Research on a CEF-YOLOv8n-Based Method for Small Object Detection in UAV Aerial Imagery

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Small object detection in Unmanned Aerial Vehicle (UAV) imagery faces challenges due to high object proportions, low resolution, and scale variations.
• Existing algorithms struggle to effectively capture and process these intricate features, limiting performance in critical applications.

Purpose of the Study:

• To develop an advanced UAV small-object detection and recognition algorithm that overcomes current limitations.
• To enhance feature extraction, fusion, and model efficiency for improved accuracy in challenging UAV scenarios.

Main Methods:

• The proposed algorithm, CEF-YOLOv8n, utilizes YOLOv8n as a baseline, incorporating a Partial Convolution-based Cross Partial Feature (CPF) module in the backbone.
• A Focusing Generalized Feature Pyramid Network (FGFPN) and a Feature Semantic Fusion Module (FSFM) with cross-attention are employed in the neck for multi-scale feature fusion.
• A Lightweight Weight-Sharing Detection Head (LWSD) is introduced to optimize computational efficiency and real-time performance.

Main Results:

• CEF-YOLOv8n achieved 37.6% mAP50 and 22.6% mAP50-95, outperforming the original YOLOv8n by 3.4% and 2.6% respectively.
• The model demonstrated improved computational efficiency, reducing parameters from 3.2 M to 2.5 M and FLOPs from 8.7 G to 6.9 G.
• Experiments on public datasets validated the algorithm's effectiveness against other state-of-the-art methods.

Conclusions:

• The proposed CEF-YOLOv8n algorithm effectively addresses the challenges of small object detection in UAV imagery.
• The integration of CPF, FGFPN, FSFM, and LWSD modules significantly enhances detection accuracy and model efficiency.
• This work provides a robust and efficient solution for real-time small object recognition in UAV applications.