YOLOv8-MPEB small target detection algorithm based on UAV images

Area of Science:

• Computer Vision
• Artificial Intelligence
• Remote Sensing

Background:

• Unmanned Aerial Vehicle (UAV) aerial imagery presents significant challenges for target detection, including scale variations, small object sizes, and complex scenes.
• Existing detection algorithms often suffer from missed or false detections in UAV applications due to these inherent difficulties.

Purpose of the Study:

• To develop an efficient and accurate algorithm for small target detection in UAV aerial images.
• To enhance the YOLOv8 model to address the specific challenges of UAV imagery, improving detection performance and model efficiency.

Main Methods:

• Proposed YOLOv8-MPEB algorithm replaces the CSPDarknet53 backbone with the lightweight MobileNetV3.
• Incorporated a dedicated small target detection layer and the Efficient Multi-Scale Attention (EMA) mechanism within the C2f module.
• Utilized a bidirectional feature pyramid network (BiFPN) in the Neck segment for improved scale variation handling.

Main Results:

• The YOLOv8-MPEB algorithm achieved a mean Average Precision (mAP) of 91.9% on a custom helmet and reflective clothing dataset.
• The model has 7.39 million parameters and a size of 14.5 MB, representing a 34% reduction in parameters and 32% reduction in size compared to standard YOLOv8.
• Demonstrated superior accuracy and speed compared to other prevalent detection algorithms.

Conclusions:

• The proposed YOLOv8-MPEB algorithm effectively enhances small target detection in UAV aerial images.
• The model offers a significant improvement in accuracy and efficiency, making it suitable for real-world UAV applications.
• The study validates the algorithm's effectiveness through ablation experiments and comparative analysis.