The key components of the train are essential for ensuring the safe operation of the train. The current detection algorithm based on deep learning has poor detectability under poor lighting conditions and small component size. To solve this problem, this study proposes a detection algorithm for key components of a train based on improved RetinaNet. First, a receptive field block module was introduced after shallow feature P3 to improve the receptive field and feature quality of the P3 feature layer. Then, the feature pyramid network was replaced with a pixel aggregation net and the positioning ability of the feature pyramid was enhanced by adding a bottom-up feature fusion path. Finally, by adjusting the experimental parameters and the location of the network detection layer, a network model suitable for detecting key components of the train was obtained. Results show that the proposed model is superior to the original RetinaNet in the open dataset PASCAL VOC. Furthermore, it is superior to the current mainstream algorithm in detecting the key components of the train.