Application of YOLO11 Model with Spatial Pyramid Dilation Convolution (SPD-Conv) and Effective Squeeze-Excitation...

Area of Science:

• Railway Engineering
• Computer Vision
• Deep Learning

Background:

• Object detection algorithms are increasingly used for railway track defect detection.
• Existing methods face challenges with low efficiency and inadequate accuracy.

Purpose of the Study:

• To develop an improved You Only Look Once (YOLO) object detection algorithm for enhanced railway track defect detection.
• To address limitations in efficiency and accuracy of current track inspection methods.

Main Methods:

• Modified the YOLO11 backbone with Spatial Pyramid Dilation Convolution (SPD-Conv) for improved low-resolution and small object detection.
• Integrated Effective Squeeze-Excitation (EffectiveSE) attention mechanism to enhance feature representation.
• Added a small target detection head to the neck network for multi-scale target capture.

Main Results:

• Achieved 95.9% mAP@0.5 on a track fastener dataset and 89.5% mAP@0.5 on a track surface dataset.
• Demonstrated superior performance compared to the original YOLO11 model and other object detection algorithms.
• Significantly improved the efficiency and accuracy of track defect detection.

Conclusions:

• The enhanced YOLO11 algorithm effectively improves railway track defect detection.
• The proposed modifications provide superior performance in identifying track anomalies.
• This approach offers a more efficient and accurate solution for railway infrastructure inspection.