Sensors最新文献

(1) Background: Given the large amount of data extracted from information technologies, principal component analysis (PCA) allows the identification of the most important variables to assess physical activity (PA). The aim of this systematic review is to highlight which variables, extracted through PCA (as a data reduction technique), provide the most information about preschool and school children's PA-related behavioral patterns during school hours. (2) Methods: The search was conducted in PubMed, SCOPUS, Web of Science, and ProQuest Central according to the PRISMA guidelines and the guidelines for performing systematic reviews in sports sciences. (3) Results: From 403 studies initially identified, seven were fully reviewed, and their outcome measures were extracted and analyzed. An analysis of these seven studies (n = 8927) revealed that volume-related components explained the majority of the variance (62.8-69.0%) in PA behaviors, while intensity components contributed less (14.4-14.8%). However, confidence intervals and heterogeneity statistics (I²) were not reported in the original studies, limiting quantitative synthesis. (4) Conclusions: This systematic review demonstrates that PCA effectively identifies multidimensional patterns in children's PA and motor development, with volume-related dimensions consistently dominating the variance structure across diverse populations and settings.

Long-term time series forecasting (LTSF) is critical for modern power systems, energy management, and grid planning. Yet virtually all existing forecasting models employ stationary activation functions that apply identical nonlinear mappings regardless of temporal context-a fundamental mismatch with real-world load data, which exhibits strongly regime-dependent dynamics such as summer demand peaks, winter heating patterns, and overnight low-load periods. We address this gap by proposing TC-KAN (Time-Conditioned Kolmogorov-Arnold Network), the first forecasting architecture to augment KAN activation functions with position-aware coefficient parameterisation. The core innovation replaces the static polynomial coefficients in standard KAN activations with position-conditioned coefficients produced by a lightweight positional-embedding MLP, providing additional learnable capacity beyond standard KAN while adding negligible parameter overhead. TC-KAN further integrates a dual-pathway processing block-combining depthwise convolution for local temporal pattern extraction with the time-conditioned KAN layer for enhanced nonlinear transformation-within a channel-independent framework with Reversible Instance Normalisation. Experiments were conducted on four standard ETT benchmark datasets and the high-dimensional Weather dataset. TC-KAN achieves superior or competitive accuracy in most configurations while requiring merely 51K parameters-approximately 40% of DLinear and ∼100× fewer than iTransformer. On ETTh2, TC-KAN reduces the mean squared error by up to 61.4% over DLinear, and matches the current state-of-the-art iTransformer on ETTm2 at a fraction of the computational cost. This extreme parameter reduction circumvents the steep memory bottlenecks endemic to massive Transformer models, positioning TC-KAN as a highly practical architecture tailored precisely for resource-constrained edge deployments-such as on-device load forecasting inside smart grid sensors and industrial IoT controllers.

Turbid Class II inland waters such as Taihu Lake exhibit a "spectral uplift" effect driven by suspended particulate matter (SPM) scattering and colored dissolved organic matter (CDOM) absorption, which can obscure chlorophyll-a (Chl-a) signals in the visible-red-edge region and challenge retrieval under small-sample, collinear feature settings. Using multispectral observations from the BC-1A satellite (carrying the Lightweight Hyperspectral Remote Sensing Imager, LHRSI) and synchronous satellite-ground in situ measurements acquired over Taihu Lake in late autumn, this study proposes Chl-a-oriented PCA-RF (COP-RF), a leakage-safe inversion framework integrating correlation screening, principal component analysis (PCA), and random forest (RF) regression. Candidate band-combination features are generated, and PCA is applied for orthogonal compression to mitigate collinearity before RF learning. A stratified five-fold cross-validation based on Chl-a quantile bins is adopted, with screening, standardization, and PCA fitted only on training folds. COP-RF achieves stable performance under the current dataset (R2=0.671, RMSE =1.80μg/L, MAE =1.25μg/L). Spatial inversion shows higher Chl-a near shores and bays and lower values in the lake center, consistent with Sentinel-2 hotspot ranks.

Convolutional neural networks (CNNs) have been widely used for acne image classification due to their effectiveness in capturing local texture of skin lesions. However, the locality of convolution operations limits their ability to model long-range dependencies. Vision Transformer (ViT) methods address this issue to some extent but their high computational complexity and reliance on large-scale pre-training present challenges. Although CNN-Transformer architecture alleviates this conflict to some extent, acne images present task-specific challenges, including indistinct lesion boundaries, subtle inter-class variations, and various facial interference factors. In this paper, we propose AcneFormer, a lesion-aware and noise-robust CNN-Transformer architecture for acne image classification. We introduce three modules especially for acne tasks: a Lesion Cue Enhancement (LCE) module to highlight discriminative multi-scale spatial patterns, a Cross-Layer Feature Transmission (CLFT) module to enhance cross-layer information flow in Transformers, and a Differential Semantic Denoising (DSD) module to suppress irrelevant responses during deep feature interaction. Extensive experiments show that AcneFormer outperforms several strong baselines. Ablation and external lesion-annotated analyses further show a consistent pattern: LCE mainly improves lesion-sensitive localization and class-balanced recognition, CLFT expands valid cross-depth lesion evidence, and DSD suppresses off-lesion semantic responses.

Smartwatches are widely used to monitor physiological responses during exercise; however, their accuracy in measuring heart rate (HR) and energy expenditure (EE) across different exercise modalities remains insufficiently characterized. This study evaluated the accuracy of HR and EE measurements obtained from four commercially available smartwatches in comparison with gold-standard reference methods. Sixty-two healthy adult men performed standardized endurance and resistance exercise protocols while simultaneously wearing four smartwatches (Apple, Galaxy, Fitbit, and Garmin). HR was measured using electrocardiography (ECG), and EE was determined using indirect calorimetry. Measurement accuracy was assessed using repeated-measures analysis of variance, Pearson's correlation analysis, intraclass correlation coefficients (ICCs), and Bland-Altman analyses. All smartwatches demonstrated high accuracy in HR measurements during both endurance and resistance exercises. During endurance exercise, HR measurements from all smartwatch brands were comparable to those obtained via ECG, whereas during resistance exercise, only the Apple Watch showed no significant difference from the ECG. HRs showed strong correlations with ECG readings (r = 0.64-0.97), excellent reliability (ICC > 0.94), and narrow limits of agreement (approximately ±10 bpm). In contrast, the EE measurements exhibited limited accuracy across all devices. During endurance exercise, EE was consistently underestimated with wide limits of agreement. EE accuracy further deteriorated during resistance exercise, showing weak correlations with indirect calorimetry (r = 0.10-0.34) and poor reliability (ICC < 0.45). Overall, smartwatches provide accurate HR measurements across endurance and resistance exercise modalities, supporting their use in exercise intensity monitoring and HR-based training. However, smartwatch-derived EE estimates do not accurately reflect the metabolic demands, particularly during resistance exercises. Future research should focus on improving EE estimation algorithms through multimodal biosignal integration and machine-learning approaches, and validating these methods across diverse populations and exercise modalities.

Vehicular Ad hoc Networks (VANETs) feature high node mobility and volatile topologies, rendering the conventional Greedy Perimeter Stateless Routing (GPSR) protocol prone to weak link stability and inefficient route discovery due to its lack of direction awareness. Existing direction-aware improvements typically rely on multi-criteria weighting or clustering, introducing heavy parameter fusion and computational overhead that conflict with the resource-constrained nature of onboard units. To overcome these limitations, this paper presents VANET-GPSR+, a lightweight enhanced routing protocol. Its key novelty is that it discards multi-parameter fusion and relies solely on movement direction, supported by a synergistic framework of three lightweight mechanisms: direction-aware neighbor classification to prioritize nodes with consistent trajectories, adaptive greedy forwarding region expansion in sparse and dynamic networks, and path deviation angle-based next-hop selection. This work builds a probabilistic link lifetime model that theoretically quantifies the stability gains of direction awareness-a novel theoretical foundation. Comprehensive urban and highway simulations show that VANET-GPSR+ improves the packet delivery ratio by 16.3% and reduces end-to-end delay by 27.5% compared with standard GPSR, and it outperforms both OP-GPSR and AK-GPSR. It introduces negligible CPU and memory overhead, with CPU usage over 50% lower than the two benchmark protocols at 80 vehicles/km, and demonstrates strong robustness against varying beacon intervals and communication radii. Retaining GPSR's stateless and distributed traits, VANET-GPSR+ delivers substantial performance gains with minimal overhead, serving as an efficient routing solution for highly dynamic VANETs.

Central nervous system (CNS) disorders represent a growing healthcare burden, and various drugs are developed for their treatment. However, the blood-brain barrier (BBB) prevents over 98% of therapeutics from reaching brain tissue. Intranasal delivery provides a promising alternative by exploiting olfactory and trigeminal nerve pathways to circumvent the BBB. This review surveys recent advances in nose-to-brain delivery technologies, from carrier design to evaluation methods. Polymeric and lipid-based nanocarriers show enhanced mucosal penetration and prolonged residence time, and microneedle platforms further enable controlled drug release with minimal discomfort. To evaluate these delivery strategies, sensor-integrated organ-on-chip models provide more physiologically relevant testing than static cultures. Although persistent challenges such as rapid mucociliary clearance and formulation stability remain, combining nanotechnology with microfluidic devices and computational modeling shows potential for developing patient-specific therapeutics.

Digitalizing essential services opens up a new risk of exposing critical infrastructure to botnet infections. In a grid topology network, the neighbor-to-neighbor paths can be used by the malicious botnet to spread the infection. Previous white-hat worm launchers used heuristics and supervised learning to exterminate botnets, which demand specific conditions or a suitable dataset to be effective. Although reinforcement learning addressed these issues, it requires a longer time to train. This article proposes a framework to shorten training and improve the effectiveness of reinforcement learning. The framework applies four key principles: (1) surveying the network status with multi-tensor input, (2) removing irrelevant actions via a novel Chebyshev-based masking strategy, (3) reinforcing key actions with rewards, and (4) optimizing rewards for winning. Four reinforcement learning algorithms are implemented to evaluate the framework, which are vanilla policy gradient, deep Q-network, proximal policy optimization, and MuZero in a stylized grid topology network simulation. An ablation study indicates that the masking used in identify accounts for the majority of the improvement, whereas multi-channel in Survey alone can reduce performance without complementary masking, rewards, and optimization. With the mean winning rate improved by 49.129% and mean win efficiency improved by 118.8031% against our previous work, the framework effectiveness is confirmed in stylized simulations.

Accurate classification of uranium slag origin and category is essential for nuclear environmental monitoring and safety. This study presents a hybrid framework combining laser-induced breakdown spectroscopy (LIBS), four preprocessing methods, and five machine learning algorithms for rapid uranium slag classification. A total of nine sample categories were collected from three mining areas, with categories defined by their U concentration levels within each origin. Standard normal variate (SNV), Savitzky-Golay smoothing (SG), and their combinations (SNV-SG, SG-SNV) were applied to evaluate preprocessing effects. To address ultra-high-dimensional spectral data (49,242 points per spectrum), principal component analysis (PCA) and random forest (RF) were employed for feature engineering, integrated with support vector machine (SVM), linear discriminant analysis (LDA), and K-nearest neighbors (KNN) classifiers. Hyperparameter optimization via five-fold cross-validation and Bayesian optimization enhanced accuracy and efficiency. RF-based hybrid models consistently outperformed PCA-based counterparts. Remarkably, the RF-LDA model with SNV-SG preprocessing achieved 100% classification accuracy across all test sets with a processing time of only 10.46 s, demonstrating exceptional discriminative power and computational efficiency. These findings establish that combining RF feature selection with advanced machine learning offers a robust solution for LIBS-based nuclear material classification, with significant implications for both nuclear safety and resource management.

Accurate calibration is essential for ensuring the performance of magnetic gradient tensor (MGT) arrays. Existing calibration methods generally rely on mechanical rotation to obtain magnetic responses under multiple orientations. However, for large-scale cubic MGT arrays, rotating the entire array using a high-precision non-magnetic turntable is often costly and impractical, while manual rotation is difficult to control and may introduce array-center offsets. To address these limitations, this paper proposes a rotation-free scalar calibration framework for cubic MGT arrays, in which a tri-axial Helmholtz coil system generates constant-magnitude magnetic fields with randomized orientations while compensating for ambient magnetic drifts. Based on the acquired data, a hierarchical calibration algorithm is developed to estimate sensor-level intrinsic errors and array-level misalignment errors. Experimental results show that the proposed method reduces the joint tensor invariant CT from 9.07×103 nT/m to 11.51 nT/m, corresponding to a 99.87% reduction. In addition, compared with a conventional rotation-based fast calibration method, the proposed framework further decreases the mean and RMS of the joint CT by 62.7% and 63.1%, respectively. These results demonstrate that the proposed framework improves the spatial consistency of the MGT array and provides a practical calibration solution for large-scale MGT array systems.