Search Results (1,672)

Estimating the relative position and velocity of micro aerial vehicles (MAVs) using visual signals is a critical issue in numerous tasks. However, traditional relative motion estimation algorithms suffer severely from non-Gaussian noise interference and have limited observability, making it difficult to meet the practical requirements of complex dynamic scenarios. To address this dilemma, this paper proposes a Multimodal Decoupled Spatiotemporal Adaptive Network (MDSAN). Designed for air-to-air scenarios, MDSAN achieves high-precision relative pose and velocity estimation of dynamic MAVs while overcoming the observability limitations of traditional algorithms. In detail, MDSAN is collaboratively composed of two core sub-modules: Modality-Specific Convolutional Normalization (MSCN) blocks and Spatiotemporal Adaptive State (STAS) blocks. Specifically, MSCN uses custom convolution kernels tailored to three modalities—visual, physical, and geometric—to separate their features. This prevents interference between modalities and reduces non-Gaussian noise. STAS, built on a state-space model, combines two key functions: it tracks long-term MAV motion trends over time and strengthens the synergy between different modal features across space. Adaptive weights balance these two functions, enabling stable estimation, even when traditional methods struggle with low observability. Furthermore, MDSAN adopts a full-vision multimodal fusion scheme, completely eliminating the dependence on wireless communication and reducing hardware costs. Extensive experimental results demonstrate that MDSAN achieves the best performance in all scenarios, significantly outperforming existing motion estimation algorithms. It provides a new technical path that balances high precision, high robustness, and cost-effectiveness for technologies such as MAV swarm perception. Full article

Accurate wildlife detection in Unmanned Aerial Vehicle (UAV)-captured imagery is crucial for biodiversity conservation, yet it remains challenging due to the visual similarity of species, environmental disturbances, and the small size of target animals. To address these challenges, this paper introduces YOLO-WL, a wildlife detection algorithm specifically designed for UAV-based monitoring. First, a Multi-Scale Dilated Depthwise Separable Convolution (MSDDSC) module, integrated with the C2f-MSDDSC structure, expands the receptive field and enriches semantic representation, enabling reliable discrimination of species with similar appearances. Next, a Multi-Scale Large Kernel Spatial Attention (MLKSA) mechanism adaptively highlights salient animal regions across different spatial scales while suppressing interference from vegetation, terrain, and lighting variations. Finally, a Shallow-Spatial Alignment Path Aggregation Network (SSA-PAN), combined with a Spatial Guidance Fusion (SGF) module, ensures precise alignment and effective fusion of multi-scale shallow features, thereby improving detection accuracy for small and low-resolution targets. Experimental results on the WAID dataset demonstrate that YOLO-WL outperforms existing state-of-the-art (SOTA) methods, achieving 94.2% mAP@0.5 and 58.0% mAP@0.5:0.95. Furthermore, evaluations on the Aerial Sheep and AI-TOD datasets confirm YOLO-WL’s robustness and generalization ability across diverse ecological environments. These findings highlight YOLO-WL as an effective tool for enhancing UAV-based wildlife monitoring and supporting ecological conservation practices. Full article

Surface defect evaluation in steel production demands both high inference speed and accuracy for efficient production. However, existing methods face two critical challenges: (1) the diverse dimensions and irregular morphologies of surface defects reduce detection accuracy, and (2) computationally intensive feature extraction slows inference. In response to these challenges, this study proposes an innovative network based on dual-branch feature enhancement and downsampling (DFED-Net). First, an atrous convolution and multi-scale dilated attention fusion module (AMFM) is developed, incorporating local–global feature representation. By emphasizing local details and global semantics, the module suppresses noise interference and enhances the capability of the model to separate small-object features from complex backgrounds. Additionally, a dual-branch downsampling module (DBDM) is developed to preserve the fine details related to scale that are typically lost during downsampling. The DBDM efficiently fuses semantic and detailed information, improving consistency across feature maps at different scales. A lightweight dynamic upsampling (DySample) is introduced to supplant traditional fixed methods with a learnable, adaptive approach, which retains critical feature information more flexibly while reducing redundant computation. Experimental evaluation shows a mean average precision (mAP) of 81.5% on the Northeastern University surface defect detection (NEU-DET) dataset, a 5.2% increase compared to the baseline, while maintaining a real-time inference speed of 120 FPS compared to the 118 FPS of the baseline. The proposed DFED-Net provides strong support for the development of automated visual inspection systems for detecting defects on steel surfaces. Full article

This study aims to address the safety and mobility challenges faced by visually impaired individuals. To this end, a lightweight, high-precision semantic segmentation network is proposed for scenes containing tactile paving and zebra crossings. The network is successfully deployed on an intelligent guide robot equipped with a high-definition camera and a Huawei Atlas 310 embedded computing platform. To enhance both real-time performance and segmentation accuracy on resource-constrained devices, an improved G-GhostNet backbone is designed for feature extraction. Specifically, it is combined with a depthwise separable convolution-based Coordinate Attention module and a redesigned Atrous Spatial Pyramid Pooling (ASPP) module to capture multi-scale contextual features. A dedicated decoder efficiently fuses multi-level features to refine segmentation of tactile paving and zebra crossings. Experimental results demonstrate that the proposed model achieves mPA of 97% and 93%, mIoU of 94% and 86% for tactile paving and zebra crossing segmentation, respectively, with an inference speed of 59.2 fps. These results significantly outperform several mainstream semantic segmentation networks, validating the effectiveness and practical value of the proposed method in embedded systems for visually impaired travel assistance. Full article

Background: Major Depressive Disorder (MDD) is a pervasive psychiatric condition. Electroencephalography (EEG) is employed to detect MDD-specific neural patterns because it is non-invasive and temporally precise. However, manual interpretation of EEG signals is labor-intensive and subjective. This problem was addressed by proposing machine learning (ML) and deep learning (DL) methods. Although DL methods are promising for MDD detection, they face limitations, including high model complexity, overfitting due to subject-specific noise, excessive channel requirements, and limited interpretability. Methods: To address these challenges, we propose MS-MDDNet, a new lightweight CNN model specifically designed for EEG-based MDD detection, along with an ensemble-like method built on it. The architecture of MS-MDDNet incorporates spatial, temporal, and depth-wise separable convolutions, along with average pooling, to enhance discriminative feature extraction while maintaining computational efficiency with a small number of learnable parameters. Results: The method was evaluated using 10-fold Cross-Subjects Cross-Validation (CS-CV), which mitigates the risks of overfitting associated with subject-specific noise, thereby contributing to generalization robustness. Across three public datasets, the proposed method achieved performance comparable to state-of-the-art approaches while maintaining lower computational complexity. It achieved a 9% improvement on the MODMA dataset, with an accuracy of 99.33%, whereas on MUMTAZ and PRED + CT it achieved accuracies of 98.59% and 96.61%, respectively. Conclusions: The predictions of the proposed method are interpretable, with interpretability achieved through correlation analysis between gamma energy and learned features. This makes it a valuable tool for assisting clinicians and individuals in diagnosing MDD with confidence, thereby enhancing transparency in decision-making and promoting clinical credibility. Full article

This paper proposes a lightweight facial expression recognition model based on an improved Mini-Xception algorithm to address the issue of deploying existing models on resource-constrained devices. The model achieves lightweight facial expression recognition, particularly for elder-oriented applications, by introducing depthwise separable convolutions, residual connections, and a four-class expression reconstruction. These designs significantly reduce the number of parameters and computational complexity while maintaining high accuracy. The model achieves an accuracy of 79.96% on the FER2013 dataset, outperforming various other popular models, and enables efficient real-time inference in standard CPU environments. Full article

To address the challenges of insufficient efficiency and accuracy in traditional detection models caused by minute target sizes, low signal-to-noise ratios (SNRs), and feature volatility in Satellite Laser Ranging (SLR) images, this paper proposes an efficient, lightweight, and high-precision detection model. The core motivation of this study is to fundamentally enhance the model’s capabilities in feature extraction, fusion, and localization for minute and blurred targets through a specifically designed network architecture and loss function, without significantly increasing the computational burden. To achieve this goal, we first design a DMS-Conv module. By employing dense sampling and channel function separation strategies, this module effectively expands the receptive field while avoiding the high computational overhead and sampling artifacts associated with traditional multi-scale methods, thereby significantly improving feature representation for faint targets. Secondly, to optimize information flow within the feature pyramid, we propose a Lightweight Upsampling Module (LUM). Integrating depthwise separable convolutions with a channel reshuffling mechanism, this module replaces traditional transposed convolutions at a minimal computational cost, facilitating more efficient multi-scale feature fusion. Finally, addressing the stringent requirements for small target localization accuracy, we introduce the MPD-IoU Loss. By incorporating the diagonal distance of bounding boxes as a geometric penalty term, this loss function provides finer and more direct spatial alignment constraints for model training, effectively boosting localization precision. Experimental results on a self-constructed real-world SLR observation dataset demonstrate that the proposed model achieves an ${\mathrm{mAP}}_{50:95}$ of 47.13% and an F1-score of 88.24%, with only 2.57 M parameters and 6.7 GFLOPs. Outperforming various mainstream lightweight detectors in the comprehensive performance of precision and recall, these results validate that our method effectively resolves the small target detection challenges in SLR scenarios while maintaining a lightweight design, exhibiting superior performance and practical value. Full article

This paper proposes a robust convolutional neural network (CNN) architecture for human activity recognition (HAR) using smartphone accelerometer data, evaluated on the WISDM dataset. We introduce two novel pooling mechanisms—Pooling A (Extrema Contrast Pooling (ECP)) and Pooling B (Center Minus Variation (CMV))—that enhance feature discrimination and noise robustness. ECP emphasizes sharp signal transitions through a nonlinear penalty based on the squared range between extrema, while CMV Pooling penalizes local variability by subtracting the standard deviation, improving resilience to noise. Input data are normalized to the [0, 1] range to ensure bounded and interpretable pooled outputs. The proposed framework is evaluated in two separate configurations: (1) a 1D CNN applied to raw tri-axial sensor streams with the proposed pooling layers, and (2) a histogram-based image encoding pipeline that transforms segment-level sensor redundancy into RGB representations for a 2D CNN with fully connected layers. Ablation studies show that histogram encoding provides the largest improvement, while the combination of ECP and CMV further enhances classification performance. Across six activity classes, the 2D CNN system achieves up to 96.84% weighted classification accuracy, outperforming baseline models and traditional average pooling. Under Gaussian, salt-and-pepper, and mixed noise conditions, the proposed pooling layers consistently reduce performance degradation, demonstrating improved stability in real-world sensing environments. These results highlight the benefits of redundancy-aware pooling and histogram-based representations for accurate and robust mobile HAR systems. Full article

Ensuring food security requires innovative, sustainable pest management solutions. The Solar Insecticidal Lamp Internet of Things (SIL-IoT) represents such an advancement, yet its reliability in harsh, variable outdoor environments is compromised by frequent component and sensor faults, threatening effective pest control and data integrity. This paper presents a comprehensive survey on fault detection (FD) for SIL-IoT systems, systematically analyzing their unique challenges, including electromagnetic interference, resource constraints, data scarcity, and network instability. To address these challenges, we investigate countermeasures, including blind source separation for signal decomposition under interference, lightweight model techniques for edge deployment, and transfer/self-supervised learning for low-cost fault modeling across diverse agricultural scenarios. A dedicated case study, utilizing sensor fault data of SIL-IoT, demonstrates the efficacy of these approaches: an empirical mode decomposition-enhanced model achieved 97.89% accuracy, while a depthwise separable-based convolutional neural network variant reduced computational cost by 88.7% with comparable performance. This survey not only synthesizes the state of the art but also provides a structured framework and actionable insights for developing robust, efficient, and scalable FD solutions, thereby enhancing the operational reliability and sustainability of SIL-IoT systems. Full article

Fatigue crack propagation is a critical failure mechanism in engineering structures, requiring meticulous monitoring for timely maintenance. This research introduces a deep learning framework for estimating fatigue fracture length in metallic plates through acoustic emission (AE) signals. AE waveforms recorded during crack growth are transformed into time-frequency images using the Choi–Williams distribution. First, a clustering system is developed to analyze the distribution of the AE image-based dataset. This system employs a CNN-based model to extract features from the input images. The AE dataset is then divided into three categories according to fatigue lengths using the K-means algorithm. Principal Component Analysis (PCA) is used to reduce the feature vectors to two dimensions for display. The results show how close together the data points are in the clusters. Second, convolutional neural network (CNN) models are trained using the AE dataset to categorize fracture lengths into three separate ranges. Using the pre-trained models ResNet50V2 and VGG16, we compare the performance of a bespoke CNN using transfer learning. It is clear from the data that transfer learning models outperform the custom CNN by a wide margin, with an accuracy of approximately 99% compared to 93%. This research confirms that convolutional neural networks (CNNs), particularly when trained with transfer learning, are highly successful at understanding AE data for data-driven structural health monitoring. Full article

Reliable and low-latency seizure detection from electroencephalography (EEG) is critical for continuous clinical monitoring and emerging wearable health technologies. Spiking neural networks (SNNs) provide an event-driven computational paradigm that is well suited to real-time signal processing, yet achieving competitive seizure detection performance with constrained model complexity remains challenging. This work introduces a hybrid spike encoding scheme that combines Delta–Sigma (change-based) and stochastic rate representations, together with two spiking architectures designed for real-time EEG analysis: a compact feed-forward HybridSNN and a convolution-enhanced ConvSNN incorporating depthwise-separable convolutions and temporal self-attention. The architectures are intentionally designed to operate on short EEG segments and to balance detection performance with computational practicality for continuous inference. Experiments on the CHB–MIT dataset show that the HybridSNN attains 91.8% accuracy with an F1-score of 0.834 for seizure detection, while the ConvSNN further improves detection performance to 94.7% accuracy and an F1-score of 0.893. Event-level evaluation on continuous EEG recordings yields false-alarm rates of 0.82 and 0.62 per day for the HybridSNN and ConvSNN, respectively. Both models exhibit inference latencies of approximately 1.2 ms per 0.5 s window on standard CPU hardware, supporting continuous real-time operation. These results demonstrate that hybrid spike encoding enables spiking architectures with controlled complexity to achieve seizure detection performance comparable to larger deep learning models reported in the literature, while maintaining low latency and suitability for real-time clinical and wearable EEG monitoring. Full article

This paper presents a novel hypersector-based method with Fuzzy Learning Vector Quantization (FLVQ) for the real-time classification of wind turbine blade defects using data acquired by unmanned aerial vehicles (UAVs). Unlike conventional prototype-based FLVQ approaches that rely on Euclidean distance in the feature space, the proposed method models each defect class as a hypersector on an n-dimensional hypersphere, where class boundaries are defined by angular similarity and fuzzy membership transitions. This geometric reinterpretation of FLVQ constitutes the core innovation of the study, enabling improved class separability, robustness to noise, and enhanced interpretability under uncertain operating conditions. Feature vectors extracted via the pre-trained SqueezeNet convolutional network are normalized onto the hypersphere, forming compact directional clusters that serve as the geometric foundation of the FLVQ classifier. A fuzzy softmax membership function and an adaptive prototype-updating mechanism are introduced to handle class overlap and improve learning stability. Experimental validation on a custom dataset of 900 UAV-acquired images achieved 95% classification accuracy on test data and 98.3% on an independent dataset, with an average F1-score of 0.91. Comparative analysis with the classical FLVQ prototype demonstrated superior performance and noise robustness. Owing to its low computational complexity and transparent geometric decision structure, the developed model is well-suited for real-time deployment on UAV embedded systems. Furthermore, the proposed hypersector FLVQ framework is generic and can be extended to other renewable-energy diagnostic tasks, including solar and hydropower asset monitoring, contributing to enhanced energy security and sustainability. Full article

The aim of this study was to accurately meet the demand of real-time detection of seedling shortage in large-scale seedling production and solve the problems of low precision of traditional models and insufficient adaptability of mainstream lightweight models. This study proposed a Light-YOLO-Pepper seedling shortage detection model based on the improvement of YOLOv8n. This model was based on YOLOv8n. The SE (Squeeze-and-Excitation) attention module was introduced to dynamically suppress the interference of the nutrient soil background and enhance the features of the seedling shortage area. Depth-separable convolution (DSConv) was used to replace the traditional convolution, which can reduce computational redundancy while retaining core features. Based on K- means clustering, customized anchor boxes were generated to adapt to the hole sizes of 72-unit (large size) and 128-unit (small size and high-density) seedling trays. The results show that the overall mAP@0.5, accuracy and recall rate of Light-YOLO-Pepper model were 93.6 ± 0.5%, 94.6 ± 0.4% and 93.2 ± 0.6%, which were 3.3%, 3.1%, and 3.4% higher than YOLOv8n model, respectively. The parameter size of the Light-YOLO-Pepper model was only 1.82 M, the calculation cost was 3.2 G FLOPs, and the reasoning speeds with regard to the GPU and CPU were 168.4 FPS and 28.9 FPS, respectively. The Light-YOLO-Pepper model was superior to the mainstream model in terms of its lightweight and real-time performance. The precision difference between the two seedlings was only 1.2%, and the precision retention rate in high-density scenes was 98.73%. This model achieves the best balance of detection accuracy, lightweight performance, and scene adaptability, and can efficiently meet the needs of embedded equipment and real-time detection in large-scale seedling production, providing technical support for replanting automation. Full article

Recent advances in satellite observations have expanded the use of Sea Surface Temperature (SST) and Sea Surface Height (SSH) data in climate and oceanography, yet their low spatial resolution limits fine-scale analyses. We propose HiT_DS, a modular hierarchical Transformer framework for high-resolution downscaling of SST and SSH fields. To address challenges in multiscale feature representation and physical consistency, HiT_DS integrates three key modules: (1) Enhanced Dual Feature Extraction (E-DFE), which employs depth-wise separable convolutions to improve local feature modeling efficiently; (2) Gradient-Aware Attention (GA), which emphasizes dynamically important high-gradient structures such as oceanic fronts; and (3) Physics-Informed Loss Functions, which promote physical realism and dynamical consistency in the reconstructed fields. Experiments across two dynamically distinct oceanic regions demonstrate that HiT_DS achieves improved reconstruction accuracy and enhanced physical fidelity, with selective module combinations tailored to regional dynamical conditions. This framework provides an effective and extensible approach for oceanographic data downscaling. Full article

Infrared images are often degraded by complex noise due to hardware and environmental factors, posing challenges for subsequent processing and target detection. To overcome the shortcomings of existing denoising methods in balancing noise removal and detail preservation, this paper proposes a Wavelet Transform Enhanced Infrared Denoising Model (WTEIDM). Firstly, a Wavelet Transform Self-Attention (WTSA) is designed, which combines the frequency-domain decomposition ability of the discrete wavelet transform (DWT) with the dynamic weighting mechanism of self-attention to achieve effective separation of noise and detail. Secondly, a Multi-Scale Gated Linear Unit (MSGLU) is devised to improve the ability to capture detail information and dynamically control features through dual-branch multi-scale depth-wise convolution and gating strategy. Finally, a Parallel Hybrid Attention Module (PHAM) is proposed to enhance cross-dimensional feature fusion effect through the parallel cross-interaction of spatial and channel attention. Extensive experiments are conducted on five infrared datasets under different noise levels (σ = 15, 25, and 50). The results demonstrate that the proposed WTEIDM outperforms several state-of-the-art denoising algorithms on both PSNR and SSIM metrics, confirming its superior generalization capability and robustness. Full article