Search Results (58)

Efficient object detection is needed in automated driving and edge perception. In these scenarios, a detector must work under limits on latency, power, and memory. YOLOv8 is a strong real-time baseline, but its computation can still be high for compact deployment. This paper proposes MV3-YOLO, a lightweight YOLOv8 variant with a stage-wise hybrid backbone. The early Conv/C2f stages are kept to retain low-level spatial details. Lightweight modules are placed in deeper stages, where feature maps are smaller and redundant computation is more common. C2fMixed is used at the stride-16 stage to balance feature capacity and cost. C2fGhostis used at the deepest stage to generate high-level features with fewer parameters. The YOLOv8 neck and head are kept unchanged for stable multi-scale fusion. On the KITTI validation set, MV3-YOLO reaches mAP@0.5 = 0.859 and mAP@0.5:0.95 = 0.610 with only 2.53 M parameters and 6.6 GFLOPs. Compared with YOLOv8n, it reduces parameters by 19.7% and GFLOPs by 25.0% while improving mAP@0.5 by 1.66% and mAP@0.5:0.95 by 1.50%. On COCO val2017, MV3-YOLO obtains 38.4 mAP@0.5:0.95, which is higher than the YOLOv8n reference result and close to YOLOv10n. These results show that MV3-YOLO reduces deployment cost while keeping competitive detection accuracy. Full article

The increasing convergence of Operational Technology (OT) and Information Technology (IT) within the Industrial Internet of Things (IIoT) brings about remarkable improvements in monitoring and automation. However, it also exposes industrial systems to large-scale Distributed Denial of Service (DDoS) attacks. Edge-based defences are essential in satisfying low-latency demands and data sovereignty rules, yet they must function under severe resource limitations and adapt to shifting traffic characteristics without cloud assistance. In this work, we introduce a lightweight hybrid deep learning architecture that fuses a Convolutional Neural Network (CNN) with a Convolutional Block Attention Module (CBAM) and a Multi-Layer Perceptron (MLP) in a single detector. A sequential transfer learning scheme is adopted, including a feature projection layer that handles differences in input dimensionality. The model is pre-trained on the CIC-DDoS2019 dataset, then adapted to the more recent CICIoT23 dataset. Evaluations are performed on both datasets while preserving their natural class imbalance. We provide extensive ablation and variance analysis under identical experimental conditions. The proposed method achieves 99.52% accuracy on CICIoT23 while maintaining 99.65% recall, which is a crucial property for critical systems. Real-time measurements on a CPU-only testbed show an average inference latency of 0.013 ms, inference-only throughput exceeding 93,000 packets/s, and end-to-end batch throughput of approximately 38,000 packets/s. The solution demonstrates effective domain adaptation, sub-millisecond latency, and suitability for resource-constrained IIoT edge gateways. Full article

Real-time video object detection on unmanned aerial vehicles (UAVs) is essential for urban inspection and autonomous perception, yet its deployment on edge devices is severely constrained by the high computational cost of accurate detectors, the quantization sensitivity of hybrid convolution-attention networks, and the system-level latency of full video processing pipelines. To address these challenges, we present DUST-YOLO, a deployment-oriented algorithm-hardware co-design framework, where structured pruning and mixed-precision quantization-aware training (QAT) are jointly optimized with TensorRT–DeepStream for efficient UAV small-object detection on edge platforms. First, we introduce a multi-dimensional structured pruning strategy that applies asymmetric channel pruning to convolutional and feature-fusion modules while compressing the Swin Transformer prediction heads and bottleneck stacks, thereby reducing parameters and computation with limited impact on multi-scale representation capability. Second, we develop a hardware-aware mixed-precision QAT scheme that maps computation-intensive backbone layers to INT8 while preserving the Transformer-related modules in FP16, improving inference efficiency while mitigating the accuracy loss caused by uniform low-bit quantization. Third, we compile the optimized network with TensorRT and integrate the resulting inference engine into a DeepStream-based asynchronous video pipeline on the edge platform, enabling end-to-end acceleration by reducing decoding, preprocessing, and memory-transfer overheads. Experimental results on the VisDrone2019-DET dataset and the NVIDIA Jetson Orin NX demonstrate that DUST-YOLO achieves 43.7% mAP@0.5 accuracy with an end-to-end latency of 36.3 ms and a throughput of 27.5 FPS. Compared with the state of the art, DUST-YOLO reduces end-to-end latency by 56.9% and improves end-to-end video throughput by $2.31\times$. Full article

Alpha/beta (α/β) pulse shape discrimination (PSD) in liquid scintillation counting (LSC) is fundamentally limited by the charge comparison method (CCM) at low energies, where the entire tritium (³H) beta spectrum resides (0–18.6 keVee). The CCM figure-of-merit drops below 0.6 in this region, rendering it inadequate for simultaneous tritium and natural uranium alpha monitoring in nuclear power plant (NPP) liquid effluents. We present a one-dimensional convolutional neural network (1D-CNN) trained on an 80,000-waveform physics-based simulation dataset using established scintillation parameters for Ultima Gold AB. The proposed network achieves 97.4% overall classification accuracy and an area under the receiver operating characteristic curve (AUC) of 0.9981 on the held-out test set, representing improvements of 13.8 percentage points and 0.046 AUC over CCM. In the critical 0–18.6 keVee region, CNN accuracy exceeds 95% compared to below 60% for CCM—a greater than 35 percentage point improvement. Pulse amplitude discrimination (PAD), evaluated as a preliminary screening method, exhibits a 6.3% alpha spillover rate into the beta window, exceeding the regulatory limit of 3%. Gradient-weighted class activation maps (Grad-CAM) confirm that the network exploits physically meaningful pulse features rather than simulation artefacts. A comprehensive background suppression strategy combining dual-SiPM coincidence (24× reduction), anti-coincidence guard detector (5.8× reduction), composite passive shielding (10× reduction), and CNN-assisted discrimination reduces the system equivalent background to 1.83 ± 0.12 cpm, yielding a tritium minimum detectable activity (MDA) of 0.21 Bq/mL (10 mL sample, 30 min count), which satisfies the GB 14587 reference limit of 0.5 Bq/mL. After 8-bit post-training quantisation, the model achieves sub-microsecond inference latency on an embedded Xilinx Artix-7 Field-programmable gate array(FPGA), enabling real-time deployment in portable online monitoring systems. Full article

Selective harvesting of greenhouse cucumbers requires accurate detection with low inference latency. In greenhouse canopies, mature cucumbers are often partly occluded and visually similar to surrounding stems and leaves, which makes harvestability recognition difficult. Existing real-time detectors still struggle to preserve fine boundary cues, capture long-range context, and remain compact enough for on-device inference under these conditions. This study proposes MCS-DETR, an efficient multi-scale context-aware detector built on RT-DETR. Instead of increasing model scale, MCS-DETR redesigns shallow feature extraction, high-level contextual interaction, and cross-scale feature aggregation within a compact framework. A shallow feature level is also retained to preserve fine contour information. On the greenhouse cucumber dataset, MCS-DETR achieved 93.4% mAP@0.5 and 76.8% mAP@0.5:0.95, outperforming RT-DETR-R18 while requiring fewer parameters and less computation. On an NVIDIA Jetson Orin NX Super (Hunan ChuangLebo Intelligent Technology Co., Ltd., Room 2003, Building C, Xinchanghai Digital Center, Changsha Economic and Technological Development Zone, Changsha, Hunan, China) platform, it reached 26.3 FPS after TensorRT acceleration. These results indicate that MCS-DETR can provide an efficient on-device perception module for real-time greenhouse cucumber detection. Full article

Tomato leaf diseases substantially reduce tomato yields and quality and remain a persistent challenge for efficient crop management. Although deep learning-based detectors have achieved strong accuracy in controlled benchmarks, many existing solutions are still difficult to transfer to resource-constrained agricultural systems because they rely on high-end GPUs, consume considerable power, and often lose performance after deployment on embedded devices. To address this practical gap, this study proposes HGS-YOLO, a system-oriented deployable lightweight adaptation of YOLOv11 for leaf-level tomato disease detection, together with an end-to-end edge sensing pipeline for low-power agricultural deployment. The main contribution lies in the coordinated system-level co-design of model structure, optimization, and deployment rather than in a novel detector architecture. Specifically, YOLOv11 is adapted through three coordinated modifications: an HGNetV2 backbone for efficient feature extraction, an HS-FPN neck with channel attention for lightweight multi-scale fusion, and an MPDIoU loss function for more stable localization optimization. Beyond the model architecture, the study establishes a complete engineering pipeline that includes training, optimization, post-training quantization, and hardware deployment with BPU acceleration on a D-Robotics RDK X5 handheld platform. Comprehensive benchmark experiments indicate that HGS-YOLO achieves 93.6% ${\mathrm{mAP}}_{50}$ and 72.1% mAP@[0.5:0.95] with 86.5% recall, only 1.3 M parameters, and a 3.1 MB model size, substantially reducing the model complexity and storage cost relative to the YOLOv11 baseline. A three-seed retraining comparison shows that HGS-YOLO trades roughly 0.5 ${\mathrm{mAP}}_{50}$ points for this compactness (a statistically significant but small concession) and recovers the cost on the deployment side: on the RDK X5 chip, HGS-YOLO is the fastest, most memory-efficient, and lowest-power model among all compared detectors. Indoor deployment tests using separately collected tomato leaf samples further achieve 90.3% ${\mathrm{mAP}}_{50}$, 82.3% recall, 89.0% precision, 25.0 ± 0.4 ms end-to-end latency, 40.0 ± 0.6 FPS, and 9.8 ± 0.4 W average system power. After PTQ, the ${\mathrm{mAP}}_{50}$ drops from 93.6% to 93.0% on the same benchmark; because this figure was measured under controlled imaging conditions, it is presented as an in-distribution reference point rather than as evidence of robustness in the open field. We also took the handheld system into a working tomato greenhouse for a small outdoor field round, where it ran end-to-end and produced on-device disease detections under natural sunlight, specular highlights, partial occlusion, background clutter, and handheld motion blur. These results show that HGS-YOLO reaches a good balance of accuracy, efficiency, and deployability and that it works in the field on an independent small-scale test; validating it more widely across sites, seasons, and weather is left to future work. Full article

Edge computing on unmanned aerial vehicles (UAVs) enables low-latency wildfire monitoring by performing visual inference onboard; however, practical deployment is constrained by limited labeled data and resource budgets that often preclude reliance on large GPU servers. This work investigates transfer learning (TL) for UAV-based wildfire smoke and flame detection and evaluates its impact on both detection accuracy and edge deployment performance. We introduce the Aerial Fire and Smoke Essential (AFSE) dataset (282 aerial-view images; classes—smoke and fire), compiled from publicly available wildfire footage and FLAME2. Lightweight YOLO models are fine-tuned using heterogeneous (MS COCO) and homogeneous (FASDD) source pretraining and are assessed using mAP@0.5 together with frames per second (FPS), average inference power, energy consumption, and the normalized energy–delay product (EDP) on an edge computing platform. Results show that TL substantially improves detection accuracy on AFSE, achieving up to 79.2% mAP@0.5, while reducing training time, and improving cross-validation stability. On the tested edge platform, TL does not materially change inference speed or energy use, indicating that accuracy gains from TL do not automatically translate to improved efficiency without additional optimization. Among the evaluated lightweight detectors, YOLOv5n achieves the best mAP@0.5 while maintaining the highest edge device throughput, processing images nearly twice as fast as YOLO11n without hardware acceleration. More broadly, the measured throughput and energy differences among lightweight YOLO variants show that edge model selection should be guided by application-specific accuracy, latency, and energy constraints. Full article

Forward-looking sonar (FLS) is an important sensing modality for autonomous underwater vehicles and other marine robotic systems operating in turbid, low-visibility, and acoustically cluttered environments. Reliable target detection in FLS imagery remains challenging because target echoes are often weak, compact targets can be obscured by background clutter, and embedded processors impose strict limits on model size, latency, and computation. To address these issues, this study presents a lightweight FLS sensing framework for embedded target detection in resource-constrained underwater systems. The framework combines a compact detection architecture, difficulty-aware supervision, and teacher–student knowledge transfer. Specifically, FPN-Mix is developed as a lightweight backbone with a Conv-Mix module to improve contextual aggregation under limited computational budgets. A target-aware dynamic weighting loss is introduced to increase the supervision weight of difficult acoustic samples associated with weak echoes, ambiguous boundaries, and clutter interference. A multi-level knowledge distillation strategy is then adopted to transfer feature-level and prediction-level knowledge from an enhanced teacher model to the compact student detector. Experiments on the public UATD benchmark and the independently collected Zhanjiang Bay No.1 field dataset show that the proposed method achieves a favorable balance between detection accuracy and efficiency and remains competitive in a real marine aquaculture environment. The proposed model contains only 2.83 M parameters and requires 6.68 GFLOPs. After ONNX export and TensorRT FP16 acceleration, the model reaches 72.23 frames per second (FPS) on an NVIDIA Jetson Orin NX platform, supporting its practical use in embedded FLS sensing systems. Full article

The Internet of Medical Things (IoMT) has become integral to modern healthcare, yet its always-connected and resource-constrained nature enlarges the attack surface and complicates timely intrusion detection. This study presents a deployment-oriented, two-stage anomaly-detection pipeline. First, Principal Component Analysis (PCA) is employed to reduce the dimensionality of network traffic data, capturing the most significant variance. Subsequently, a One-Class Support Vector Machine (OC-SVM) is trained exclusively on these principal components of normal traffic. This approach prioritizes computational efficiency for resource-constrained IoMT devices while maintaining high model robustness. By modeling the principal components of normal behavior, our method achieves state-of-the-art performance across diverse attack families. We adopt a uniform protocol across four public IoMT corpora—BoT-IoT, CICIoMT2024, ECU-IoHT, and IoMT-TrafficData. The model’s hyperparameters, including the optimal number of principal components determined by explained variance, are tuned via randomized search. Despite using no attack labels during training, the proposed PCA-enhanced detector achieves state-of-the-art performance across diverse attack families: on BoT-IoT we obtain 99.92% F1-score (99.84% accuracy), on CICIoMT2024 we obtain 99.88% F1-score (99.77% accuracy), on ECU-IoHT 99.25% F1-score (98.58% accuracy), and on IoMT-TrafficData 99.19% F1-score (98.66% accuracy). The compact model size, enabled by PCA, makes the approach highly amenable to edge or gateway deployment in clinical networks, while the normal-only training paradigm improves robustness to zero-day threats. The results demonstrate that modeling the principal components of routine network behavior is a highly effective and efficient strategy for reliable, low-latency threat detection in realistic IoMT settings. Full article

Autonomous driving perception demands low latency, high temporal resolution, and stringent hardware efficiency. While event-based spiking neural networks (SNNs) offer bio-inspired sparse computation, their deployment on edge field-programmable gate arrays (FPGAs) is obstructed by irregular execution patterns and temporal state storage overhead. To address this, we propose HAPQ, a unified hardware-aware pruning and quantization pipeline for compact event-based object detection. Starting from an end-to-end adaptive sampling SNN detector (EAS-SNN), HAPQ conducts hardware-aware configuration search within discrete digital signal processor (DSP) and block RAM (BRAM) budgets, applies single-instruction-multiple-data (SIMD)-aligned structured pruning for computational regularity, and jointly quantizes synaptic weights and membrane potentials via a shift-friendly fixed-point recurrence. Evaluation on the Prophesee Gen1 dataset and an FPGA accelerator shows that HAPQ improves detection accuracy from 0.284 to 0.425 in mean average precision ($mA{P}_{50:95}$) and achieves 0.722 $mA{P}_{50}$. Hardware implementation reveals a reduction in lookup table (LUT) usage to 1680, complete DSP elimination, and a maximum operating frequency of 920.81 MHz at 0.630 W. These results confirm that effective temporal SNN deployment requires joint optimization of model architecture, state precision, and hardware-aligned workload organization. Full article

Sleep disorders, such as insomnia, obstructive sleep apnea (OSA), narcolepsy, REM sleep behavior disorder, and circadian rhythm disturbances, represent a rapidly expanding global health burden that is strongly associated with cardiovascular, metabolic, neurological, and psychiatric diseases. Advancements in wearable sensing technologies and Internet of Medical Things (IoMT) infrastructures have expanded the possibilities for continuous, home-based sleep assessment beyond conventional polysomnography laboratories. These Sleep Health Internet of Things (S-HIoT) systems combine multimodal physiological sensing (EEG, ECG, SpO₂, respiratory effort and actigraphy) with wireless communication and cloud-based analytics for automated sleep-stage classification and disorder detection. Nonetheless, the digitization of sleep medicine brings about significant cybersecurity concerns. The constant transmission of sensitive biomedical information makes S-HIoT networks open to anomalous traffic flows, signal manipulation, replay attacks, spoofing, and data integrity violation. Existing studies mostly focus on analyzing physiological signals and network intrusion detection independently, resulting in a systemic vulnerability of cyber–physical sleep monitoring ecosystems. With the aim of addressing this empirical deficiency, this review integrates emerging advances (2022–2026) in the AI-assisted categorization of sleep phases and IoMT anomaly detector designs on the finer analysis of CNN, LSTM/BiLSTM, Transformer-based systems, and a component part of federated schemes and the lightweight, edge-deployable intruder assessor models available. The aim of this study is to uncover a gap in the literature: integrated architectures to trade off audiences of faithfulness of physiological modeling with communication-layer security. To counter it, we present a single framework to include CNN-based spatial feature extraction, Bidirectional Long Short-Term Memory (BiLSTM)-based temporal models and Random Forest-based ensemble classification using a dual task-learning approach. We propose a multi-objective optimization framework to jointly optimize the performance of sleep-stage prediction and that of network anomaly detection. Performance on publicly available datasets (Sleep-EDF and CICIoMT2024) confirms that hybrid integration can be tailored to achieve high accuracy [99.8% sleep staging; 98.6% anomaly detection] whilst being characterized by low inference latency (<45 ms), which is promising for feasibility in real-time deployment in view of targeting edge devices. This work presents a comprehensive framework for developing secure, intelligent, and clinically robust digital sleep health ecosystems by bridging chronobiological signal modeling with cybersecurity mechanisms. Furthermore, it highlights future research directions, including explainable AI, federated secure learning, adversarial robustness, and energy-aware edge optimization. Full article

Industrial Internet of Things (IIoT) and SCADA-connected networks are increasingly vulnerable to Distributed Denial of Service (DDoS) attacks, which can disrupt time-sensitive industrial processes and compromise operational continuity. Effective mitigation requires accurate and low-latency attack detection at the network edge, where industrial gateways operate under strict constraints in computation, memory, and energy. This study investigates Deep Reinforcement Learning (DRL) for real-time binary DDoS detection and proposes a detector based on Proximal Policy Optimisation (PPO) for deployment in resource-constrained IIoT environments. Four DRL agents, namely Deep Q-Network (DQN), Double DQN, Dueling DQN, and PPO, are trained and evaluated within a unified experimental pipeline incorporating automatic label mapping, numerical feature selection, robust scaling, and class balancing. Experiments are conducted on three representative benchmark datasets: CIC-DDoS2019, Edge-IIoTset, and CICIoT23. Performance is assessed using accuracy, precision, recall, F1-score, false positive rate, false negative rate, and CPU inference latency. The reward function is asymmetric: +1 for correct classification, −1 for false positive, and −2 for false negative, penalising missed attacks more heavily for IIoT safety. The results show that PPO provides a competitive accuracy–latency tradeoff across all three datasets, achieving the highest mean accuracy of 97.65% and ranking first on CIC-DDoS2019 with a score of 95.92%, while remaining competitive on Edge-IIoTset (99.11%) and CICIoT23 (97.92%). PPO also converges faster than the value-based baselines. Inference latency is below 0.8 ms per sample on a standard CPU (Intel i7-11800H), confirming real-time feasibility. To support practical deployment, the trained PPO policies are exported to ONNX format (≈9 KB per model), enabling lightweight and PyTorch-independent inference on industrial edge gateways. Full article

The rapid growth of 5G necessitates wireless receivers capable of high-speed, low-latency communication under complex channel conditions. Traditional receivers struggle with the performance–complexity trade-off in massive MIMO systems, where linear detectors underperform and maximum likelihood (ML) detection becomes computationally prohibitive. Deep-learning-based model-driven approaches have demonstrated a favorable balance between detection performance and computational cost. However, despite their algorithmic promise, the transition of these learned detectors into practical, real-time systems is critically hampered by inefficient hardware mapping, resulting in suboptimal throughput, high resource overhead, and limited scalability. To bridge this gap, this paper presents LwAMP-Net, a dedicated FPGA accelerator for a lightweight learned AMP detector. We propose a modular and multi-mode hardware architecture for LwAMP-Net, featuring an outer-product-based dataflow that mitigates pipeline stalls and multi-mode processing elements that adapt to diverse computation patterns. These innovations jointly enhance computational parallelism and resource utilization on the FPGA. Implemented on a Xilinx XC7VX690T FPGA for a 128 × 8 MIMO system with 16QAM, the accelerator achieves a 49.2% higher normalized throughput per iteration, an 85.4% improvement in throughput per LUT slice, and a 12.7% improvement in throughput per DSP compared to the state-of-the-art methods. This work provides a complete architectural solution for deploying high-performance, hardware-efficient learned MIMO detectors in real-world systems. Full article

The automatic reading recognition of digital instruments is crucial for achieving metering automation and intelligent inspection. However, in non-standardized industrial environments, the masking effect caused by the coupling of non-uniform low-light conditions and the reflective surfaces of instrument panels severely degrades the displayed information, significantly limiting the recognition performance. Conventional image processing methods, while aiming to restore the imaging quality of instrument panels through low-light enhancement, inevitably introduce overexposure and indiscriminately amplify background noise during this process. To address the two key challenges of illumination recovery and noise suppression in the process of restoring panel image quality under non-uniform low-light conditions, this paper proposes a coarse-to-fine cascaded perception framework (CFCP). First, a lightweight YOLOv10 detector is employed to coarsely localize the meter reading region under non-uniform illumination conditions. Second, an Adaptive Illumination Correction Module (AICM) is designed to decouple and correct the illumination component at the pixel level, effectively restoring details in dark areas. Then, an Illumination-invariant Feature Perception Module (IFPM) is embedded at the feature level to dynamically perceive illumination-invariant features and filter out noise interference. Finally, the refined detection results are fed into a lightweight sequence recognition network to obtain the final meter readings. Experiments on a self-built industrial digital instrument dataset show that the proposed method achieves 93.2% recognition accuracy, with 17.1 ms latency and only 7.9 M parameters. Full article

Continuous healthcare monitoring systems generate non-stationary physiological data streams, where evolving statistical properties and patterns often invalidate static models and fixed user classifications. To address this challenge, we propose drift-aware adaptive architecture that integrates concept drift detection into a distributed edge–cloud data analytics pipeline. In the proposed design, a concept drift is elevated from a maintenance signal to the primary mechanism governing user-state adaptation, model evolution, and inference consistency. Within the proposed system, the edge tier performs low-latency inference and preliminary drift screening under strict resource constraints, while the cloud tier executes advanced drift detection and validation, orchestrates user reclassification and model retraining, and manages model evolution. A feedback loop synchronizes edge and cloud operations, ensuring that detected drift triggers appropriate system transitions, either reassigning a user to an updated state category or initiating targeted model updates. This architecture reduces reliance on static group assignments, improves personalization, and preserves model fidelity under evolving physiological conditions. We analyze the drift types most relevant to healthcare data streams, evaluate the suitability of lightweight and cloud-grade drift detectors, and define the system requirements for stability, responsiveness, and clinical safety. Evaluation across 21 concurrent users demonstrates that drift-aware adaptation reduced prediction MAE by 40.6% relative to periodic retraining, with an end-to-end adaptation latency of 66 ± 37 s. Hierarchical cloud validation reduced the false-positive retraining rate from 88.9% (edge-only triggering) to 27.3%, while maintaining uninterrupted inference throughout all adaptation events. Full article