Abstract:As Unmanned Aerial Vehicle (UAV) remote sensing technology progresses, the utilization of deep learning in UAV imagery object detection has become more prevalent. However, detecting small targets in complex backgrounds and distinguishing dense targets remains a major challenge. To address these issues and improve object detection efficiency, this study proposes an UAV imagery object detection method called YOLO-UAV by optimizing YOLOv5. YOLO-UAV first reconstructs the backbone and feature fusion networks by simplifying the network structure and reducing computational burden. The employment of a Dense_CSPDarknet53 backbone network, fashioned via the incorporation of dense connections, facilitates the extraction of latent image information through the recurrent utilization of features. In the Neck structure, an efficient feature fusion block with structural re-parameterization and ELAN strategies is integrated to effectively reduce interference from complex background noise while extracting more accurate and rich features. In addition, by proposing GS-Decoupled Head, this approach diminishes the parameter count of the decoupled head without compromising accuracy. It also separates classification tasks from regression tasks to lessen the influence of task disparities on prediction bias. To tackle the discrepancy between positive and negative samples in bounding box regression tasks, this study introduces a new loss function, Focal-ECIoU, capable of expediting network convergence and improve model positioning ability. Experimental findings from the public VisDrone2019 dataset indicate that YOLO-UAV outperforms other advanced object detection methods in comprehensive performance. Compared with the baseline model YOLOv5s, YOLO-UAV increased mAP0.5 from 35.1% to 46.7%, while mAP0.5:0.95 increased from 19.1% to 27.4%. For small-scale targets, AP $_{small}$ increased from 10.2% to 17.3%. The experiment proves that YOLO-UAV performs well in improving object detection accuracy and has strong generalization ability, satisfying the practical requirements of UAV imagery object detection tasks.

UAV-DETR: Efficient End-to-End Object Detection for Unmanned Aerial Vehicle Imagery

DroneNet: Rescue Drone-View Object Detection

AF-DETR: efficient UAV small object detector via Assemble-and-Fusion mechanism

DV-DETR: Improved UAV Aerial Small Target Detection Algorithm Based on RT-DETR

AMFEF-DETR: An End-to-End Adaptive Multi-Scale Feature Extraction and Fusion Object Detection Network Based on UAV Aerial Images

Drone-DETR: Efficient Small Object Detection for Remote Sensing Image Using Enhanced RT-DETR Model

MCG-RTDETR: Multi-Convolution and Context-Guided Network with Cascaded Group Attention for Object Detection in Unmanned Aerial Vehicle Imagery

Efficient and Accurate Giraffe-Det for UAV Image Based Object Detection

DFS-DETR: Detailed-Feature-Sensitive Detector for Small Object Detection in Aerial Images Using Transformer

RemDet: Rethinking Efficient Model Design for UAV Object Detection

Detection of Military Targets on Ground and Sea by UAVs with Low-Altitude Oblique Perspective

APNet: Accurate Positioning Deformable Convolution for UAV Image Object Detection

End to end polysemantic cooperative mixed task trainer for UAV target detection

Object Detection for UAV Aerial Scenarios Based on Vectorized IOU

SRDD: a lightweight end-to-end object detection with transformer

UavTinyDet: Tiny Object Detection in UAV Scenes

Delving into Robust Object Detection from Unmanned Aerial Vehicles: A Deep Nuisance Disentanglement Approach

PTCDet: advanced UAV imagery target detection

YOLO-UAV: Object Detection Method of Unmanned Aerial Vehicle Imagery Based on Efficient Multi-Scale Feature Fusion

OVA-DETR: Open Vocabulary Aerial Object Detection Using Image-Text Alignment and Fusion

Object Detection Deployed on UAVs for Oblique Images by Fusing IMU Information