EDNet: Edge-Optimized Small Target Detection in UAV Imagery - Faster Context Attention, Better Feature Fusion, and Hardware Acceleration DOI
Zijian Song, Yuan Zhang, Abd Al Rahman M. Abu Ebayyeh

et al.

Published: Dec. 2, 2024

Detecting small targets in drone imagery is challenging due to low resolution, complex backgrounds, and dynamic scenes.We propose EDNet, a novel edge-target detection framework built on an enhanced YOLOv10 architecture, optimized for real-time applications without postprocessing.EDNet incorporates XSmall head Cross Concat strategy improve feature fusion multiscale context awareness detecting tiny diverse environments.Our unique C2f-FCA block employs Faster Context Attention enhance extraction while reducing computational complexity.The WIoU loss function employed improved bounding box regression.With seven model sizes ranging from Tiny XL, EDNet accommodates various deployment environments, enabling local inference ensuring data privacy.Notably, achieves up 5.6% gain mAP@50 with significantly fewer parameters.On iPhone 12, variants operate at speeds 16 55 FPS, providing scalable efficient solution edge-based object imagery.The source code pre-trained models are available at: https://github.com/zsniko/EDNet.

Language: Английский

Surveying the deep: A review of computer vision in the benthos DOI Creative Commons
Cameron Trotter, Huw J. Griffiths, Rowan J. Whittle

et al.

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 102989 - 102989

Published: Jan. 1, 2025

Language: Английский

Citations

1
Drone-assisted adaptive object detection and privacy-preserving surveillance in smart cities using whale-optimized deep reinforcement learning techniques DOI Creative Commons
Ahmed Abu‐Khadrah, Ahmad Al–Qerem,

Mohammad R. Hassan

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 22, 2025

Drone/ unmanned aerial vehicles (UAV) surveillance for object/ human detection is familiar in large gatherings the modern cities era. Artificial intelligence algorithms and computer-aided processing will handle images extracted from videos to reveal object. This article proposes a novel object technique (ODT) that assimilates whale optimization deep reinforcement learning. The algorithm detects spreading image features origin end of x×y pixels. feature extraction performed until complete pixels are covered identify their existence least position. forging behaviour whales implied highly overlapping detection/ classification. If increase, whale's movement updated last-known highest pixel learning recommends new agents validate low features, such agent moves towards feature. Therefore, feature-based differentiation pursued using searching objects through collated

Language: Английский

Citations

1
Bearing-DETR: A Lightweight Deep Learning Model for Bearing Defect Detection Based on RT-DETR DOI Creative Commons

Minggao Liu,

Haifeng Wang, Luyao Du

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(13), P. 4262 - 4262

Published: June 30, 2024

Detecting bearing defects accurately and efficiently is critical for industrial safety efficiency. This paper introduces Bearing-DETR, a deep learning model optimised using the Real-Time Detection Transformer (RT-DETR) architecture. Enhanced with Dysample Dynamic Upsampling, Efficient Model Optimization (EMO) Meta-Mobile Blocks (MMB), Deformable Large Kernel Attention (D-LKA), Bearing-DETR offers significant improvements in defect detection while maintaining lightweight framework suitable low-resource devices. Validated on dataset from chemical plant, outperformed standard RT-DETR, achieving mean average precision (mAP) of 94.3% at IoU = 0.5 57.5% 0.5–0.95. It also reduced floating-point operations (FLOPs) to 8.2 G parameters 3.2 M, underscoring its enhanced efficiency computational demands. These results demonstrate potential transform maintenance strategies quality control across manufacturing environments, emphasising adaptability impact sustainability operational costs.

Language: Английский

Citations

4
Ldstd: low-altitude drone aerial small target detector DOI

Yuheng Sun,

Zhenping Lan,

Yanguo Sun

et al.

The Journal of Supercomputing, Journal Year: 2025, Volume and Issue: 81(2)

Published: Jan. 21, 2025

Language: Английский

Citations

0
YOLO-SIFD: YOLO with Sliced Inference and Fractal Dimension Analysis for Improved Fire and Smoke Detection DOI Open Access

Mr. Muhammad Ishtiaq,

Jong-Un Won

Computers, materials & continua/Computers, materials & continua (Print), Journal Year: 2025, Volume and Issue: 82(3), P. 5343 - 5361

Published: Jan. 1, 2025

Language: Английский

Citations

0
MFYOLO: Improved UAV lightweighting algorithm for wind turbine blade surface visibility damage detection DOI

Jiale Xiao,

Lei Xu,

Changyun Li

et al.

Computers & Electrical Engineering, Journal Year: 2025, Volume and Issue: 123, P. 110225 - 110225

Published: March 14, 2025

Language: Английский

Citations

0
Flight Evolution: Decoding Autonomous UAV Navigation—Fundamentals, Taxonomy, and Challenges DOI Open Access

Geeta Sharma,

Sanjeev Jain

Transactions on Emerging Telecommunications Technologies, Journal Year: 2025, Volume and Issue: 36(4)

Published: March 19, 2025

ABSTRACT Due to the adaptability and effectiveness of autonomous unmanned aerial vehicles (UAVs) in completing challenging tasks, research on UAVs has increased quickly during past few years. An UAV refers drone navigation an unknown environment with minimal human interaction. However, when used a dynamic environment, confront numerous difficulties including scene mapping localization, object recognition avoidance, path planning, emergency landing, so forth. Real‐time demand quick responses situations; as result, this is crucial feature that requires further research. This article presents different novel taxonomies briefly explain communication architecture utilized ground stations. Popular databases for UAVs, fundamentals latest ongoing detection avoidance methods, planning techniques, trajectory mechanisms are also explained. Later, we cover benchmark dataset available kinds simulators UAVs. Furthermore, several challenges covered. From literature, it been found algorithms based deep reinforcement learning (DRL) employed more frequently than other intelligence field navigation. To best our knowledge, first covers aspects related

Language: Английский

Citations

0
Drone-Based Detection and Classification of Greater Caribbean Manatees in the Panama Canal Basin DOI Creative Commons
Javier E. Sánchez-Galán,

Kenji Contreras,

Allan Denoce

et al.

Drones, Journal Year: 2025, Volume and Issue: 9(4), P. 230 - 230

Published: March 21, 2025

This study introduces a novel, drone-based approach for the detection and classification of Greater Caribbean Manatees (Trichechus manatus manatus) in Panama Canal Basin by integrating advanced deep learning techniques. Leveraging high-performance YOLOv8 model augmented with Sliced Aided Hyper Inferencing (SAHI) improved small-object detection, our system accurately identifies individual manatees, mother–calf pairs, group formations across challenging aquatic environment. Additionally, use AltCLIP zero-shot enables robust demographic analysis without extensive labeled data, enhancing adaptability data-scarce scenarios. For this study, more than 57,000 UAV images were acquired from multiple drone flights covering diverse regions Gatun Lake its surroundings. In cross-validation experiments, achieved precision levels as high 93% mean average (mAP) values exceeding 90% under ideal conditions. However, testing on unseen data revealed lower recall, highlighting challenges detecting manatees variable altitudes adverse lighting Furthermore, integrated demonstrated top-2 accuracy close to 90%, effectively categorizing manatee groupings despite overlapping visual features. work presents framework technology, offering scalable, non-invasive solution real-time wildlife monitoring. By enabling precise classification, it lays foundation enhanced habitat assessments effective conservation planning similar tropical wetland ecosystems.

Language: Английский

Citations

0
SSOD-MViT: A novel model for recognizing alfalfa seed pod maturity based on semi-supervised learning DOI
Fuyang Tian, Y. Zhang, Shakeel Ahmed Soomro

et al.

Computers and Electronics in Agriculture, Journal Year: 2025, Volume and Issue: 236, P. 110439 - 110439

Published: April 23, 2025

Language: Английский

Citations

0
Precision enhancement in wireless capsule endoscopy: a novel transformer-based approach for real-time video object detection DOI Creative Commons
Tsedeke Temesgen Habe,

Keijo Haataja,

Pekka Toivanen

et al.

Frontiers in Artificial Intelligence, Journal Year: 2025, Volume and Issue: 8

Published: April 30, 2025

Wireless Capsule Endoscopy (WCE) enables non-invasive imaging of the gastrointestinal tract but generates vast video data, making real-time and accurate abnormality detection challenging. Traditional methods struggle with uncontrolled illumination, complex textures, high-speed processing demands. This study presents a novel approach using Real-Time Detection Transformer (RT-DETR), transformer-based object model, specifically optimized for WCE analysis. The model captures contextual information between frames handles variable image conditions. It was evaluated Kvasir-Capsule dataset, performance assessed across three RT-DETR variants: Small (S), Medium (M), X-Large (X). RT-DETR-X achieved highest precision. RT-DETR-M offered practical trade-off accuracy speed, while RT-DETR-S processed at 270 FPS, enabling performance. All models demonstrated improved computational efficiency compared to baseline methods. framework significantly enhances precision in WCE. Its clinical potential lies supporting faster more diagnosis. Future work will focus on further optimization deployment endoscopic analysis systems.

Language: Английский

Citations

0