The invention discloses a semi-supervised railway track foreign matter detection method based on an inverse perspective drawing. The method comprises the following steps: firstly, obtaining the inverse perspective drawing with the same resolution ratio as a camera, and fusing the inverse perspective drawing into a Teamer-Student model architecture; in the model training stage, only normal railway line sample data is adopted, meanwhile, after scale zooming is conducted on an inverse perspective drawing, point-by-point multiplication is conducted on the inverse perspective drawing and a middle layer feature graph of the Student network, and then knowledge distillation operation is conducted on the inverse perspective drawing and a feature graph corresponding to the Teamer network; in the test stage, if the image contains the track foreign matter, the intermediate layer features of the Teamer network and the Student network show obvious differences, the feature differences are utilized to construct a multi-scale anomaly graph, and finally the specific position of the foreign matter in the image is determined through the anomaly graph. According to the invention, the foreign matter detection capability at a long distance can be effectively enhanced, and the interference of non-foreign matter targets at a short distance can be slowed down.
本发明公开了一种基于逆透视图的半监督铁路轨道异物检测方法:首先获得与摄像机分辨率相同的逆透视图并将其融入Teacher‑Student模型架构;在模型训练阶段只采用正常的铁路线路样本数据,同时将逆透视图经尺度缩放后与Student网络的中间层特征图进行逐点相乘,再与Teacher网络所对应的特征图进行知识蒸馏操作;在测试阶段,如果图像中含有轨道异物,Teacher网络与Student网络的中间层特征就会表现出明显的差异,利用这些特征差异构建多尺度的异常图,最终通过异常图来确定异物在图像中的具体位置。本发明能有效增强远距离处异物的检出能力,减缓近距离处非异物目标的干扰。
Semi-supervised railway track foreign matter detection method based on inverse perspective drawing
一种基于逆透视图的半监督铁路轨道异物检测方法
2024-04-16
Patent
Electronic Resource
Chinese
Track foreign matter detection method based on YOLO algorithm
European Patent Office | 2025
|Railway foreign matter clearance intrusion detection system and method
European Patent Office | 2015
|High-speed railway line foreign matter detection method and device
European Patent Office | 2023
|