Enhanced feature representation for real time UAV image object detection using contextual information and adaptive...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Real-time object detection in Unmanned Aerial Vehicle (UAV) imagery faces challenges like small objects, occlusion, and uneven distribution.
• Existing algorithms, including YOLO variants, show performance degradation when directly applied to UAV datasets.
• Current solutions lack comprehensive approaches for real-world UAV deployment scenarios.

Purpose of the Study:

• To develop an enhanced real-time object detection network (YOLO-UD) for improved performance on UAV imagery.
• To address the specific challenges of detecting small objects, occlusion, and varying target distributions in UAV data.
• To achieve a superior balance between accuracy and inference speed for UAV-based downstream tasks.

Main Methods:

• Introduced YOLO-UD, built upon the YOLO11 architecture, incorporating a novel C3kHR module with dilated convolutions for multi-scale feature representation.
• Designed an Efficient Adaptive Feature Fusion Network (EAFN) to filter and prioritize multi-scale feature information.
• Integrated a Small Object Detection Layer (SMDL) to specifically enhance the detection of small targets.

Main Results:

• YOLO-UD demonstrated significant improvements in real-time object detection accuracy and speed on UAV imagery.
• The C3kHR module effectively captured rich contextual and multi-scale features.
• EAFN successfully filtered and prioritized relevant information, while SMDL enhanced small object detection capabilities.

Conclusions:

• YOLO-UD offers a robust and effective solution for real-time object detection in challenging UAV environments.
• The proposed network achieves a strong balance between accuracy and inference speed, outperforming existing methods.
• The findings validate the effectiveness of integrating contextual information and adaptive fusion for UAV image analysis.