Drone technologies offer significant promise for enhancing medical logistics, particularly in rural and remote areas. However, despite several pilot trials across the UK, there remains a lack of evidence regarding patient perceptions of drone integration in routine healthcare delivery. This study aimed to explore patient perspectives on the use of drones in medical logistics, with a focus on cost, safety, and impact on quality of life amongst three medical areas (chemotherapy, lab diagnostics and blood products). A qualitative study using semi-structured interviews and small focus groups was conducted with 26 participants from five Scottish health boards. Participants were recruited via NHS engagement leads, community networks, and snowball sampling. Data were collected via Microsoft Teams and analysed thematically using an inductive approach. Eight key themes emerged: Acceptance, Knowledge of drones, Potential use within the NHS, Potential impact, Concerns, Public perception, Support networks, and Future innovation. Participants expressed cautious optimism, highlighting potential benefits including improved access in rural areas, reduced travel burdens, and environmental gains. Concerns included technical reliability, regulatory complexity, data privacy, and the risk of increasing health inequities. Drones were viewed by patients as a promising complement to existing medical logistics, especially for emergency and rural applications. However, successful integration demands transparent communication, co-design with stakeholders, robust regulatory planning, and sustainable funding models to ensure equitable and safe implementation.
Mosquito larval source management remains a critical strategy for malaria control, but traditional methods for mapping breeding sites are labor-intensive and slow. While drones offer a promising tool, their application is often hindered by reliance on costly orthomosaic generation (stitching drone images into precise, georeferenced maps) and multispectral sensors, creating significant barriers for resource-limited health programs. This study developed an alternative workflow using a deep learning approach for water body segmentation. We utilized the DeepLabV3 + architecture with an EfficientNetV2 backbone on still RGB and Grayscale drone imagery, bypassing the need for orthomosaics by directly georeferencing individual images using GPS metadata. The model was trained and evaluated on a dataset of over 4,400 images from Pangandaran Regency, Indonesia, with performance assessed via the Intersection over Union (IoU) metric. Predictions were validated through field inspections to confirm water presence and larval abundance. The model achieved a mean IoU of 0.86 on RGB images and 0.80 on Grayscale data, demonstrating robust performance even with minimal spectral information. Field validation of 47 predicted sites confirmed water presence in all cases (100% confirmation of water presence), with 31.9% found to contain mosquito larvae, including the primary local vectors Anopheles vagus and An. sundaicus. The entire workflow, from cloud-based processing to field localization, was executed without specialized hardware or software. Our findings demonstrate that a deep learning-based analysis of still drone imagery provides a scalable, cost-effective, and rapid alternative to multispectral and orthomosaic-dependent methods for larval habitat mapping. This approach democratizes advanced surveillance technology, significantly shortens the delay between data collection and intervention, and holds great promise for enhancing the efficiency of vector control programs in malaria-endemic regions.
The objective was to determine the test-retest reliability and concurrent validity of a drone system in comparison to a radar device. Seventeen male collegiate soccer players participated in two maximal 30-meter sprint runs. The test-retest reliability of the drone system was evaluated using intraclass correlation coefficients (ICC3,1), coefficient of variation (CV%), and standard error of measurement (SEM). Subsequently, the systematic bias and consistency of the two devices on various force-velocity (F-V) variables (e.g., maximal velocity [Vmax], theoretical maximal velocity [V0], theoretical maximal horizontal force [F0], the slope of the F-V relationship [SFV]) were evaluated using linear mixed model (LMM) and Bland-Altman analysis. The drone system demonstrated moderate to excellent test-retest reliability across all variables (0.59 ≤ ICC ≤ 0.95; CV% < 10%). While LMM analysis detected significant systematic differences for Vmax (p = 0.013) and V0 (p = 0.012), Bland-Altman analysis confirmed high practical agreement with minimal bias (≤ 1.12%) and narrow limits of agreement (LoA < 10%). Pmax, split times (T5m-T20m) and average accelerations (A10m-A20m) demonstrated greater consistency (%Bias ≤ 0.76%) with no significant systematic bias (p > 0.05). Conversely, early-acceleration and model-derived metrics (Tau, Amax, F0, SFV) exhibited significant bias (p ≤ 0.028) and wide LoA exceeding 10% (e.g., F0: -13.37% to 8.56%; SFV: -11.54% to 18.18%). In conclusion, although the drone system exhibits high monitoring value in the maximum speed phase, early-acceleration metrics (Amax, F0, and T5m) should be interpreted with caution for individual-level monitoring. The tracking instability during the early acceleration phase necessitates further algorithm optimization.
Background: Modern warfare has introduced novel mechanisms of injury, particularly drone-induced blast trauma, resulting in complex craniomaxillofacial injuries. These injuries differ substantially from typical ballistic wounds and require adapted surgical strategies. This study was conducted to evaluate the clinical characteristics, management approaches, and long-term outcomes of midfacial blast injuries. Methods: A retrospective analytical study was conducted on 41 patients with drone-induced midfacial blast injuries treated at a tertiary referral center in Armenia following the 2020 Nagorno-Karabakh War. All patients underwent surgical management after initial stabilization and were followed for 5 years. Clinical outcomes, complications, and reconstructive needs were assessed. Results: All patients presented with comminuted midfacial fractures, which were frequently associated with polytrauma (87.8%). Burns were observed in 82.9% of cases. Surgical management included radical debridement and early definitive osteosynthesis using titanium fixation systems. No cases of postoperative osteomyelitis, bone sequestration, or implant failure were observed during the 5-year follow-up period. Patients with extensive soft tissue defects, particularly nasal and lip amputations, required multiple reconstructive procedures. Long-term follow-up revealed progressive soft tissue thinning over titanium meshes, especially in the zygomatico-orbital region, necessitating secondary interventions such as lipofilling. Conclusions: Drone-induced midfacial blast injuries represent a distinct and severe form of trauma. Early definitive reconstruction following adequate debridement was associated with favorable outcomes. However, soft tissue reconstruction remains challenging and often requires staged procedures. Long-term follow-up is essential to manage delayed complications and optimize aesthetic outcomes.
Clove oil, primarily extracted from the flower buds of the clove plant, is widely used in food, cosmetics, and traditional medicine due to its rich composition of phenolic acids and flavonoids, with eugenol being a major component. This essential oil exhibits numerous biological activities, including antibacterial, antifungal, and anti-inflammatory properties. In the context of drone semen cryopreservation, challenges such as ice crystal formation, and oxidative stress hinder fertility outcomes. To address these issues, the study explores the effects of commercial clove oil preparation on the post-cryopreservation quality of drone semen, its antioxidant and antibacterial qualities. The collected semen samples were pooled and divided among semen extenders containing different concentrations of clove oil (C-10, C-20, C-25) and a control extender containing no clove oil. Motility, plasma membrane integrity, acrosomal integrity, and mitochondrial membrane potential were evaluated using microscopic and flow-cytometric methods. Malondialdehyde and glutathione concentrations and catalase activity were also assessed. In addition, the minimum inhibitory concentration and antimicrobial effect of clove oil were evaluated. At the post-thaw stage, the highest motility was observed in the C-25 group; the C-20 group also showed a significant advantage compared with the control. The C-20 and C-25 groups provided a similar level of membrane integrity protection but better than the control. Acrosomal integrity protection was highest in C-25, and C-20 was also significantly higher than the Control group. The highest mitochondrial membrane potential values were observed in C-20 and C-25. Malondialdehyde concentration was highest in the Control group and lowest in the C-25 group; C-20 and C-25 remained similar to each other. Higher glutathione concentrations and catalase activity were observed in the C-25 group compared with the Control. Catalase activity in the C-20 group did not differ significantly from that in the C-10 or Control groups. The microbiological examination revealed no growth in any of the groups. The findings indicate that supplementation with a commercial clove oil preparation at the tested doses may improve post-thaw drone semen quality.
Supervisory control of autonomous drones in cluttered urban environments requires operators to update beliefs about dynamic hazards, such as localized wind changes, from imperfect and time-varying cues. To examine how interface design shapes this belief-updating process, four feedback conditions were compared: a baseline information panel (Control), augmented visual cues (Visual), upper-body haptic cues (Haptic), and their combination (Multimodal). Thirty participants supervised an autonomous drone in a Unity-based high-fidelity Virtual Reality simulation, with integrated eye tracking sampled at 90 Hz to derive an objective cognitive-load index. Belief updating was captured through change reporting and quantified using reaction latency and a Bayesian Beta-Binomial "local probability" metric that estimates time-resolved correctness, alongside subjective workload measured via NASA-TLX. Reaction latency decreased monotonically from Control (3.23 s) to Visual (2.61 s) to Haptic (1.67 s) to Multimodal (1.08 s). Multimodal was faster than Control (p < 0.001) and faster than Haptic (p = 0.045). Time-resolved correctness similarly improved, with mean local probability rising from 0.17 (Control) to 0.24 (Visual) and 0.22 (Haptic), reaching 0.43 under Multimodal. Eye-tracking comparisons indicated higher cognitive load in Haptic relative to Visual (p = 0.0201) and Multimodal (p = 0.0026). Together, the findings indicate a multimodal synergy that improves both speed and reliability of belief updates without the cognitive-load elevation observed under haptic-only feedback, supporting multimodal interface design for safer and more dependable human-autonomy teaming in urban drone operations.
We report the first in situ observation of mating behaviour in the endangered Groovebelly Stingray (Dasyatis hypostigma), endemic to the Southwestern Atlantic. Recorded by drone in May 2025 off Ilhabela, Brazil, the event involved one female and four males, lasting ~3 min and comprising close following and pre-copulatory biting, without agonistic interactions. The interaction comprised a partial behavioural sequence of close following and pre-copulatory biting, without agonistic interactions between males. This record provides novel insights into the species' reproductive strategy and highlights the value of non-invasive drone surveys for elasmobranch research.
Polycyclic aromatic hydrocarbons (PAHs) in urban estuaries exhibit sharp concentration shifts during rainfall events, yet their transient redistribution and compositional restructuring remain poorly resolved due to the mismatch between laboratory specificity and field-scale monitoring frequency. Conventional chromatography provides chemical resolution but lacks temporal coverage, whereas autonomous underwater drones deliver high-frequency measurements without molecular specificity. Here we bridge this monitoring gap by transferring laboratory-derived spectral information into sensor-based field models using knowledge distillation, enabling process-resolving PAHs assessment at scale. To overcome limited sample availability under rainfall conditions, a variational autoencoder expanded 142 observations thirtyfold, stabilizing model transfer. The integrated framework achieved an R² of 0.92 for ΣPAHs, improving predictive performance by 28%. Large-scale deployment across 59,392 drone measurements revealed rainfall-triggered surges dominated by high-molecular-weight PAHs and dynamic hotspot migration within the estuary. Interpretable analyses further indicate how spectral signatures reorganize along specific sensor pathways under hydrological perturbation. By coupling laboratory specificity with autonomous sensing, this approach establishes a scalable strategy for resolving pollutant dynamics in rainfall-impacted urban waters.
Chemical, Biological, Radiological, and Nuclear (CBRN) emergencies represent one of the most complex contexts for defense systems and public health. The use of drones now provides concrete tools to reduce direct personnel exposure, enhance environmental data collection, and optimize healthcare logistics. This paper analyzes the technological evolution of drones in the healthcare sector, highlighting the main types employed and their operational prospects in CBRN emergencies. A central section is devoted to the training of CBRN UAS (Unmanned Aircraft System) operators, considered a strategic prerequisite for the effective integration of these technologies into safety and response protocols. Through a hypothetical/propositional approach, training models based on integrated modules, realistic simulations, and joint-force exercises are outlined, identifying the regulatory and organizational challenges that remain open. The conclusions emphasize the need for a unified training system capable of leveraging the expertise already present within Italian military structures and promoting inter-agency cooperation for a safer and more efficient response to CBRN emergencies.
In cooperative counter-drone defense, dynamic interception task allocation presents a fundamental challenge: targets may appear, vanish, or shift trajectory unpredictably throughout an engagement. Current multi-agent reinforcement learning approaches generally lack explicit mechanisms to capture the changing correspondence between interceptors and targets as the operational topology evolves. We propose DT-GAT-MARL, a hierarchical framework pairing a Dynamic-Topology Graph Attention Network (DT-GAT) at the strategic allocation level with Multi-Agent Proximal Policy Optimization (MAPPO)-based maneuver control at the tactical level. Three design choices define DT-GAT: a masking mechanism that handles node additions and removals without rebuilding the graph, learnable edge-feature biases that embed spatial reachability and interception urgency directly into attention computation, and a Gumbel-Softmax allocation head that enables differentiable discrete assignment while preserving end-to-end gradient flow. A dual-frequency architecture separates the allocation and maneuvering timescales, which helps suppress assignment oscillations. We evaluate the framework across balanced, numerically disadvantaged, and dynamic intrusion scenarios ranging from 4v4 to 12v12. In the dynamic intrusion setting, DT-GAT-MARL exceeds the strongest baseline by 10.3 percentage points, reaches an 87.3% effective reallocation rate, and keeps oscillation at just 9.6%. Ablation results confirm that edge-feature bias is the single most critical component.
A core bottleneck of forestry remote sensing lies in accurate, real-time pine wilt disease monitoring on UAV-borne edge hardware, which suffers from constrained computing capacity and complicated field forest backgrounds. To fill this technical gap, we developed an ultra-lightweight real-time detection architecture named Edge-Forest YOLO in this work. Methods: Built upon the baseline YOLOv8n network, three targeted optimizations were embedded into the proposed model: (1) a domain-adaptive data augmentation workflow to mitigate poor generalization induced by variable illumination and uneven lesion sizes in complex woodland; (2) scale-aware asymmetric channel redistribution, cutting 37.5% shallow channels and expanding 75% deep channels to remove redundant spatial features and strengthen high-level pathological feature extraction; (3) Cross-layer ECA attention adopting 1D convolution to capture inter-channel correlation and concentrate on diseased regions with minimal computation overhead. All model validation was performed on the public high-resolution UAV pine wilt PDT dataset. Edge-Forest YOLO only occupies 2.31 M storage with mAP@0.5 up to 92.7%. Its single-image inference costs 4.2 ms on regular computing equipment and runs at around 26 FPS on the Jetson Nano edge platform. Compared with YOLOv8s and customized YOLO-DP, our model cuts over half parameter quantity while retaining competitive detection precision. The proposed lightweight detector supplies a low-power, practically deployable solution for on-board UAV real-time forest disease monitoring, supporting rapid in-field pine wilt diagnosis and facilitating scientific decision-making for forest health management and disease prevention.
暂无摘要（点击查看详情）
To address the issues of low detection accuracy and insufficient model lightweighting in UAV-based transmission line foreign object detection, this paper proposes SDS-YOLOv8n, an optimized foreign object detection algorithm. Built upon the YOLOv8n architecture, the proposed model incorporates three targeted improvements to balance detection accuracy and computational efficiency. First, an Enhanced SPPF module integrating parallel global average and max pooling layers is designed to improve the model's focus on target edge details while suppressing environmental background interference. Second, the standard detection head is replaced with a Dynamic Detection Head that unifies scale-aware, spatial-aware, and task-aware attention mechanisms, significantly enhancing feature adaptability for irregular objects such as kites and bird nests. Third, a Parameter-free Attention Mechanism (SimAM) is embedded to further refine feature extraction without increasing model parameters. Experimental results on a self-constructed dataset demonstrate that SDS-YOLOv8n achieves a mAP@0.5 of 95.8% and a mAP@0.5:0.95 of 75.1%, outperforming the baseline model by 1.1% and 2.6%, respectively. Furthermore, the model exhibits strong generalization capabilities on an additional untrained dataset. With a parameter count of 2.75 M and high inference speed verified on the NVIDIA Jetson Orin Nano platform, the proposed method provides a robust and efficient solution for real-time intelligent grid inspection.
Apis cerana (A. cerana) is a native and widely managed honey bee species in China. Body size and body weight are crucial breeding traits, as colonies possessing individuals with large body weight tend to be healthier and exhibit high productivity. This study aimed to clarify the relationships between body size and body weight in A. cerana and to evaluate their associations with geographic, climatic, and colony productive traits for selective breeding. Body size and body weight were measured in virgin queens, drones, and workers from Jinfo Mountain, Chongqing, and additional measurements of queens and drones were implemented in five other regions across China. Linear mixed-effects models confirmed that body size had a significant positive effect on body weight in virgin queens, drones, and workers. However, correlations of body-size and body-weight traits among different bee groups were weak and non-significant after FDR correction, indicating that drones or workers cannot be used as direct substitutes for queen body-size traits in the present dataset. Standardized model estimates showed that queen and drone body-size and body-weight traits were consistently negatively associated with annual minimum and annual mean temperatures, but positively associated with latitude after FDR adjustment. Annual precipitation was also negatively associated with queens' body size, queens' body weight, and drones' body size, whereas annual maximum temperature, longitude, and elevation showed no significant associations after FDR adjustment. Moreover, queens' body size and body weight were significantly positively associated with honey yield, honey yield during the main nectar flow, and colony gentleness after FDR correction, whereas their associations with the number of effective eggs laid by queens, colony strength, and robbery were not significant after FDR correction. These findings suggest that queen body-type traits may serve as useful auxiliary indicators for selecting colonies with higher honey production and gentler behavior, but their relationships with other colony traits should be interpreted cautiously. This research is beneficial for initiating a body size-weight selective breeding program for A. cerana, as it can help optimize breeding objectives and accelerate genetic progress.
Detecting objects in drone-captured aerial imagery is particularly formidable due to challenges such as the prevalence of numerous small targets and their dense spatial distribution. To bridge this gap, this paper introduces YOLO-UTD (YOLO-UAV Traffic Detection), a dedicated small object detector tailored for drone traffic surveillance. Built upon the YOLOv8 framework, the proposed model incorporates three principal enhancements. First, a specialized small-object detection head replaces the original large-object head to increase the sensitivity to fine-grained features. Second, we introduce a shallow-augmented feature pyramid network (SFPN) into the neck module. The SFPN enriches the semantic content of high-resolution shallow features via dense multiscale interactions and CARAFE upsampling, boosting performance on small targets. Finally, a C2fA layer is integrated into the deep backbone stages to adaptively fuse spatial details and semantic context through a dual-path architecture and a cross-attention mechanism, thereby dynamically refining features critical for small objects. Extensive experiments on the VisDrone2019 dataset validate that YOLO-UTD achieves a 3.6% higher mean average precision (mAP) than YOLOv8 while preserving a low parameter footprint, with a particularly significant gain of 5.3% in vehicle detection accuracy. These findings confirm the model's efficacy and strong potential for application in smart city drone surveillance.
The relationship between attentional resource allocation and early motor skill learning remains unclear. This study examined how proficiency in drone operation influences attention and which factors contribute to individual differences. Twenty-three participants with no prior drone experience performed a simulator task consisting of 12 three-minute trials, divided into three phases of four trials each. We estimated attentional resource allocation using auditory-evoked potentials (AEPs) elicited by task-irrelevant auditory probes. Performance improved significantly across phases (ps < 0.001), and N1 and P2 amplitudes were significantly reduced (ps < 0.05), indicating that participants allocated more attentional resources to drone operation as proficiency increased. Furthermore, overall task performance significantly correlated with smaller P2 amplitudes (r = -0.42, p = .044). Given that variability in P2 amplitude of AEPs elicited by task-irrelevant probes reflects endogenous attention, the findings suggest that endogenous attention might play a key role in early motor skill learning and underlie individual differences.
Anterior cruciate ligament (ACL) injuries remain a leading cause of morbidity in athletic populations, with 70-80% occurring through non-contact mechanisms driven by biomechanical risk factors including knee valgus (>10°), low knee flexion (<30°), tibial internal rotation (>20°), and loading asymmetry (>15°), yet implementation of evidence-based neuromuscular training (which reduces injury risk by 50-70%) remains limited due to barriers in identifying at-risk individuals through accessible field-based screening. This narrative review synthesizes motion analysis technologies spanning laboratory-based optical systems (marker-based), wearable inertial measurement units (IMUs), computer vision and marker-less pose estimation, force plate and pressure-sensitive insole systems, and integrated drone-based field assessment platforms to address this critical gap. We present a three-tier clinical screening framework that progresses from basic anthropometric and single-plane video analysis to multi-modal biomechanical assessment using real-time kinematic feedback. As an illustrative example of emerging field-deployable technology, an integrated drone-based motion capture and smart insole system combining 4K video capture, AI-driven 3D motion reconstruction, and plantar pressure mapping is described to demonstrate how laboratory-quality biomechanical assessment can be achieved in ecologically valid field settings. This evidence-based review addresses current gaps between laboratory research and practical field deployment, with emphasis on cost-effectiveness, accessibility, and clinical utility for ACL injury prevention in diverse sporting environments.
Russia's 2022 full-scale invasion of Ukraine has precipitated the largest and most destructive conflict in Europe since World War II, yet traditional notions of linear battlefields and clearly delineated combatants no longer apply. Russia's widespread use of drones, precision-guided munitions, digital disinformation, ecocide, and systematic attacks on civilian infrastructure has dispersed the risk of death and injuries across the entire Ukrainian population. The close proximity of major population centers to active combat, the large proportion of Ukrainians mobilized as military personnel or first responders, and the persistent threat of drone and missile strikes further blur distinctions between combatants and noncombatants. The U.S. military is incorporating strategic and tactical lessons from this conflict into its doctrine for large-scale combat operations (LSCO) with near peer adversaries, including revised approaches to battlefield healthcare and combat stress control. This case study examines current wartime conditions in Ukraine from a psychological perspective, identifies shared and unique war stressors among Ukrainian military personnel and civilians, and discusses Ukraine's actions to mitigate these stressors. The analysis synthesizes peer-reviewed research, governmental and non-governmental reports, technical documents, news media reports, and field observations. Current U.S. doctrine for managing combat stress has a relatively narrow focus, emphasizing actions at the military unit level. In contrast, the Ukrainian experience illustrates the need for a whole-of-society approach for LSCO level warfare. Insights from Ukraine offer critical understanding of the psychological demands of LSCO and can inform future U.S. doctrine, training, and policies for high-intensity conflict.
Access to safe blood and blood products is a global health priority, requiring concerted public health efforts to ensure universal access to reduce the global disease burden. However, there are inequities in access to safe and adequate blood and blood products in sub-Saharan Africa (SSA), leading to marked challenges in meeting the clinical needs of patients with diverse pathologies. This study used the WHO health system building blocks to systematically synthesize the literature to understand the barriers to blood and blood product usage in SSA. We used the PRISMA guideline to systematically search relevant articles using six electronic databases: Web of Science, MEDLINE, PubMed, PscyInfo, Google Scholar, and Global Health databases between 2005 and 2023. The risk of bias for included studies was assessed using a modified Joanna Briggs Checklist. Inductive thematic analysis was performed to thematize the extracted data based on the WHO health system building blocks. Sixty-five studies representing eighteen countries in SSA were included for review. Barriers included transfusion delays, low transfusion rates, inappropriate practices, untimely referrals, blood stock-outs, poor adherence to guidelines, lack of standardized protocols, poor transfusion management systems, poor leadership/governance, and health workforce shortages. Facilitators included drone technology use, prompt referrals, appropriate transfusions, availability of skilled workforce, health information systems, and effective leadership and governance. Numerous barriers to blood and blood product access in SSA exist. Addressing service delivery barriers such as transfusion delays, product stock-outs, and ineffective governance of transfusion systems would be vital. Further, innovative technology, boosting referral systems, and practicing appropriate use of blood would engender sustainable access to safe blood services in SSA. Innovations such as the successful deployment of drone technology in Rwanda are worth emulating and scaling across the continent. Also, financing and effective leadership should be further researched as potential facilitators of safe blood accessibility and use in Africa. https://www.crd.york.ac.uk/PROSPERO/view/CRD42023434335.
This study presents an approach for the object detection of multiple weeds in tomato production systems based on deep learning. A comprehensive dataset has been collected in three provinces of Türkiye (Balıkesir, Ankara, and Aksaray) under real-world field conditions. The data set has 32,607 images and 44,165 bounding boxes annotations. The two weed species included in the dataset are, to our knowledge, underrepresented in the current deep learning-based agricultural object detection literature. Drone and smartphone cameras took pictures at different times of the day (morning, noon, and afternoon) of different soil textures, light levels, and weather conditions, such as rain, mud, and shadows. The dataset reflects agricultural diversity as it exists in the real world, unlike previous studies that relied on controlled experimental environments. The model was trained using YOLO-based deep learning algorithms within the PyTorch framework. The metrics Precision, Recall, mAP@0.5, and mAP@[0.5:0.95] were used to evaluate the performance of the models. In this study, seven different YOLO architectures were comparatively evaluated on the TomatoWeedDet dataset created under real field conditions. The results show that the YOLOv8l model demonstrates high performance in the multi-class weed detection task and has significant potential for precision weed management applications. The model that was created could be used in mobile or embedded systems to monitor weeds in real time with drones. The proposed system enables targeted herbicide application and less use of chemicals. This study advances research on weed detection using deep learning. It also helps to make precision and sustainable farming systems a reality.