Search Results (1,335)

Detecting small and distant objects in road traffic scenarios remains challenging owing to limited pixel resolution, cluttered backgrounds, and resource constraints on edge computing platforms. This work presents LCA-Net, a computationally efficient framework for small object detection that balances accuracy with model complexity. The framework incorporates three complementary designs: an Adaptive Deformable Downsampling Module (ADDM) that merges asymmetric and deformable convolution operations to improve spatial feature encoding while explicitly accounting for the parameter and computational cost of offset and modulation-mask prediction; a Cross-Scale Feature Fusion Pyramid (CSFFP) specifically engineered for minute objects, which augments multi-scale feature learning and enhances detection of far-field small targets; and a Lightweight Feature-Gated Detection Head (LFGDH) that employs channel–spatial attention to selectively emphasize informative features, thereby reducing both parameter count and computational cost. On Udacity, LCA-Net improves mAP@0.5 by 2.3 percentage points; on VisDrone2019, it improves mAP@0.5 by 1.7 percentage points. Across both benchmarks, the complete model reduces the parameter count by 25.58% and GFLOPs by 16.05% relative to YOLOv8-N. On the RTX A6000, LCA-Net-N reduces forward-pass latency from 1.82 to 1.63 ms, increases throughput from 549 to 613 FPS, and lowers peak GPU memory from 1180 to 1015 MiB. These results demonstrate a favorable accuracy–efficiency trade-off for real-time traffic perception. Full article

Timely forecasting of maize productivity is essential to support precision agriculture and optimize management practices. In this study, we analyzed the potential of integrating ground-based measurements and UAV-derived spectral data for predicting maize grain yield (GY) under different fertilization conditions. Field data were collected at two key phenological stages: early vegetative stage (V7) and pre-harvest (R4). Ground-based measurements included SPAD, above-ground biomass (AGB), and leaf area index (LAI), while multispectral and hyperspectral imagery was acquired by drone. A series of Ordinary Least Squares (OLS) models was developed to evaluate the predictive performance of individual variables and their combinations. Model robustness was assessed using two validation strategies: Leave-One-Treatment-Out (LOTO) to assess model performance across the treatments included in the experimental design and random sampling to assess performance within the dataset. The results showed that yield prediction was less accurate during the early growth stages, where data fusion significantly improved the model’s accuracy (${\mathrm{R}}^2$ = 0.82; MAE = 6.36 q ${\mathrm{ha}}^{-1}$; $MAPE\approx$7 %). The predictive performance of VIs alone increased substantially in the pre-harvest stage, with the combination of red-edge indices and LAI proving to be the best model for late yield prediction (${\mathrm{R}}^2$ = 0.86; MAE = 6.56 q ${\mathrm{ha}}^{-1}$; $MAPE\approx$7%). Comparison of multispectral and hyperspectral data revealed comparable predictive performance, suggesting that multispectral sensors may already capture the key spectral information needed for yield forecasting. Furthermore, random validation consistently produced more optimistic results than the LOTO method, highlighting the importance of using validation strategies that explicitly account for the experimental design when evaluating model performance across the treatments included in the study. Overall, the present study demonstrates that yield prediction is highly dependent on the phenological stage and validation approach, and that integrating complementary data sources can improve model performance, particularly during the early growth stages. These findings should be interpreted as a proof-of-concept based on a single-site, single-season experiment with a limited sample size (n = 12), and therefore require further validation across multiple environments and growing seasons. Full article

Drones are increasingly being used for tasks such as power grid maintenance, routine inspections, and forest monitoring. These tasks often involve large areas that need to be surveyed, requiring drones to be recharged frequently. However, these areas, such as mountainous regions and forests, have rugged terrain and humid environments, making it unsuitable to place charging systems on the ground. Furthermore, misalignment can easily occur when drones are docked. Therefore, we propose an underground wireless charging system for drones with a frustum magnetic coupling structure to adapt to rugged terrain and enhance the wireless charging capability of drones. On the transmitting side, a frustum magnetic coupling structure based on a tap coil is proposed. To enhance the drone’s alignment capability, the receiving coil is designed as a circular planar coil that matches the frustum structure of the transmitting coil, achieving self-alignment through its physical structural characteristics. The parameters of the magnetic coupling structure are designed and optimized through finite element simulation analysis, and the output characteristics of the equivalent T/S type compensation network are analyzed. A physical prototype is built, achieving a maximum efficiency of 90.4%. The feasibility of the proposed system has been verified. Full article

Multimodal object detection leveraging RGB and infrared imagery has become essential for robust all-weather perception in unmanned aerial vehicle (UAV) applications. However, existing methods still struggle with effective cross-modal feature fusion, spatial misalignment between modalities, and scale variation of objects in aerial views. In this paper, we propose CMGFDet, a Cross-Modal Gated Fusion Network with Multi-Receptive Field Aggregation designed for RGB–infrared aerial object detection. Our framework introduces three coordinated modules: (1) a Cross-Modal Feature Fusion Network (CMFFN) that employs a gated attention mechanism to selectively aggregate complementary information from both modalities during encoding; (2) a Global–Local Attention Module (GLAM) that performs hierarchical cross-modal feature alignment by jointly modelling global channel statistics and local spatial correlations in the decoder; and (3) a Multi-Receptive Field Aggregation Network (MRFAN) that captures multi-scale contextual information through parallel depthwise convolutions with diverse kernel sizes. Additionally, we incorporate a deep supervision strategy and a composite loss function to enhance training efficiency. Extensive experiments on four public benchmarks (DroneVehicle, RGBTDronePerson, VEDAI, and VTUAV) show that CMGFDet improves the previous best mAP@0.5 by 1.6%, 2.2%, 1.9%, and 2.2%, respectively. The implementation code will be released upon acceptance to support reproducibility. Full article

While drone delivery has gained significant scholarly and industrial interest, rural residents’ acceptance of these systems remains underexplored, despite the region’s acute logistics challenges and service inequities. Using survey data from rural U.S. residents, this study first estimates an ordered logistic regression (OLR) model to identify factors associated with five-level drone delivery acceptance. The study then compares OLR, multinomial logistic regression (MNL), Random Forest (RF), XGBoost, and LightGBM under matched feature sets to evaluate whether nonlinear machine-learning models improve prediction beyond the interpretable statistical baseline. Open-ended responses are coded into LLM-derived sentiment labels and added as supplementary predictors to test whether unstructured feedback improves acceptance prediction. Results show that willingness to pay is the strongest predictor of acceptance, while equitable same-day delivery demand and post-pandemic attitude adjustment are also positively associated with higher acceptance. Household disability status and urban accessibility are not significant after adjustment. In the five-level analysis, OLR provides a strong ordinal baseline, while XGBoost and other tree-based models improve selected class-level prediction metrics. In the binary high-acceptance analysis, machine-learning models show stronger predictive performance, especially when structured predictors are combined with sentiment features. This study contributes to rural drone-delivery literature by linking service equity, perceived value, and LLM-derived sentiment within a comparable statistical and machine-learning framework for rural service design. Full article

Accurate underwater acoustic localization is a key enabling capability for autonomous underwater vehicles and underwater drones, which cannot rely on satellite positioning while submerged and therefore depend on acoustic methods to determine their position. Localization based on time-of-arrival (TOA) measurements requires precise time synchronization between the source and the receivers, which is difficult to maintain in practical deployments. The time-difference-of-arrival (TDOA) representation removes this requirement but discards part of the absolute timing information, reducing localization accuracy. This study investigates a physics-based feedforward multilayer perceptron (FF-MLP) framework for two-dimensional range–depth underwater localization that learns directly from the arrival-time structure induced by sound-speed variability and multipath, with the receiver-array geometry treated as a central design variable for improving synchronization-free TDOA localization. Using multi-receiver arrival times generated with the BELLHOP beam-tracing model under a representative Mediterranean underwater environment, synchronous TOA, biased TOA, and TDOA measurement representations are compared on a common footing, and the effects of the receiver depth distribution, the number of receivers, and the reference-receiver position are systematically examined through Monte Carlo evaluation. The results show that the receiver-array geometry, rather than the measurement representation alone, is decisive for TDOA-based localization: with an appropriately designed geometry, synchronization-free TDOA localization achieves a median two-dimensional RMSE of 11.28 m, approaching the accuracy attainable with synchronous TOA, which requires precise time synchronization. These findings indicate that careful receiver-geometry design can make synchronization-free TDOA a practical alternative to synchronous TOA for the acoustic localization of autonomous underwater vehicles. Full article

Low-altitude unmanned aerial vehicle (UAV)-based traffic object detection is challenged by substantial scale variations from aerial perspectives, the extremely small pixel proportions of distant traffic participants, complex road background interference, unstable illumination, and severe occlusion in dense traffic scenes. To address these problems, this paper proposes ACP2-YOLO, an improved YOLO11-based detection framework for low-altitude UAV traffic scenarios, with the goal of enhancing the detection of vehicles, pedestrians, and non-motorized traffic participants. The proposed framework introduces two key improvements. First, a lightweight hybrid ACmix module that integrates convolution and self-attention is embedded into the network, enabling the model to jointly capture local detailed features and global contextual dependencies and thereby strengthen feature representation under complex backgrounds. Second, a P2 small-object detection layer is added to the original three-scale detection structure of YOLO11 to construct a four-scale P2–P5 feature pyramid. By allowing shallow high-resolution features to directly participate in object prediction, this design effectively reduces spatial information loss caused by deep downsampling and improves small-object perception. Experiments on the VisDrone2019 dataset show that the improved model achieves 53.1% Precision, 41.1% Recall, 42.9% mAP@50, and 26.3% mAP@50–95, outperforming the baseline YOLO11 by 4.2, 4.2, 5.0, and 3.6 percentage points, respectively. Comparisons with mainstream YOLO-series detectors further demonstrate its superior overall accuracy, small-object detection capability, and adaptability to complex scenes, indicating its potential for UAV-based traffic monitoring, road safety inspection, and intelligent transportation perception. Full article

The widespread application of drones in complex environments imposes higher demands on flight safety. However, traditional anomaly detection methods often rely on mathematical models or large amounts of labeled samples, which makes them ill suited to address practical challenges such as unmanned aerial vehicle systems, which exhibit strong nonlinear characteristics, a scarcity of fault samples, and highly variable operating conditions. To overcome the above difficulties, this work puts forward a one-dimensional selective kernel convolutional autoencoder (1-D SKCAE) based on Bayesian optimization for unsupervised drone anomaly detection. Relying solely on normal operation data, this model achieves accurate anomaly identification by leveraging the sudden changes in reconstruction error. For the model architecture, this paper designs a multi-scale selective kernel convolution module and combines an attention mechanism to achieve adaptive feature weighting for various receptive fields. This method effectively improves the model’s ability to represent complex operational conditions and subtle fault characteristics. Simultaneously, Bayesian optimization is embedded into the model training process as a hyperparameter search strategy, enabling the adaptive configuration of key hyperparameters to further enhance detection performance. Extensive experiments were conducted using the RflyMAD simulation dataset and the 3DR Solo real flight dataset. The results demonstrate that the 1-D SKCAE outperforms multiple comparative models, exhibiting superior robustness particularly in complex scenarios such as mixed multi-fault superposition. This method enables drone anomaly detection without fault labels, showcasing strong potential for engineering applications. Full article

High-rise glass curtain wall cleaning has long relied on dangerous manual high-altitude operations, which suffer from low efficiency, high labour costs and prominent safety hazards. Existing cleaning robots generally lack a systematic user demand-driven design framework, making it difficult to balance safety, obstacle-crossing capability and maintenance performance. To fill this gap, this study proposes a four-stage integrated sustainable design methodology integrating the KANO model, Analytic Hierarchy Process (AHP), Quality Function Deployment (QFD) and TRIZ. This methodology covers requirement attribute classification, priority weight quantification, demand-technology mapping and engineering conflict resolution, which can assist in sorting out multi-constraint design contradictions for the studied high-altitude cleaning equipment. Applying this methodology, the study develops the modular NetWing curtain wall cleaning drone with a vertically separable structure, retractable obstacle-crossing cleaning arm and quick-release cleaning head, and the feasibility of the conceptual scheme is preliminarily verified through numerical simulation. This work provides a quantifiable and replicable design paradigm for high-altitude service equipment, and the proposed solution can effectively reduce occupational risks and support the sustainable development of the facade cleaning industry. Full article

To address contactless delivery and fully account for customers’ expected delivery time preferences, this paper presents a mixed-integer programming model for minimizing truck–drone cooperative delivery costs. The model takes truck stopping points, truck routes, drone routes, and customer service time windows as decision variables and minimizes the sum of truck operating costs, drone operating costs, and time-window penalty costs as the objective. Based on the characteristics of the model, a two-stage algorithm is designed. In Stage 1, an improved K-Means clustering method partitions customers into sub-regions, with each cluster centroid serving simultaneously as a temporary truck stop and a drone launch point. In Stage 2, a variable neighborhood simulated annealing (SAVN) algorithm jointly optimizes truck and drone cooperative delivery routes to achieve minimum delivery cost. The correctness and effectiveness of the model and algorithm are verified by benchmarking against an exact solver and two traditional heuristic algorithms. A real-world case study in Chongqing, China, further shows that the two-stage algorithm achieves moderate cost savings and substantial solution-time reduction compared with simultaneous truck–drone route generation. Full article

Industrial-grade suspended-load logistics drones have now been widely used in commercial activities. The swing of their suspended loads has always been a major challenge in flight control. At present, most related studies take small quadrotor drones as the research object and employ approaches such as designing novel flight control systems, followed by analysis through theoretical and simulation-based validation. However, research on industrial-grade drones remains lacking, and the results of such studies cannot be quickly applied in engineering practice. Therefore, this paper proposes a set of flight operation guidelines for the existing flight control system of industrial-grade logistics drones. This study conducts flight experiments on commonly used industrial-grade logistics drones to investigate slung-load stability under varying built-in parameters, velocity profiles, and payload weights. The swing parameters are measured and analyzed. The results show that by adjusting relevant parameters, the swing angle and settling time are significantly improved. Finally, based on the experimental analysis results, a set of flight strategies is proposed, which can quickly improve the stability of the slung load of industrial-grade logistics drones using the existing conditions in engineering applications. Full article

Maritime transportation, a traditionally polluting sector, is engaged in sustainable innovation; advanced detection of marine pollution incidents can help control improvements in this sector. However, coastal and marine environmental governance is increasingly constrained by fragmented monitoring architectures in which environmental, operational, and regulatory datasets remain distributed across institutions and jurisdictions. Comprehensive governance mechanisms that cross the sea–land continuum are limited. Public authorities, port administrations, research institutes, and private operators independently monitor marine pollution, vessel movements, coastal pressures, and urban and ecological risks; however, these data streams remain siloed across organizational, sectoral, and jurisdictional boundaries. The absence of interoperability, real-time exchange, and coordinated analytics generates a gap between monitoring capacity and regulatory effectiveness, resulting in delayed detection of multi-risks, such as pollution incidents. Weak governance and assignment of responsibility and largely reactive enforcement practices further reinforce the problem. Anticipatory interventions are needed to improve coastal and marine sustainability. This paper examines the EMMERA (East Med Cross-border Marine Environmental Risk Assessment through E-Platform Integrated Data Management) platform established by three port authorities as a data-centric intervention designed to address some of these structural limitations. Implemented in port and coastal environments in Cyprus, Greece, and Israel, EMMERA integrates heterogeneous static and dynamic data sources—including satellite observations, administrative records, vessel information, and drone-based monitoring—into a unified operational framework accessible to competent authorities. Through data fusion, cross-validation, and automated anomaly detection combined with targeted drone verification, the platform aims to transform fragmented monitoring streams into coherent, actionable environmental intelligence, strengthening the evidentiary basis for regulatory intervention. The paper presents the platform design and provides a baseline assessment. It argues that EMMERA’s primary contribution lies not in the introduction of additional monitoring tools, but in enabling more effective coastal and marine environmental governance through integrated data infrastructures. EMMERA is proposed as a governance-oriented integrated data infrastructure whose anticipated contribution lies in improving institutional interoperability, risk visibility, and evidence generation for environmental oversight, even as operational effectiveness will require future evaluation following sustained deployment. More broadly, the paper proposes that integrated data architectures can recalibrate environmental governance, shifting emphasis from post hoc documentation toward anticipatory, coordinated, and performance-oriented regulatory practices. Full article

This paper presents a lightweight MCU–EdgeTPU platform—a microcontroller unit (MCU) paired with an Edge Tensor Processing Unit (EdgeTPU) accelerator—for onboard drone-perception experiments, extended from an open-source baseline originally limited to Quarter Video Graphics Array (QVGA) single-camera operation. Rather than treating hardware, runtime, model, and data as separate problems, they are developed as parts of the same continuous perception pipeline. The platform extends the hardware baseline toward dual 5 Mpx sensing, onboard inertial measurement unit (IMU) support, real-time embedded inference, and a high-level MicroPython control layer. In parallel, lightweight You Only Look Once (YOLO) detectors are trained and selected on a synthetic aerial-person dataset generated under the visual conditions expected by the drone camera, including target resolution, viewpoint, object scale, weather, lighting, and time-of-day variation. The resulting workflow starts from both ends: the detector must be small and quantization-stable enough for the EdgeTPU path, while the dataset must match the images that the onboard sensor is expected to observe. To evaluate the system, the full path from camera capture and image conversion to TPU transfer, model execution, and post-inference processing is analyzed. In the tested setup, the optimized single-camera pipeline runs stably with no timeouts or inference failures at about 26 detections per second with standard RGB input; because each EdgeTPU invocation is bounded by the USB transfer of the input image, feeding the camera’s native YUV420 format instead halves that transfer and raises throughput to about 40 detections per second at the same accuracy, while the selected 8-bit-integer (INT8) person detector preserves most of its 32-bit floating-point (FP32) accuracy. Detections are exposed to drone-control workflows (MAVLink/PX4 and Crazyflie) through the scriptable layer as an integration interface rather than a validated autonomy stack. The central contribution is therefore a co-designed embedded perception pipeline in which the board, runtime, detector, dataset, and even the camera pixel format are aligned around the same operating conditions. Full article

Small objects in UAV and remote sensing images often occupy only a few to several dozen pixels, and their edge and texture information is easily further weakened after processing by the network backbone. Although high-frequency information helps capture details of small objects, the background texture will also generate high-frequency responses. To address this issue, this paper proposes HFG-YOLO, a high-frequency-guided network. First, we propose the High-Frequency Feature Guidance Module (HFFGM), which extracts high-frequency information from images to compensate for the loss of small-object information caused by backbone downsampling. Second, we design the Background Suppression Fusion Module (BSFM) to reduce the impact of irrelevant high-frequency information on the model’s detection results. Experimental results on the VisDrone2019 and VEDAI datasets demonstrate the effectiveness of the proposed design. On VisDrone2019, HFG-YOLO achieves a 31.10% mAP50 and 6.08% mAP_s, outperforming the YOLO11n baseline by 2.70 and 1.71 percentage points, respectively. On VEDAI, HFG-YOLO achieves a 33.40% mAP50:95 and 22.93% mAP_s, improving over YOLO11n by 2.40 and 3.26 percentage points, respectively. Full article

Sandwich structures have gained much interest in drone manufacturing structures based on their lightweight design and excellent mechanical characteristics. In this work, a new solution for lightweight drone wing structures consisting of a thin sandwich skin, a main spar, and ribs was proposed. Seven sandwich structures based on prepreg-based CFRP skins and different cores were proposed for the wing drone sandwich skin. Thus, sandwiches with different chemical configurations and densities, such as ROHACELL 51, AIREX T92.100, balsa, AIREX R82.150, AIREX C71.75, NOMEX ECA-I, and Soric XF, were autoclave-manufactured and investigated. All the samples were tested under three-point bending. Also, microscopic analysis of the fracture zones was performed to establish a direct link between macroscopic flexural behavior and local failure mechanisms. A statistical analysis based on ANOVA with Box–Cox transformation followed by Tukey’s Honestly Significant Difference test was performed for flexural strength and flexural modulus. The results show that the sandwiches containing Soric XF foam with 62.5 kg/m³ density had the best mechanical properties, with a 71.66 MPa flexural strength and a 10,039 MPa flexural modulus. Full article