Search Results (1,354)

On-device plant disease perception must reconcile two competing demands: enough spatial detail to localise diseased tissue within a field image and a memory and compute budget small enough for microcontroller-class hardware. Single-network solutions that jointly learn a pixel-wise mask and a fine-grained disease label tend to oversubscribe both the Flash and activation SRAM of such devices. This paper presents Chroma-Sense 2.0, a two-stage lightweight pipeline that decouples the two subproblems and sizes each stage for its own budget. The two stages run sequentially on the same frame: Stage 1 is a per-channel convolutional classifier, derived from Chroma-Sense, that names the disease, and Stage 2 is a compact ESPNet segmenter that produces a binary diseased-versus-healthy mask localising it. Because the stages run one after the other rather than concurrently, the peak working memory of the pipeline is the maximum of the two stages rather than their sum. We evaluate the pipeline on the in-the-wild PlantSeg dataset using a curated 10-species, 34-class subset and a leakage-controlled protocol in which all training crops are derived from PlantSeg’s official training images and all reported metrics are measured on a held-out test set of 5002 crops built from the official test images. The segmentation stage attains a mean foreground recall of 0.97 (mean foreground IoU of 0.49; 0.53 pooled over pixels), a deliberately recall-oriented operating point. Against Fast-SCNN, a small U-Net, LR-ASPP, and DeepLabV3+, ESPNet is the smallest-footprint model (140k parameters, 193 KB Int8 Flash) while retaining the highest foreground recall; the per-channel classifier reaches accuracies comparable to much larger ImageNet-pretrained backbones (MobileNetV3 and EfficientNet) using 10–13× fewer parameters. End to end, the coupled pipeline classifies the disease correctly on 87.8% of the test crops. An on-device profile on the OpenMV H7 and H7 Plus shows that the binding constraint at 256 × 256 is the segmenter’s ≈4 MB contiguous activation arena, rather than parameter Flash: Even on the 32 MB-SDRAM H7 Plus, the usable interpreter heap is only about 4 MB, and the arena cannot be allocated as a single contiguous block from it, so on the tested firmware, the classifier runs on microcontrollers while the segmenter does not; the full pipeline instead fits the gigabyte-scale single-board-computer tier (for example, Raspberry Pi or NVIDIA Jetson Nano), and enabling the segmenter to run on microcontrollers remains the open gap. Full article

Patients with congenital heart disease (CHD) represent a special healthcare challenge due to their high complexity, which accompanies them throughout all life stages. Consequently, this population faces increased morbidity and mortality rates, often linked to hemodynamic shifts in pulmonary flow or cardiac output. These risks are further compounded by potential arrhythmias and heart failure decompensation, which may lead to the progressive progression toward advanced stages of the disease. Unfortunately, standard outpatient follow-up is often not capable of responding to the continuous monitoring needs presented by these patients and their families. This selective review frames information on the wearable devices that have emerged as a key solution for continuous and remote monitoring. Beyond clinical tracking, research is increasingly focusing on their role in assessing physical activity—a critical determinant of health outcomes in the CHD population. This review examines the existing literature on wearable technology in both pediatric and adult patients while also addressing the current limitations that hinder their integration into routine clinical practice. Full article

Deploying object detection models on low-power embedded devices requires a careful balance between detection accuracy, runtime performance, and energy efficiency, especially for aerial images that contain small objects. We present a hardware-measured evaluation of Edge-TPU-compatible detectors on an MCU-class NXP i.MX RT1176 (bare-metal FreeRTOS, Cortex-M7) driving a Coral Edge TPU over an internal USB 2.0 link. We benchmark quantized YOLOv5-based detectors at input resolutions from 256 to 1024 px, reporting the accuracy (size-stratified AP_S on the VisDrone person split) together with the measured per-frame and per-pixel energy, sampled at 100 Hz with an automatic window and N ≥ 5 repetitions. We use the energy per pixel as a resolution-normalized view, not as a new metric, to show where the analytical model fails. The measured per-pixel energy departs from the compute (MAC)-based model because the inference is transfer-bound: it is dominated by the instruction stream and input activations re-sent over USB on every invoke, while the parameters stay cached on-chip. An Edge TPU clock sweep and a compiler-byte analysis agree independently that about 79% of each invocation is the USB transfer. This gives concrete design rules (a lower clock is more energy-efficient; ReLU is preferred to SiLU), and shows that the USB 2.0 ceiling is architecturally fundamental for the MCU class. Full article

This survey presents a comprehensive review of Human Pose Estimation spanning 2D and 3D settings, unifying prior work through a taxonomy of body representations (2D keypoints, 3D skeletons, dense meshes), processing flows (top-down vs. bottom-up), problem formulations (regression vs. detection/heatmaps), and modern learning architectures (CNNs, Transformers, GCNs). We compare reported benchmark results of representative methods across widely used datasets (e.g., COCO, MPII, Human3.6M, 3DPW) and evaluation metrics (AP/OKS, PCK/AUC, MPJPE/PA-MPJPE, PVE), highlighting trade-offs between accuracy, robustness, and efficiency. Despite substantial progress driven by deep learning and temporal modeling, we identify persistent challenges, including costly and biased annotations, domain shift, occlusion, depth ambiguity, multi-person association, and real-time constraints on edge devices. We synthesize emerging directions that target these gaps, data-centric learning, stronger temporal and kinematic priors, and whole-body modeling, and outline deployment-oriented frontiers including generative motion priors, model compression, and on-device inference, framing their implications for engineering systems that demand reliable, low-latency human motion analysis. Full article

Ideathons and pitch competitions have expanded rapidly as experiential learning devices, but the textual artefacts they produce—presentation slides—remain under-examined. This study applies interpretable computational text analysis to 104 ideathon decks (1289 content slides) from four cohorts: Cambodia 2023, 2024, 2025 and Thailand 2025. Measures include lexical frequency, TF-IDF, lexicon-based sentiment, a ten-component pitch-completeness proxy, numerical density, and Jaccard similarity. Because sector designation was absent in Cambodia 2023 and present from 2024 onward, the longitudinal Cambodian data support an observational cohort comparison with an institutional change between cohorts; year effects, programme evolution, and sector designation cannot be separated. The Cambodia 2025 vs. Thailand 2025 contrast is a single-year cross-country comparison, not a longitudinal one. Between Cambodia 2023 and 2024, presentations show large Cohen’s d differences with 95% bootstrap confidence intervals (CIs) in total words, unique words, slide count, pitch completeness, and market-related vocabulary, alongside a small decline in type–token ratio. At fixed sector composition, Cambodia 2025 and Thailand 2025 differ sharply in surface vocabulary (top-50 Jaccard = 0.176): Cambodia leans toward agriculture, rural markets, and community development, while Thailand leans toward AI, learning, and cassava-centric agronomy. AI use was not directly measured, so all claims about generative AI are hypothesis-generating; the drop in within-cohort pairwise Jaccard from 2024 to 2025 (0.061 → 0.041) is consistent with—but does not establish—an augmentative rather than homogenising effect of AI assistance. Findings are reported as descriptive associations and interpreted through the lens of constraint-based creativity and institutional theory. We discuss implications for curriculum designers who wish to balance structural templates with exercises that promote diverse problem framings. Full article

Bulk heterojunction (BHJ) active layers in organic photovoltaics (OPVs) are nanostructured composites in which electron-donating and electron-accepting semiconductors form interpenetrating phases for exciton dissociation and charge transport. The power conversion efficiency (PCE) of these organic-organic nanocomposites is governed by structural features spanning multiple length scales: molecular packing and energy-level alignment at donor/acceptor (D/A) interfaces, phase-separation morphology and crystallite connectivity, and thin-film optical and charge-transport characteristics. Rational design of high-performance OPV nanocomposites requires multiscale numerical modelling that bridges quantum chemistry, mesoscale morphology simulation, and device-scale optoelectronic modelling. This review surveys and critically compares recent advances in the structural design and numerical simulation of OPV BHJ nanocomposites. At the molecular scale, we examine density functional theory and non-adiabatic molecular dynamics approaches for resolving charge-separation driving forces, interfacial energy-level alignment, and exciton dynamics. At the mesoscale, we discuss molecular dynamics, kinetic Monte Carlo, and electronic coarse-graining methods for describing phase separation, crystallization kinetics, morphology evolution, and charge transport. At the device scale, we review exciton-diffusion, optical transfer-matrix, and drift-diffusion models that quantitatively link morphology to photovoltaic performance metrics. The review also evaluates how machine learning, high-throughput screening, surrogate models, and generative design accelerate donor–acceptor selection and morphology optimization, while distinguishing benchmark predictions from experimentally validated design rules. Across these scales, we compare the strengths, assumptions, and validation limits of the principal modelling approaches. Finally, we highlight emerging multiscale integration frameworks, including sequential parameter-passing pipelines and differentiable digital-twin concepts. By framing OPV BHJ layers as nanocomposites whose performance bottlenecks map onto composite-design challenges such as interface integrity, phase connectivity, multiscale charge transfer, and degradation-aware design, this review connects OPV modelling with broader structural-composites thinking for next-generation organic solar cells. Full article

Touchless interfaces for embedded systems require models that are both compact and evaluated under protocols that reflect realistic generalization. Public low-resolution gesture datasets can obscure this requirement because repeated samples, temporal neighbors, or subject identity may leak across train and test partitions. This study presents a leakage-aware multi-corpus benchmark in which four public corpora are evaluated within their own dataset-specific partitions after conversion to a common 8 × 8 depth representation when needed: a public IEEE DataPort ToF CSV with 8400 frames from four gestures, DS.GestureRecognition_TOF.1.0 with 3109 frames extracted from 40 low-resolution ToF sequences and three gestures, the depth component of the public Kinect+Leap dataset with 1400 samples from 14 subjects and 10 gesture classes, and the public Senz3D depth archive with 1320 samples from 4 subjects and 11 gesture classes. No blurred RGB image dataset, deblurring target, or synthetic blur augmentation is produced in the experiments; the only spatial reduction applied to external depth corpora is deterministic block averaging for a shared low-resolution depth input. The term multi-corpus is used here to clarify scope: the experiments do not train on one dataset and test on another, but instead compare optimistic random splits against structure-aware alternatives matched to each corpus: chronological block-aware, sequence-aware, and subject-wise evaluation. On the original IEEE ToF benchmark, the best macro F1-score decreases from 1.0000 under random splitting to 0.8494 under block-aware evaluation. On DS.GestureRecognition_TOF.1.0, random splitting again reaches 1.0000, whereas the best sequence-aware result falls to 0.6697. On the downsampled Kinect+Leap depth benchmark, random frame classification reaches 0.7639 macro F1, while subject-wise evaluation drops to 0.0566. On Senz3D reduced to 8 × 8, random frame classification reaches 0.9068 macro F1, while subject-wise evaluation drops to 0.0717. Device-side validation on an NVIDIA Jetson Nano Developer Kit preserved a macro F1-score of 0.8667 for the selected GRU on the primary IEEE ToF dataset with 14.48 ms/window steady-state CPU latency. The results show that low-resolution gesture recognition conclusions are strongly protocol- and corpus-dependent: compact temporal models remain viable for genuine ToF streams, but naive random splitting can substantially overstate generalization, and subject-wise recognition after aggressive 8 × 8 conversion remains difficult for the evaluated datasets and architectures. Full article

In dense indoor Internet of Things (IoT) environments, simultaneously transmitting and reflecting intelligent surfaces (STAR-IRS) can provide full-space coverage by steering energy to both sides of the surface. However, most STAR-IRS works focus on the physical layer and overlook medium access control (MAC) challenges, such as how to collect channel state information (CSI), dynamically assign each STAR-IRS element to either the reflection or transmission mode, and coordinate concurrent users. We propose a two-stage MAC protocol that couples carrier sense multiple access with collision avoidance (CSMA/CA) for contention and time division multiple access (TDMA) for scheduled data transmission. During contention, devices send extended request-to-send (eRTS) frames to enable CSI acquisition; during transmission, the access point (AP) configures the STAR-IRS, allocates TDMA slots, and pairs one reflection-region device (RD) with one transmission-region device (TD) for non-orthogonal multiple access (NOMA) with successive interference cancellation (SIC). Analysis and simulations show that the proposed scheme increases throughput and reduces delay compared with reflective-only IRS baselines without NOMA, while remaining practical for dynamic indoor IoT. Full article

Human pose estimation (HPE) has become a core technology for human-centered visual understanding and is widely used in sports analysis, rehabilitation assessment, human–computer interaction, autonomous driving, and industrial safety monitoring. Although benchmark-driven research has substantially improved pose estimation accuracy, strong performance on standard datasets does not necessarily ensure reliable deployment in real-world environments. Practical HPE systems are commonly deployed in camera-based, mobile, wearable, and edge-computing environments. In these settings, sensor quality, device placement, frame rate, illumination conditions, hardware limitations, temporal instability, and application-specific reliability requirements jointly influence system performance. Using a structured narrative synthesis of representative literature, this review examines HPE from a deployment-oriented perspective instead of treating pose estimation solely as a model-level task. The review first summarizes the major benchmark-to-deployment gaps, including data, system, task, and temporal gaps; it then analyzes the end-to-end HPE pipeline, including image or video acquisition, human detection, region localization, pose inference, tracking, temporal smoothing, coordinate decoding, and task-level decision-making. Representative CNN-based, Transformer-based, temporal modeling, and lightweight optimization methods are reviewed in terms of deployment cost, optimization potential, and application suitability. This review also discusses practical application requirements and future research directions related to deployment-aware evaluation, system-level optimization, efficient temporal modeling, and application-level reliability. By linking benchmark performance with practical sensing and system constraints, this review provides a structured reference for the selection, evaluation, and deployment of HPE methods in real-world environments. Full article

Real-time video anomaly detection (VAD) under realistic edge constraints—sub-second latency, ≤25 W power, no cloud dependency, and human-interpretable output—remains an open problem. Existing lightweight video convolutional neural networks (X3D, MoViNets) are bound to closed-set training distributions, while recent vision–language-model-based VAD methods (LAVAD, VERA, Holmes-VAD) achieve 80–89% area under the curve (AUC) but rely on datacenter-grade GPUs and Chain-of-Thought (CoT) reasoning that pushes per-segment latency well above one second. This paper reframes the design target from peak accuracy to practical edge deployability and contributes two tightly coupled designs: (i) an edge-optimized inference stack that compresses Qwen3-VL-2B with 4-bit Activation-aware Weight Quantization (INT4 AWQ) and serves it through a TensorRT-LLM C++ runtime on NVIDIA Jetson Orin NX (16 GB, 25 W); and (ii) a fully automatic, CoT-free verbalized prompt optimization in which an 8B optimizer iteratively refines a natural-language definition block $D_t$ using class-balanced (stratified) development batches on a disjoint development subset, with no human editing and no runtime cost on the edge device. Three findings support this framing: (a) the inference stack reduces per-segment latency to 0.25 s, a 7.4× speed-up and 55% memory reduction over a Python/PyTorch baseline; (b) verbalized prompt optimization improves zero-shot AUC from 71.82% (manual prompt) to 76.39%, outperforming GPT-4- and Gemini-Pro-generated prompts (74.12% and 74.35%) under the same edge backbone; and (c) single-frame input attains the highest mean AUC among one-, five-, and eight-frame windows—statistically comparable to the five-frame setting—while offering the lowest latency, making it the preferred operating point under the edge budget. While the absolute AUC (76.39%) is below recent server-side methods (CLIP-TSA 87.58%, VadCLIP 88.02%, Holmes-VAD 89.51%), our framework is the only one in this comparison that operates entirely on a ≤25 W edge device, providing a deployment-oriented operating point on the accuracy–feasibility frontier of VLM-based VAD. Full article

In the context of the large-scale integration of high-proportion renewable energy into power grids, Power Conversion Systems (PCS) and Static Var Generator (SVG), as the core power electronic devices for ensuring grid frequency stability and power quality, demand high-level safety and reliability. To tackle challenges such as the wide range of defect sizes in PCS and SVG, the low recognition accuracy for microscopic fuzzy defects, and complex background interference, this paper presents a lightweight and high-precision defect recognition model (RD-YOLO) based on the latest YOLOv11 benchmark. First, an improved Mosaic algorithm is introduced. This algorithm utilizes conflict relationship tables to preserve physical context semantics during traditional non-discriminative data augmentation. Second, to surmount the limitations of scale-aware feature extraction, the YOLOv11 is re-engineered within the backbone network by integrating a Res2Net multi-scale cascaded mechanism. This enhances the network’s capacity to capture both fine-grained defect features and large-scale defect boundaries. Third, Focal Loss is employed for difficult sample detection. Nonlinear gradient modulation is utilized to guide the model to focus on ambiguous defect edges. Finally, the Soft-NMS post-processing strategy significantly enhances the regression accuracy in densely corroded regions. Experimental validation on a self-developed dataset consisting of 8500 high-resolution PCS and SVG defect images reveals that the enhanced RD-YOLO attains an average precision of 89.6% and a frame inference rate of 98 FPS (in RTX 3090), offering robust technical support for intelligent visual maintenance in renewable energy facilities. Full article

Driver behavior detection is essential for improving road safety in Intelligent Transportation Systems (ITSs). This paper proposes an attention-enhanced CNN integrated with Explainable Artificial Intelligence (XAI) to achieve reliable and interpretable driver behavior classification. Furthermore, the model also uses a multi-head attention mechanism to ensure that it captures local spatial features as well as long-range dependencies in the activities of the driver. The State Farm Distracted Driver dataset is experimented with using stratified 5-fold cross-validation. The proposed MHA-CNN model has a mean accuracy of 99.48% on all folds and an overall high levels of precision, recall, and F1-score in all ten distraction classes. Grad-CAM is used to produce attribution maps to improve interpretability by showing semantically important parts of the image, including hands, face, and mobile devices, and can give a visual explanation of why models make decisions. In addition, an ablation investigation is performed with k-fold validation to assess the value of the multi-head attention mechanism. Its model can run at 122 frames per second with 42.92 million parameters, showing that the suggested method is a reasonable tradeoff between accuracy, efficiency, and transparency and can be applied in real-time driver monitoring applications. Full article

Small unmanned aerial vehicles (UAVs) often occupy only a limited number of pixels in an image and can be easily confused with surrounding objects in cluttered scenes, which makes reliable detection difficult. To address this challenge, we develop PCCS-YOLOv8, an enhanced object detector tailored to small UAV targets. A P2 prediction branch is added to retain fine spatial information associated with tiny objects. The cross-stage partial pyramid convolution (CSPPC) module is introduced to offset the additional computational burden caused by the detection branch with high resolution, while the spatial pyramid pooling with efficient layer aggregation network (SPPELAN) combines multiscale pooling with efficient feature aggregation. The convolutional block attention module (CBAM) is further integrated to emphasize features related to targets and reduce interference from complex backgrounds. Experiments were conducted on a UAV dataset containing 7785 images collected from TIB-UAV, Anti-UAV, and self-collected sources. PCCS-YOLOv8 achieved an mAP@0.5 of 94.0% and an mAP@0.5:0.95 of 50.6%, outperforming the YOLOv8 baseline by 2.9 and 2.2 percentage points, respectively. After training, the model was exported, converted to RKNN format, and then deployed on an Orange Pi 5 Pro development board. In real-world detection tests, the embedded system achieved an average frame rate of 27.7 FPS and an average runtime of 46.3 ms per frame. These results demonstrate the potential of the proposed method for real-time UAV detection on edge devices. Full article

Internet-of-Things (IoT) devices in smart homes are vulnerable to passive traffic fingerprinting, where an adversary captures encrypted IEEE 802.11 frames and identifies devices using MAC-layer metadata such as packet sizes and inter-arrival times. Existing defenses based on padding, traffic shaping, or synthetic cover traffic can remain vulnerable because artificial timing signatures are detectable by machine learning classifiers. This paper proposes a game-theoretic framework for evaluating Wi-Fi MAC-layer cover-traffic injection defenses. We introduce donor-based mimicry injection, in which the access point injects a replica of a paired device’s authentic traffic into each device’s stream. We compare donor mimicry with fixed-rate, exponential, and uniform synthetic baselines across 198 scenario instances (156 unique defender configurations) and eight classifiers using 10-fold cross-validation. Donor mimicry at 100% bandwidth overhead reduces the best attacker’s balanced accuracy to 33.5%, whereas synthetic methods at equal overhead reach 93.9%, showing that behavioral realism, rather than injected volume alone, drives effectiveness. Modeling the interaction as a finite two-player zero-sum game yields a mixed-strategy Nash equilibrium with game value 0.247 within the evaluated strategy space; a deployable deterministic defense holds the best pairing-unaware attacker to 25.9% balanced accuracy, near the four-class random baseline of 25%. A pairing-aware robustness analysis shows that an attacker who can orient the donor-induced identity swap recovers near-baseline accuracy, so the four-class protection presumes pairing secrecy and the durable effect is pair-level anonymity. The defense operates at the access point and requires no IoT device modifications. Full article

High-salinity waters, including seawater, brines, and saline radioactive effluents, contain recoverable uranium or require uranium removal, but their complex chemistry complicates both capture and regeneration. Under seawater-relevant conditions, U(VI) occurs mainly as uranyl carbonate complexes and ternary uranyl carbonate complexes containing Ca²⁺ or Mg²⁺ rather than as free UO₂²⁺. Selective separation therefore depends on coupled transport, hydration-shell reorganization, carbonate displacement, and interfacial coordination. Conventional sorbents largely optimize binding strength and adsorption capacity, often at the expense of harsh stripping and secondary waste. This review frames thermoresponsive uranium separation as a coupled aqueous-speciation, interfacial-state, and process-design problem. It examines how LCST and UCST transitions, polymer-brush reorganization, hydration-layer reconstruction, pore gating, and localized photothermal heating regulate access to binding sites and release pathways. Polymer brushes, hydrogels and microgels, membranes and nanochannels, ion-imprinted magnetic composites, and MXene-based hybrids are critically compared using cycle-level criteria, including U/V selectivity, switching time, regeneration demand, energy input, fouling resistance, material loss, synthesis reproducibility, and environmental performance. Particular emphasis is placed on distinguishing genuine structural gating from the generic effects of temperature on diffusion, ligand exchange, and adsorption equilibria. Current evidence supports the feasibility of programmable capture–release interfaces but remains limited by matrix-dependent transition windows, incomplete mechanistic attribution, scarce quantitative energy and temperature-gradient data, short cycling tests, and limited device-scale validation. Progress will require standardized testing in realistic saline matrices and complete capture–release mass and energy balances rather than evaluation by maximum adsorption capacity alone. Full article