Small Invasion Obstacle Detection on the Railway Track Based on BD-YOLOv5

With the increase of railway operation kilometers and the increase of operation speed, ensuring the safety of train operation has become the focus of attention. Among them, the consequences of the collision event between the train and the obstacles invading the railway gauge are particularly serious. Therefore, the detection of railway obstacles is an important guarantee for the safe operation of trains. This paper proposes a BD-YOLOv5 model, which introduces the BiFormer attention mechanism to better integrate the context information around the target, so that the model can capture the characteristics of small targets more accurately. At the same time, the ODConv is used to replace the original Conv layer, which has a new multidimensional attention mechanism with a parallel strategy, learning the complementary attention of the convolution kernel along all four dimensions of the kernel space in any convolution layer. It reduces the number of parameters and also enhances the feature extraction ability of CNN basic convolution operation. The experimental results show that the Precision can reach as high as 97.5%, the Recall value achieves 98.4%, and the frame per second (FPS) is 100. BD-YOLOv5 can meet the practical application requirements of railway obstacle detection.