SOD-YOLOv8-Enhancing YOLOv8 for Small Object Detection in Aerial Imagery and Traffic Scenes.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Object detection is vital for autonomous systems and smart cities, but detecting small objects from high-altitude cameras remains a significant challenge.
• Existing models struggle with small object detection due to factors like object size, distance, varied shapes, and cluttered backgrounds.

Purpose of the Study:

• To develop a novel model, small object detection YOLOv8 (SOD-YOLOv8), specifically designed to enhance the detection of numerous small objects in challenging imaging conditions.
• To improve the accuracy and efficiency of small object detection in computer vision applications.

Main Methods:

• Proposed SOD-YOLOv8 model enhances YOLOv8 by integrating features across different levels using efficient generalized feature pyramid networks (GFPNs).
• Introduced a fourth detection layer for high-resolution spatial information utilization and an efficient multi-scale attention module (EMA) for enhanced feature extraction.
• Implemented powerful-IoU (PIoU) loss function to improve bounding box regression accuracy and convergence speed.

Main Results:

• SOD-YOLOv8 demonstrated significant improvements in small object detection metrics, including recall (40.1% to 43.9%), precision (51.2% to 53.9%), mAP_0.5 (40.6% to 45.1%), and mAP_0.5:0.95 (24% to 26.6%).
• The model showed superior performance compared to widely used models without substantial increases in computational cost or latency.
• Real-world traffic scene experiments confirmed SOD-YOLOv8's reliability and effectiveness in diverse environmental conditions.

Conclusions:

• SOD-YOLOv8 effectively addresses the challenges of small object detection from high-altitude cameras, offering enhanced accuracy and efficiency.
• The proposed model provides a robust solution for critical applications like traffic management, autonomous driving, and smart city initiatives.
• SOD-YOLOv8 represents a significant advancement in object detection technology, particularly for scenarios with numerous small objects.