YOLO-Enhanced Lightweight Learning for Rapid and Accurate Object Detection

Reliable information exchange is essential for train operation systems, especially in tunnels where leaky cables provide stable signal coverage and flexible deployment. However, manual inspection of cable clamps is inefficient and subjective, while existing automated methods struggle with lighting changes, background noise, and real-time performance. This paper proposes a defect detection method based on a lightweight network and adaptive data augmentation. A dedicated dataset with diverse scenarios and lighting conditions is constructed, and a hybrid augmentation strategy is applied. The network uses MobileNetV3-small to rebuild the YOLOv5s backbone, with optimizations like channel compression, SE module removal, DWConv, and C3Ghost modules. Experiments show that the method achieves high accuracy and fast inference, effectively handling lighting variations, background interference, and small target detection in tunnels. This approach offers a practical solution for intelligent tunnel maintenance.