Abstract: The track foreign object detection algorithm based on improved YOLOv5 is proposed for problems such as unclear images in railroad scenes and the existence of small targets and obscured targets that are not detected. Firstly, the adaptive parameters are adjusted according to the maximum value of the line characteristics of the scene, and the rail area is divided by the weight of the boundary points. Then, for the problem that the image is not clear, the CLAHE(Contrast-limited Adaptive Histogram Equalization)is used to improve the contrast of the picture and reduce the external influence. Finally, on the basis of the original YOLOv5 algorithm, the CBAM module is introduced to improve the feature extraction ability and the leakage detection of occluded targets and small targets. The CIOU loss function is used to replace the GIOU loss function as the bounding box regression loss function to accelerate the convergence speed of the model and improve the accuracy of border localization. The experimental results show that the model achieves an average detection accuracy of mean average precision of 0.941 and detection speed of 39 frames per second, which can quickly and accurately detect the foreign objects present on the railroad and meet the requirements of real-time target detection.