometre-scale context and long temporal dependencies without heavy tiling and downsampling, while Transformers incur quadratic costs in token count and often rely on aggressive patching or windowing. Recently proposed visual state-space models, typified by Mamba, offer linear-time sequence processing with selective recurrence and have therefore attracted rapid interest in remote sensing. This survey analyses how far that promise is realised in practice. We first review the theoretical substrates of state-space models and the role of scanning and serialization when mapping two- and three-dimensional EO data onto one-dimensional sequences. A taxonomy of scan paths and architectural hybrids is then developed, covering centre-focused and geometry-aware trajectories, CNN– and Transformer–Mamba backbones, and multimodal designs for hyperspectral, multisource fusion, segmentation, detection, restoration, and domain-specific scientific applications. Building on this evidence, we delineate the task regimes in which Mamba is empirically warranted—very long sequences, large tiles, or complex degradations—and those in which simpler operators or conventional attention remain competitive. Finally, we discuss green computing, numerical stability, and reproducibility, and outline directions for physics-informed state-space models and remote-sensing-specific foundation architectures. Overall, the survey argues that Mamba should be used as a targeted, scan-aware component in EO pipelines rather than a drop-in replacement for existing backbones, and aims to provide concrete design principles for future remote sensing research and operational practice. Full article

The North China Plain, a major grain production base in China, is facing the chronic threat of climate-change-induced delays in winter wheat sowing, with late sowing constraining yields by shortening the pre-winter growth period, and soil moisture at sowing potentially serving as a key factor to alleviate late-sowing losses. However, previous studies have mostly independently analyzed the effects of sowing time or water stress, and there is still a lack of systematic quantitative evaluation on how the interaction effects between the two determine long-term yield potential and risk. To fill this gap, this study aims to quantify, in the context of long-term climate change, the independent and interactive effects of different sowing dates and baseline soil moisture on the growth, yield, and production risk of winter wheat in the North China Plain, and to propose regionally adaptive management strategies. We selected three representative stations—Beijing (BJ), Wuqiao (WQ), and Zhengzhou (ZZ)—and, using long-term meteorological data (1981–2025) and field trial data, undertook local calibration and validation of the APSIM-Wheat model. Based on the validated model, we simulated 20 management scenarios comprising four sowing dates and five baseline soil moisture levels to examine the responses of phenology, aboveground dry matter, and yield, and further defined yield-reduction risk probability and expected yield loss indicators to assess long-term production risk. The results show that the APSIM-Wheat model can reliably simulate the winter wheat growing period (RMSE 4.6 days), yield (RMSE 727.1 kg ha⁻¹), and soil moisture dynamics for the North China Plain. Long-term trend analysis indicates that cumulative rainfall and the number of rainy days within the conventional sowing window have risen at all three sites. Delayed sowing leads to substantial yield reductions; specifically, compared with S1, the S4 treatment yields about 6.9%, 16.2%, and 16.0% less at BJ, WQ, and ZZ, respectively. Moreover, increasing the baseline soil moisture can effectively compensate for the losses caused by late sowing, although the effect is regionally heterogeneous. In BJ and WQ, raising the baseline moisture to a high level (P85) continues to promote biomass accumulation, whereas in ZZ this promotion diminishes as growth progresses. The risk assessment indicates that increasing baseline moisture can notably reduce the probability of yield loss; for example, in BJ under S4, elevating the baseline moisture from P45 to P85 can reduce risk from 93.2% to 0%. However, in ZZ, even the optimal management (S1P85) still carries a 22.7% risk of yield reduction, and under late sowing (S4P85) the risk reaches 68.2%, suggesting that moisture management alone cannot fully overcome late-sowing constraints in this region. Optimizing baseline soil moisture management is an effective adaptive strategy to mitigate late-sowing losses in winter wheat across the North China Plain, but the optimal approach must be region-specific: for BJ and WQ, irrigation should raise baseline moisture to high levels (P75-P85); for ZZ, the key lies in ensuring baseline moisture crosses a critical threshold (P65) and should be coupled with cultivar selection and fertilizer management to stabilize yields. The study thus provides a scientific basis for regionally differentiated adaptation of winter wheat in the North China Plain to address climate change and achieve stable production gains. Full article

This work describes the lead optimization of a promising class of antibacterial compounds, derived from a previously reported N-[4-(4-fluorophenoxy)phenyl]-6-(methylsulfonyl)-2,6-diazaspiro [3.4]octane-8-carboxamide (LK1819), through systematic scaffold simplification. A novel series of amide derivatives were designed and synthesized, exploring key structural variations, including the replacement of the diphenyl ether core with a biphenyl system. All compounds were evaluated for in vitro antibacterial activity against the ESKAPE panel of pathogens. The most potent simplified analogs demonstrated exceptional, broad-spectrum activity, with minimum inhibitory concentrations (MICs) that were 10 to 100 times lower than the control antibiotic ciprofloxacin against many strains. Mechanistic studies using a reporter system and enzymatic assays revealed that the compounds do not inhibit protein synthesis but disrupt DNA replication, exhibiting a dose-dependent inhibitory effect on bacterial topoisomerase I and DNA gyrase. The compounds showed moderate toxicity against human cell lines, consistent with their DNA-targeting mechanism, but cytotoxicity assays indicated a sufficient selectivity window. We conclude that scaffold simplification successfully yielded highly potent antibacterial agents with a defined mechanism of action, presenting a promising foundation for further development as antibiotics and potentially as anticancer agents. Full article

Aiming to address the core issue that the current intelligence evaluation for Unmanned Ground Vehicles (UGVs) overly rely on static performance metrics and lack dynamic quantitative characterization of learning internalization capability (LIC), this study proposes a dynamic evaluation framework based on time series analysis. The framework begins by constructing a multidimensional test scenario parameter system and collecting externally observable performance sequence data. It then introduces a sliding window-based slope-standard deviation collaborative analysis technique to achieve unsupervised division of learning phases, from which five core evaluation metrics are extracted to comprehensively quantify the multidimensional dynamic characteristics of LIC in terms of efficiency, stability, and overall effectiveness. Simulation experiments were carried out using UGVs equipped with three types of path-planning algorithms in low-, medium-, and high-difficulty scenarios. Results demonstrate that the proposed algorithm can effectively distinguish multi-dimensional differences in LIC among different UGVs, exhibiting strong discriminative power and interpretability. This study provides a standardized evaluation tool for UGV intelligent selection, algorithm iteration optimization, and training strategy design, and offering significant reference value for the evaluation of the learnability of autonomous driving systems. Full article

The Automatic Identification System (AIS) frequently suffers from data loss and irregular report intervals in real maritime environments, compromising the reliability of downstream navigation, monitoring, and trajectory reconstruction tasks. To address these challenges, we propose BAARTR (Boundary-Aware Adaptive Regression for Kinematically Consistent Vessel Trajectory Reconstruction), a novel kinematically consistent interpolation framework. Operating solely on time, latitude, and longitude inputs, BAARTR explicitly enforces boundary velocities derived from raw AIS data. The framework adaptively selects a velocity-estimation strategy based on the AIS reporting gap: central differencing is applied for short intervals, while a hierarchical cubic velocity regression with a quadratic acceleration constraint is employed for long or irregular gaps to iteratively refine endpoint slopes. These boundary slopes are subsequently incorporated into a clamped quartic interpolation at a 1 s resolution, effectively suppressing overshoots and ensuring velocity continuity across segments. We evaluated BAARTR against Linear, Spline, Hermite, Bezier, Piecewise cubic hermite interpolating polynomial (PCHIP) and Modified akima (Makima) methods using real-world AIS data collected from the Mokpo Port channel, Republic of Korea (2023–2024), across three representative vessels. The experimental results demonstrate that BAARTR achieves superior reconstruction accuracy while maintaining strictly linear time complexity ($\mathit{O}\left(\mathit{N}\right)$). BAARTR consistently achieved the lowest median Root Mean Square Error (RMSE) and the narrowest Interquartile Ranges (IQR), producing visibly smoother and more kinematically plausible paths-especially in high-curvature turns where standard geometric interpolations tend to oscillate. Furthermore, sensitivity analysis shows stable performance with a modest training window (n ≈ 16) and minimal regression iterations (m = 2–3). By reducing reliance on large training datasets, BAARTR offers a lightweight, extensible foundation for post-processing in Maritime Autonomous Surface Ship (MASS) and Vessel Traffic Service (VTS), as well as for accident reconstruction and multi-sensor fusion. Full article

Left ventricular (LV) apical obliteration represents a convergent imaging phenotype arising from diverse cardiac conditions, including thrombotic, hypertrophic, infiltrative, congenital, and neoplastic diseases. These conditions, despite sharing overlapping morphological features, require profoundly different management strategies. In this context, an accurate characterization of the LV apex is a cornerstone point, and can be performed through various techniques. Advances in multimodality imaging have substantially improved diagnostic precision, allowing clinicians to differentiate true obliteration from mimicking conditions such as hypertrabeculation, apical hypertrophy, or subendocardial fibrosis. This review provides a comprehensive overview of the anatomical variability of the LV apex and its implications for imaging interpretation. We appraise the role of echocardiography, including contrast-enhanced and speckle-tracking studies—alongside cardiac magnetic resonance (CMR), computed tomography (CT), and selective nuclear imaging in the evaluation of apical pathology. For each principal cause of apical obliteration—LV thrombus, apical hypertrophic cardiomyopathy, left ventricular non-compaction, endomyocardial fibrosis, cardiac amyloidosis, and intracardiac tumors—we outline key diagnostic clues, imaging red flags, and distinguishing tissue characteristics. Special emphasis is given to the incremental value of CMR for tissue characterization, thrombus detection, and fibrosis mapping, as well as to the interpretative challenges posed by apical foreshortening, near-field artefacts, and suboptimal acoustic windows. A practical, stepwise imaging framework is proposed to guide clinicians through the differential diagnosis of apical obliteration using an integrated multimodality approach. Future directions include the incorporation of 4D flow, advanced mapping techniques, and artificial intelligence-powered analysis to refine apical phenotyping and identify early disease signatures. Recognizing the full spectrum of apical pathology and its imaging manifestations is essential to prevent misdiagnosis, enable timely therapeutic decisions, and improve risk stratification. Full article

Time-Sensitive Networking (TSN), particularly the Time-Aware Shaper (TAS) specified by IEEE 802.1Qbv, is critical for real-time communication in Industrial Sensor Networks (ISNs). However, many TAS scheduling approaches rely on centralized computation and can face scalability bottlenecks in large networks. In addition, global-only schedulers often generate fragmented Gate Control Lists (GCLs) that exceed per-port entry limits on resource-constrained switches, reducing deployability. This paper proposes a two-phase distributed genetic-based algorithm, 2PDGA, for TAS scheduling. Phase I runs a network-level genetic algorithm (GA) to select routing paths and release offsets and construct a conflict-free baseline schedule. Phase II performs per-switch local refinement to merge windows and enforce device-specific GCL caps with lightweight coordination. We evaluate 2PDGA on 1512 configurations (three topologies, 8–20 switches, and guard bands ${\mathsf{\delta }}_{gb}\in \{0, 100, 200\} \mathrm{ns}$). At ${\mathsf{\delta }}_{gb}=0$ ns, 2PDGA achieves 92.9% and 99.8% CAP@8/CAP@16, respectively, compliance while maintaining a median latency of 42.1 $\mathsf{\mu }$s. Phase II reduces the average max-per-port GCL entries by 7.7%. These results indicate improved hardware deployability under strict GCL caps, supporting practical deployment in real-world Industry 4.0 applications. Full article

Plastic streams dominated by polyethylene (PE) including PE HD/MD (High Density/Medium Density) and PE LD/LLD (Low Density/Linear Low Density), polypropylene (PP), and polyvinyl chloride (PVC) across Europe demand a design framework that links synthesis with end of life reactivity, supporting circular economic goals and European Union waste management targets. This work integrates polymerization derived chain architecture and depolymerization mechanisms to guide selective valorization of commercial plastic wastes in the European context. Catalytic topologies such as Bronsted or Lewis acidity, framework aluminum siting, micro and mesoporosity, initiators, and strategies for process termination are evaluated under relevant variables including temperature, heating rate, vapor residence time, and pressure as encountered in industrial practice throughout Europe. The analysis demonstrates that polymer chain architecture constrains reaction pathways and attainable product profiles, while additives, catalyst residues, and contaminants in real waste streams can shift radical populations and observed selectivity under otherwise similar operating windows. For example, strong Bronsted acidity and shape selective micropores favor the formation of C₂ to C₄ olefins and Benzene, Toluene, and Xylene (BTX) aromatics, while weaker acidity and hierarchical porosity help preserve chain length, resulting in paraffinic oils and waxes. Increasing mesopore content shortens contact times and limits undesired secondary cracking. The use of suitable initiators lowers the energy threshold and broadens processing options, whereas diffusion management and surface passivation help reduce catalyst deactivation. In the case of PVC, continuous hydrogen chloride removal and the use of basic or redox co catalysts or ionic liquids reduce the dehydrochlorination temperature and improve fraction purity. Staged dechlorination followed by subsequent residue cracking is essential to obtain high quality output and prevent the release of harmful by products within European Union approved processes. Framing process design as a sequence that connects chain architecture, degradation chemistry, and operating windows supports mechanistically informed selection of catalysts, severity, and residence time, while recognizing that reported selectivity varies strongly with reactor configuration and feed heterogeneity and that focused comparative studies are required to validate quantitative structure to selectivity links. In European post consumer sorting chains, PS and PC are frequently handled as separate fractions or appear in residues with distinct processing routes, therefore they are not included in the polymer set analyzed here. Polystyrene and polycarbonate are outside the scope of this review because they are commonly handled as separate fractions and are typically optimized toward different product slates than the gas, oil, and wax focused pathways emphasized here. Full article

Visual social commerce platforms now mediate much of brand communication and conversion, yet managers still lack clear guidance on how brands and creators should technically design posts that consistently achieve high user engagement under budget and platform constraints. Prior research explains why users engage with brands online, but it mainly focuses on individual motives and message features and largely treats the brand–creator–platform relationship and the post-design process as a black box. Drawing on the Technology Affordance Actualization (TAA) framework—which conceptualizes how platform-provided action possibilities (affordances) are selectively enacted through user practices—we develop a Creator–Content–Timing (CCT) perspective on how brands and creators actualize visibility, interactivity, and commercial collaboration affordances into user engagement outcomes. We analyze 138,713 image–text posts from 100 beauty brands on Xiaohongshu using machine learning, text mining, computer vision, and regression and clustering models. The results show that creator tier, brand status, sponsorship, content cues, and posting time have systematic effects on both engagement intensity and a cost-normalized metric, Int_per_cost (interactions per 1000 CNY of estimated advertising cost). Smaller creators and non-sponsored posts achieve higher engagement per impression and higher Int_per_cost than top-tier creators and sponsored posts; moderate text length, non-exclusive brand mentions, human faces, and specific temporal windows are also associated with superior outcomes. The study extends TAA to a creator–brand–platform context by operationalizing affordance actualization as observable CCT configurations at the post level and provides configuration-level guidance on how brands can align creator selection, content design, and scheduling to improve engagement on visual social commerce platforms. Full article

The selection of data smoothing methods is one of the key steps in extracting land surface phenology parameters from time-series remote sensing data. However, existing studies often use default parameters for denoising the time-series data, neglecting the sensitivity of phenology extraction to different combinations of smoothing parameters. Therefore, this study systematically evaluated three parametric smoothing methods—Savitzky–Golay (SG), Whittaker Smoother (WS), and Harmonic Analysis of Time-Series (HANTS)—and two non-parametric methods—Asymmetric Gaussian (AG) and Double-Logistic (DL)—on the accuracy of Start of Season (SOS) and End of Season (EOS) extraction at eight ground phenology observation sites in Inner Mongolia, based on time-series MOD13Q1- Normalized Difference Vegetation Index data and using the derivative method as the background for phenology parameter extraction at the site scale. The results showed that (1) DL and HANTS yielded similar accuracy for phenology extraction in desert steppe, while parametric smoothing methods outperformed non-parametric methods in phenology simulation in typical and meadow steppe regions. (2) We proposed the optimal phenology parameter combination for different steppe types in Inner Mongolia. For desert steppe, DL or HANTS was recommended. For SOS extraction in typical steppe ecosystems, the WS parameter combination was used. For EOS and phenology in meadow steppe, the HANTS parameter combination yielded better simulation results. (3) In desert and meadow steppes, the window radius in SG contributed more to phenology accuracy than polynomial order. The opposite was true for typical steppe. In WS, the contribution of the differential order to SOS and EOS extraction in desert and typical steppes was higher than that of the smoothing factor. The opposite was observed in meadow steppe. In HANTS, the fitting tolerance error was the key factor controlling phenology extraction accuracy. (4) Based on the optimal phenology extraction scheme, the smallest extraction error occurred in meadow steppe at the site scale. This was followed by typical steppe. Desert steppe showed relatively larger errors. This study overcomes the reliance on default parameters in previous studies and proposes a practical framework for phenology extraction for different grassland ecosystems. The findings provide new empirical evidence for method selection and parameter setting in remote sensing phenology monitoring. Full article

This study aimed to evaluate the comparative predictive efficacy of the SARIMA statistical model and the Prophet machine learning model for forecasting monthly traffic accidents across the 24 provinces of Ecuador, addressing a critical research gap in model selection for geographically and socioeconomically heterogeneous regions. By integrating classical time series modeling with algorithmic decomposition techniques, the research sought to determine whether a universally superior model exists or if predictive performance is inherently context-dependent. Monthly accident data from January 2013 to June 2025 were analyzed using a rolling-window evaluation framework. Model accuracy was assessed through Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) metrics to ensure consistency and comparability across provinces. The results revealed a global tie, with 12 provinces favoring SARIMA and 12 favoring Prophet, indicating the absence of a single dominant model. However, regional patterns of superiority emerged: Prophet achieved exceptional precision in coastal and urban provinces with stationary and high-volume time series—such as Guayas, which recorded the lowest MAPE (4.91%)—while SARIMA outperformed Prophet in the Andean highlands, particularly in non-stationary, medium-to-high-volume provinces such as Tungurahua (MAPE 6.07%) and Pichincha (MAPE 13.38%). Computational instability in MAPE was noted for provinces with extremely low accident counts (e.g., Galápagos, Carchi), though RMSE values remained low, indicating a metric rather than model limitation. Overall, the findings invalidate the notion of a universally optimal model and underscore the necessity of adopting adaptive, region-specific modeling frameworks that account for local geographic, demographic, and structural factors in predictive road safety analytics. Full article

Neuroplasticity-based active movement opens an avenue for functional recovery in post-stroke patients. Active rehabilitation techniques have attracted wide attention based on their abilities to enhance patient involvement, facilitate precise personalized intervention, and provide comprehensive treatment via cross-domain approaches. Emerging evidence suggests that active rehabilitation methods can respond to patients’ motor intentions in real-time and significantly increase motivation and engagement, leading to efficient utilization of critical recovery windows and better rehabilitation outcomes. In this review, we focus on the physiological basis of active rehabilitation, including mechanisms of neuroplasticity, and discuss recent advances in intent detection and feedback devices. We also examine treatment options for different stages of stroke recovery, providing a comprehensive reference for engineers to design optimized rehabilitation techniques and for clinicians to select appropriate rehabilitation protocols. These developments create new opportunities to improve the lives of stroke patients and offer greater hope for their recovery. Full article

Passenger-gathering behavior often triggers safety incidents such as stampedes due to overcrowding, posing significant challenges to public order maintenance and passenger safety. Traditional early warning algorithms for passenger-gathering behavior typically perform only global modeling of image appearance, neglecting the analysis of individual passenger actions in practical 3D physical space, leading to high false-alarm and missed-alarm rates. To address this issue, we decompose the modeling process into two stages: human pose estimation and gathering behavior recognition. Specifically, the pose of each individual in 3D space is first estimated from images, and then fused with global features to complete the early warning. This work focuses on the former stage and aims to develop an accurate and efficient human pose estimation model capable of real-time inference on resource-constrained devices. To this end, we propose a 3D human pose estimation framework that integrates a hybrid spatio-temporal Transformer with three collaborative agents. First, a reinforcement learning-based architecture search agent is designed to adaptively select among Global Self-Attention, Window Attention, and External Attention for each block to optimize the model structure. Second, a feedback optimization agent is developed to dynamically adjust the search process, balancing exploration and convergence. Third, a quantization agent is employed that leverages quantization-aware training (QAT) to generate an INT8 deployment-ready model with minimal loss in accuracy. Experiments conducted on the Human3.6M dataset demonstrate that the proposed method achieves a mean per joint position error (MPJPE) of 42.15 mm with only 4.38 M parameters and 19.39 GFLOPs under FP32 precision, indicating substantial potential for subsequent gathering behavior recognition tasks. Full article