Search Results (194)

Satellite -derived environmental features can predict county-level life expectancy (LE) across the contiguous United States with a mean absolute error of 1.08 years over two decades, without using any census or sociodemographic inputs. We assembled 61,680 county-year observations across 3084 counties from 2000–2019, integrating features from 11 satellite and gridded data streams. The data streams include the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface temperature and vegetation indices, Sentinel-1 synthetic aperture radar, Sentinel-2 and Landsat optical imagery, the United States Department of Agriculture (USDA) Cropland Data Layer, the European Commission Joint Research Centre (JRC) Global Surface Water layer, the Copernicus Digital Elevation Model, the European Space Agency Climate Change Initiative (ESA CCI) soil moisture record, and the Food and Agriculture Organization (FAO) gridded livestock densities. After a supervised pruning step that removed low-importance variables, a Random Forest regressor was trained and evaluated using 5-fold cross-validation grouped by county. The grouping places all 20 years of each county exclusively in either the training set or the test set, which prevents spatial information leakage between folds. Coefficient of determination, mean absolute error, and root mean squared error are reported as $R^2=0.631\pm 0.013$, MAE $=1.08\pm 0.02$ years, and RMSE $=1.48\pm 0.04$ years. Moran’s I, a measure of residual spatial autocorrelation, is $0.0988$ ($p=0.001$), which supports geographic generalisation. Multimodal fusion reduces unexplained variance by approximately one-third relative to the strongest single-modality baseline (MODIS land surface temperature alone, $R^2=0.442$). TreeSHAP attribution analysis reveals a feature hierarchy in which nighttime land surface temperature features carry roughly $6.16\times$ the cumulative attribution weight of all daytime channels combined. The model response shows a protective inflection near a minimum overnight temperature of about 7.5 °C. Because all input streams are globally available, the framework is architecturally extensible to regions where civil registration and vital statistics systems are incomplete; however, the trained model and its thresholds require recalibration against local mortality data before application outside the contiguous United States. With that caveat, the approach supports satellite-based monitoring of United Nations Sustainable Development Goal (UN SDG) Target 3.9. Full article

To address the low efficiency of manual quality grading for Pleurotus ostreatus in factory-scale production and the difficulty existing computer vision models face in balancing high localization accuracy with real-time edge deployment for food processing, a lightweight non-destructive detection model named POC-DETR-Prune is proposed. Based on an improved RTDETR framework, FasterNet is introduced to optimize feature extraction, reducing memory access latency while ensuring deep feature representation for complex food morphologies. A Small Object Enhancement Pyramid (SOEP) module is designed to mitigate the loss of subtle features caused by dense mushroom clustering. Furthermore, the Inner-MPDIoU loss function is proposed to significantly improve bounding box localization accuracy in highly overlapped food sorting scenarios. To adapt to industrial hardware constraints, a Random channel pruning strategy compresses computational overhead. Experimental results demonstrate that POC-DETR-Prune achieves a mAP@0.5:0.95 of 83.7% with a computation load of only 38.2 GFLOPs. Deployment testing on the NVIDIA Jetson Orin Nano Super edge computing platform achieves a real-time detection rate of 30.2 FPS. This emerging technology provides a certain level of visual algorithm support for automated quality grading equipment in the edible fungi industry. Full article

In intensive aquaculture, non-invasive real-time monitoring of morphological traits and body weight of largemouth bass (Micropterus salmoides) is essential for precision feeding and yield estimation. Manual measurement is laborious and stressful, whereas vision-based methods are challenged by insufficient dual-view feature fusion, poor robustness to underwater noise, and over-parameterized models unsuitable for edge deployment. To address these issues, a lightweight framework, MECT-MobileViT, is proposed based on MobileViT-xxs. A Morphometric-Guided Multi-Scale Fusion module is designed to couple physical priors with dual-branch visual features, strengthening shape–weight association. An ECA-NL attention block employing instance normalization, GLU gating, and threshold filtering is embedded to enhance feature robustness against visual disturbances typical in aquaculture and to accentuate critical morphological features. A three-stage synergistic pruning strategy—attention head pruning, structured channel pruning, and depthwise separable attention substitution—is applied to achieve substantial compression while preserving representational capacity. Experiments on a self-built lateral–dorsal dual-view dataset show that the proposed model significantly outperforms mainstream benchmarks. The pruned version attains an R² of 0.8266 and an RMSE of 16.4201, with less than 2% accuracy degradation relative to the best unpruned model, and contains only 7.34 M parameters. This study demonstrates a promising prototype for contactless, stress-free weight estimation in largemouth bass and offers new technical insights into feature fusion, noise suppression, and collaborative model compression for aquaculture visual perception. Full article

Reliable condition monitoring of wind turbine gearboxes is critical to reducing unplanned downtime and maintenance costs in wind farms. However, this task presents significant challenges due to the non-stationary nature of vibration signals, in which fault-relevant features are sparsely and unevenly distributed across the time–frequency map. Although time–frequency analysis has been widely adopted to represent nonlinear and non-stationary vibration signals, existing deep learning methods typically process the full spectrogram directly, without distinguishing redundant or uninformative regions. This leads to high input dimensionality and exposes the model to substantial spectral noise. Consequently, it increases computational burden and potentially reduces the diagnostic reliability. To address this issue, this paper proposes a two-stage hybrid framework based on complementary selection mechanisms operating on two distinct feature spaces. In the first stage, eXtreme Gradient Boosting (XGBoost) importance scores are used to identify and permanently prune uninformative time–frequency features from the input spectrogram, reducing the input map size by 25%. In the second stage, a Squeeze-and-Excitation (SE) block, inserted after the deepest residual layer, performs soft channel-wise recalibration of the abstract feature maps produced by the residual convolutional neural network (ResCNN), thereby amplifying discriminative representations prior to classification. The proposed method was evaluated in an eight-class variable-speed fault classification task using the MCC5-THU benchmark, where data were collected from a 2.2 kW motor-driven gearbox test rig. The proposed method achieves a mean accuracy of 97.81% ± 0.33% under 5-fold stratified cross-validation (CV), while reducing classifier training time by approximately 23% compared to a baseline model trained on the full spectrogram. These results demonstrate that explicit input-level spectrogram pruning, combined with model-level channel attention, yields a robust and computationally efficient diagnostic framework for wind turbine gearbox condition monitoring. Full article

This study proposes an adaptive multi-mode graph search algorithm that integrates spatial previewing with terminal analytics to address node proliferation and terminal oscillation in path planning for four-wheel independent steering (4WIS) vehicles under complex, low-speed conditions. By employing line-of-sight checking and the Douglas–Peucker algorithm to extract the environmental topological skeleton, the proposed method generates Predictive Spatial Profiling (PSP) fields that precisely quantify channel safety margins. Departing from conventional soft-weight arbitration, a dynamic driving state machine leverages these rigid spatial constraints to deterministically prune redundant expansion branches—including Ackermann steering, crab steering, and in-place rotation—prior to node generation. Furthermore, a comprehensive cost function incorporating a mode-switching penalty and a gradient-heading heuristic is formulated to accelerate search convergence. To circumvent reliance on traditional empirical distance thresholds, a topology-triggered, multi-dimensional terminal analytical strategy is introduced, enabling a seamless transition from discrete search node expansion to continuous curve generation near the target. Extensive simulations demonstrate that the proposed algorithm reduces both the node expansion scale and optimization time by over 80% compared with conventional unconstrained methods, while effectively mitigating chaotic motion-mode transitions. Ultimately, integrating environmental spatial dimensionality reduction with terminal analytics yields a highly efficient and smooth global path-planning solution for 4WIS vehicles. 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

To enable the deployment of road distress detection models on resource-constrained embedded platforms, this paper presents a compression case study based on the LRDD-YOLOv8n detector designed for real-time 1080p video input. Layer-adaptive magnitude-based pruning (LAMP) was integrated with channel-wise knowledge distillation. First, LAMP performs structured pruning adaptive global sparsity allocation to reduce parameters and FLOPs. Then, a larger teacher model (LRDD-YOLOv8s) with high structural similarity guides the pruned student to recover feature representations. Compared to the baseline LRDD-YOLOv8n (64.4% mAP@0.5, 2.02 M parameters, 5.9G FLOPs, and 55.5 ms GPU inference time on Jetson TX2), our compressed model under a 1/1.4 target compression ratio achieves a mAP@0.5 of 65.1% (an slight accuracy increment of 0.7%), while reducing parameters by 36.1% (to 1.29 M) and FLOPs by 30.5% (to 4.1 G). Deployed on the BOXER-8120AI edge platform (Jetson TX2), the optimized model achieves an average inference time of 48.3 ms per frame (a 13.0% latency reduction compared to the baseline model). In addition, a 20 FPS detection rate was sustained under the 30 FPS maximum hardware acquisition limit of the industrial camera stream. Kinematic and geometric analysis validates that this processing rate utilizes 66.7% of all physically available frames and establishes a 95.4% consecutive frame-to-frame spatial overlap at typical inspection vehicle speeds (40–60 km/h). Full article

Birds, as essential components of biodiversity, serve as critical indicators of ecosystem health and stability through their population dynamics and spatial distribution. However, the complexity of natural habitats, occlusion caused by avian behavior, and logistical challenges in field monitoring pose significant difficulties for fine-grained bird detection. To address these issues, this paper presents NEB-YOLO, a lightweight bird detection network built upon an improved YOLOv8 (You Only Look Once version 8) architecture. First, to enhance detection capability for fine-grained bird images in complex backgrounds, an Efficient Multi-scale Attention (EMA) mechanism is integrated into the Neck network of the YOLOv8n architecture to construct the teacher network. Second, to accommodate resource constraints in practical scenarios, channel-wise Group_Slim pruning is applied to reduce the parameter count and computational overhead of the student model. Third, to achieve high accuracy with lightweight models, a joint offline distillation approach is adopted that combines logic-based knowledge distillation (BCKD) with feature-based knowledge distillation (CWD). This design facilitates effective transfer of discriminative features for bird imagery from the teacher model to the student model. Experiments demonstrate that the proposed framework achieves a performance gain of over 1.3% in $mAP$ on the fine-grained CUB-200-2011 bird dataset, while reducing model size by 67% compared to the original model. These results confirm that NEB-YOLO strikes an effective balance between model size and accuracy, indicating its suitability for resource-constrained scenarios. Full article

The operational health state of motor bearings is critical to the operational safety of harbor portal slewing cranes. However, in harsh industrial environments with strong noise and time-varying rotational speeds, existing bearing fault diagnosis methods still suffer from the problems of incomplete fault feature extraction from single-sensor signals and the excessively large size of multi-source fusion models, which makes them unable to adapt to edge deployment. To address these issues, this paper proposes a Multi-source Feature Fusion Lightweight Network (MTFL-Net) integrated with targeted structured channel pruning. First, vibration and current signals are preprocessed via differentiated time-frequency transformation and converted into 2D time-frequency images, to fully preserve transient impact and spectral fault features. Second, a multi-branch feature extraction architecture embedded with residual connections, multi-scale convolution and channel attention gating is designed, to alleviate feature degradation and adaptively enhance fault-sensitive features. Third, targeted structured channel pruning is performed on the feature extraction branches, to remove redundant channels while retaining the multi-source fusion logic and core feature extraction structure. Experiments on two public bearing datasets show that the original model achieves 99% diagnostic accuracy, and the pruned model still maintains an accuracy of 95%. The results demonstrate that MTFL-Net can significantly reduce model size and computational cost while retaining high diagnostic precision. Full article

Metro Computer Interlocking Systems are crucial for ensuring the safe and efficient operation of rail transit. However, existing vision-based signal detection methods face challenges including small target sizes, high target density, low image resolution, and the need for deployment on resource-constrained devices. To address these issues, this paper proposes a two-stage lightweight signal detection framework for Metro Computer Interlocking Systems. First, based on YOLOv8, a small object detection layer together with feature fusion modules is introduced to form the YOLOv8-SFF architecture. A Scale Sequence Feature Fusion (SSFF) module is added to adjust the resolution of feature maps and retain critical fine-grained information, enhancing the detection of small visual signals. A Triple Feature Encoding (TFE) module is designed to enhance the recognition of dense small signals while replacing some traditional feature concatenation and upsampling operations, yielding a more compact network. Second, to enable practical deployment, a joint optimization strategy combining Layer-Adaptive Magnitude-based Pruning (LAMP) and Channel-wise Knowledge Distillation (CWD) is applied, in which the unpruned YOLOv8-SFF serves as the teacher and the pruned model serves as the student. In addition, an automatic annotation subsystem based on digital image processing is developed, leveraging color and morphological features to generate high-quality labels. Experimental results show that YOLOv8-SFF achieves a mean average precision (mAP@50) of 98.7%, improving mAP@50 by 11.3 percentage points and recall by 22.1 percentage points over the original YOLOv8. After joint pruning and distillation, the final compact model retains 98.0% mAP while reducing the parameter count by 86.2% and the model size to 1.1 MB, making it well suited for real-time deployment in resource-constrained metro dispatching systems. Full article

To address challenges such as severe occlusion caused by the dense growth of blueberry fruits in natural environments, complex backgrounds, and the limited computational resources of agricultural edge devices, this study proposes BR-DETR-Prune, a lightweight object detection model oriented towards edge computing environments. Based on the RT-DETR architecture, the model introduces a PConv-based FasterNet as the backbone network, which effectively reduces memory access latency and floating-point operation costs. Furthermore, it utilizes a “Gather-and-Distribute” (GD) mechanism to reconstruct the feature fusion neck. Through the unified aggregation and multi-branch distribution of global information, it significantly enhances the model’s feature extraction capability for dense and overlapping targets. An AIFI-RepBN encoder is designed, integrating re-parameterization technology into the attention module to further reduce computational redundancy. For lightweight processing, a random channel pruning strategy based on the “Lottery Ticket Hypothesis” is adopted to perform structural compression and fine-tuning on the model, achieving a significant reduction in the number of parameters while inversely improving accuracy. The experimental results demonstrate that BR-DETR-Prune achieves an mAP@0.5 of 97.1% on a self-built blueberry dataset, with only 15.52 M parameters and a computational load reduced to 34.0 GFLOPs. Its comprehensive performance is superior to mainstream models such as YOLOv8, YOLO11, and the original RT-DETR. Particularly, deployment testing on the NVIDIA Jetson Orin Nano Super embedded edge computing platform reveals that the model achieves a real-time inference speed of 20.5 FPS under FP16 precision, exhibiting smooth detection frames and strong robustness against occlusion. This study provides an effective optimization solution for the deployment of high-precision Transformer architectures on low-computational-power devices, offering an efficient and reliable visual perception approach for automated blueberry harvesting and yield estimation. Full article

Unmanned aerial vehicles (UAVs) have wide application prospects in disaster search and rescue, ecological monitoring and environmental inspection tasks, where target search is a key link to realize autonomous task execution. UAVs often face challenges related to limited onboard computational resources and inefficient environmental coverage when used for target search. To address these issues, this paper proposes an autonomous search method for UAVs based on combined lightweight target detection and coverage path planning. In this method, the target search task was decomposed into two core parts: target recognition and path planning. Firstly, in terms of target recognition, the YOLOv8n model was subjected to channel pruning and INT8 quantization to reduce its computational complexity, while HSV space data augmentation was incorporated to enhance recognition robustness in complex environments. Secondly, path planning was formulated as a dual-layer task comprising “spatial coverage + target confirmation.” A grid-based search environment model was constructed, and a coverage path planning strategy was put forward that integrated breadth-first search (BFS) with local greedy optimization to achieve efficient traversal of predefined search areas. Simultaneously, the A* algorithm was employed for path backtracking to cover omitted regions. Finally, a simulation platform for UAV target search was built to validate the recognition performance and search efficiency of the proposed method. The experimental results demonstrated that the proposed method significantly improved the UAV target search efficiency and reduced the path redundancy while ensuring the recognition accuracy, thereby offering an effective solution for autonomous UAV search on resource-constrained embedded platforms. Full article

Environmental sound classification is an important component of smart city sensing systems, supporting applications such as urban noise analysis, public safety monitoring, and real-time situational awareness. However, high-accuracy models are often difficult to deploy on low-power edge devices because of memory, computational, and latency constraints. This study aims to address this deployment gap by developing a lightweight compression pipeline for a hybrid convolutional and Kolmogorov–Arnold Network-based model. The proposed pipeline consists of three stages. First, structured channel pruning is applied to remove redundant convolutional filters while preserving hardware-efficient dense operations. Second, selective quantization-aware training is applied to the most computation-dominant layers, namely the third convolutional layer and the fully connected layer. Third, knowledge distillation is used to recover accuracy by training the compressed model under the guidance of the baseline model. Experiments were conducted on ESC-10, ESC-50, FSC22, and UrbanSound8K. The proposed pipeline reduced the average parameter count from 511,033 to 50,774 and reduced the model size while maintaining competitive accuracy across all benchmarks. The final model preserved the baseline accuracy of 96.75% on ESC-10, while accuracy decreased only from 88.25% to 86.50% on ESC-50, from 87.92% to 86.38% on FSC22, and from 85.13% to 84.52% on UrbanSound8K. These results show that the proposed compression pipeline provides an effective accuracy–efficiency trade-off for real-time audio classification on resource-constrained devices. Therefore, the resulting compressed model supports the scalable deployment of distributed acoustic sensing systems for real-time smart city monitoring and decision-making. Full article

Real-time psychological stress detection on wearable and edge devices requires models that are accurate, computationally efficient, and small enough for on-device deployment. This paper proposes a Micro-Attention CNN Hybrid Architecture for stress recognition using wearable bio-signals. The model uses six sensor channels, namely tri-axial acceleration, electrodermal activity, heart rate, and skin temperature, and classifies three stress levels: no stress, low stress, and high stress. This study is conducted on a public wearable sensor dataset collected from 15 nurses during hospital work, providing a realistic benchmark for continuous stress monitoring under practical conditions. The proposed architecture combines one-dimensional and depthwise separable convolutions with a lightweight attention module to emphasize the most informative temporal patterns in short multivariate signal segments. To support deployment on resource-constrained devices, we further apply structured pruning, selective quantization-aware training, and post-training quantization. The full-precision model achieves a Macro-F1 score of 99.63%, while the final compressed model retains 98.03% Macro-F1 with a model size of 1.76 kilobytes and a CPU inference latency of 0.40 ms. Additional analyses show that most residual errors occur near the boundary between low stress and neighboring classes, while simple post-compression calibration improves reliability. These results demonstrate that accurate and low-latency stress detection using wearable bio-signals is feasible on compact edge hardware without transmitting raw sensor streams off-device. Full article

Real-time and accurate crop row extraction is a fundamental requirement for vision-based perception in autonomous agricultural machinery. In maize fields, however, row detection is easily affected by variable illumination, leaf occlusion, weed interference, and uneven soil backgrounds, which can reduce the reliability of both GNSS- and image-based navigation methods. To address these challenges, this study proposes a plant-oriented crop row perception framework that reconstructs row structures from individual maize plant detections. A lightweight detection model, named Maize–YOLO, was developed based on YOLOv11n for maize seedling detection. Three key improvements were introduced to enhance the balance between accuracy and efficiency. First, the C3k2_Faster_CGLU module replaces the original C3k2 block to reduce redundant convolutional computation while improving selective feature representation through convolutional gated linear units, thereby enhancing robustness under complex field backgrounds. Second, a lightweight shared detection head, Detect_LSH, was designed to share convolutional parameters across multi-scale feature maps and adaptively adjust feature amplitudes, reducing detection-head redundancy while maintaining multi-scale prediction capability. Third, a Layer-Adaptive Magnitude-Based Pruning strategy was applied to remove low-contribution channels and further improve computational efficiency for CPU-based deployment. Experimental results on field-collected maize seedling images showed that Maize–YOLO achieved an mAP@0.5 of 97.6%, reduced GFLOPs by 61.9%, and maintained a CPU inference speed of 84.4 FPS. After plant detection, row centerlines were estimated using an IPM–DBSCAN–LSM pipeline, which transformed detected plant centers into a quasi-top-view space, clustered them into crop rows, and fitted continuous centerlines. The extracted crop rows reached a positional accuracy of 98.6%, with a mean angular deviation of 0.44°. These results demonstrate that the proposed method can provide accurate, lightweight, and real-time crop row perception for autonomous agricultural navigation and precision field operations. Full article