In the context of unmanned aerial vehicle(UAV)perspectives over railway track beds,where the target objects are relatively small and the background is complex,existing object detection networks often suffer from issues such as false positives and misses.This paper proposes improvements to the existing YOLOv5l model.Firstly,data augmentation techniques are employed to increase the effective number of samples.Sec-ondly,to mitigate interference from complex backgrounds and enable the model to focus more on target infor-mation,a bidirectional routing attention mechanism is introduced.Finally,the Wise-IoU(WIoU)loss function is utilized instead of the original loss function to address the balance between high-quality and low-quality sam-ples,thereby enhancing the model's detection performance.Experimental results on a custom dataset of missing cover plate demonstrate that the improved model achieves an average detection accuracy of 89.5%,compared to the original YOLOv5l's 74.2%,meeting the requirements for detecting missing cover defects in track beds.
关键词
无人机影像/YOLOv5l/注意力机制/损失函数
Key words
UAV imagery/YOLOv5l/attention mechanism/loss function
维普
万方数据