OBC-YOLOv8: an improved road damage detection model based on YOLOv8

Area of Science:

• Computer Vision
• Machine Learning
• Road Infrastructure Management

Background:

• Effective pavement distress detection is crucial for road maintenance and safety.
• Existing methods often struggle with complex and diverse damage features.
• Accurate road condition assessment requires robust and efficient detection systems.

Purpose of the Study:

• To propose a novel road damage detection method, OBC-YOLOv8, for improved pavement distress identification.
• To enhance the feature extraction capabilities and detection accuracy of deep learning models for road damage.
• To provide a more efficient and reliable solution for road maintenance and management.

Main Methods:

• The proposed OBC-YOLOv8 model integrates omni-dimensional dynamic convolution (ODConv) for adaptive feature learning.
• Incorporation of Bottleneck Transformer (BoTNet) in the backbone to simultaneously extract global and local features.
• Utilizing a coordinate attention mechanism (CA) in the Neck section to refine detection and reduce interference.

Main Results:

• The OBC-YOLOv8 model demonstrated superior performance on the RDD2022-China dataset.
• Achieved a 1.8% increase in mean average precision 50 (mAP@0.5) compared to baseline models.
• Showcased a 1.6% improvement in F1-score, indicating enhanced detection efficacy.

Conclusions:

• OBC-YOLOv8 offers a significant advancement in automated pavement distress detection.
• The integration of ODConv, BoTNet, and CA effectively boosts model performance.
• This method provides a promising tool for efficient and accurate road maintenance strategies.