Research on road surface damage detection based on SEA-YOLO v8

Area of Science:

• Computer Vision
• Artificial Intelligence
• Road Infrastructure Management

Background:

• Existing road damage detection methods lack accuracy, real-time capabilities, and adaptability.
• Traffic safety and road maintenance are critical areas impacted by these limitations.

Purpose of the Study:

• To develop an advanced road damage detection model (SEA-YOLO v8) addressing current technological shortcomings.
• To enhance detection accuracy, real-time processing, and model adaptability for complex urban environments.

Main Methods:

• Constructed the SBS module to optimize computational complexity and reduce model parameters for a lightweight design.
• Integrated the EMA attention mechanism into the neck component for improved feature representation and selective focus.
• Proposed an adaptive attention feature pyramid structure to enhance feature fusion capabilities.
• Introduced the lightweight shared convolutional detection head (LSCD-Head) to further refine feature representation and reduce parameters.

Main Results:

• The SEA-YOLO v8 model achieved 63.2% mAP50 on the RDD2022 dataset.
• Demonstrated superior performance compared to the standard YOLOv8 model and other mainstream target detection models.
• Validated high detection accuracy and adaptability in complex urban traffic scenarios.

Conclusions:

• The SEA-YOLO v8 model offers accurate road damage localization and detection.
• The model contributes to saving manpower and material resources in road assessment and maintenance.
• Promotes sustainable urban road development through improved infrastructure management.