BD-YOLOv8s: enhancing bridge defect detection with multidimensional attention and precision reconstruction

Area of Science:

• Computer Vision
• Structural Health Monitoring
• Artificial Intelligence

Background:

• Object detection networks like YOLO show promise but struggle with bridge defect detection due to complex backgrounds and variable defect appearances.
• Existing methods face challenges including low accuracy, missed detections, and false positives in bridge imagery.
• Accurate detection of concrete bridge defects is crucial for infrastructure maintenance and safety.

Purpose of the Study:

• To develop an advanced object detection methodology, BD-YOLOv8s, specifically tailored for concrete bridge defect detection.
• To enhance the accuracy and reliability of detecting bridge defects by addressing limitations of existing YOLO models.
• To improve the robustness of defect detection algorithms in challenging real-world bridge environments.

Main Methods:

• Utilized YOLOv8s as the baseline architecture for bridge defect detection.
• Integrated ODConv (Orthogonal Deep Convolution) into the second convolutional layer to enhance feature extraction.
• Incorporated the CBAM (Convolutional Block Attention Module) into C2F modules to leverage spatial and channel attention mechanisms.
• Replaced traditional upsampling with CARAFE (Content-Aware Reassembly of Features) for improved feature map reconstruction.

Main Results:

• BD-YOLOv8s achieved a mean Average Precision (mAP) of 86.2% at IoU threshold 0.5.
• The model obtained 56% mAP@0.5:0.95, indicating strong performance across various overlap thresholds.
• BD-YOLOv8s demonstrated significant improvements over the baseline, with increases of 5.3% and 5.7% in mAP@0.5 and mAP@0.5:0.95, respectively.
• The proposed method substantially reduced false positives and missed detections.

Conclusions:

• BD-YOLOv8s offers a superior approach for concrete bridge defect detection compared to standard YOLOv8s.
• The integration of ODConv, CBAM, and CARAFE effectively enhances the model's ability to handle complex bridge defect imagery.
• The improved accuracy and reduced detection errors make BD-YOLOv8s a valuable tool for structural health monitoring of bridges.