Abstract: As an important aspect of train safety maintenance, the detection of high-speed railway catenary support components (CSCs) usually faces problems such as multiple types of targets, large differences in target scales, and small dimensions of some components. In response to the above issues, traditional object detection algorithms based on deep learning are prone to insufficient feature fusion. This article proposes a detector based on multi-scale fusion pyramid focus network, which is used to detect all components of CSCs at different scales. Firstly, the designed separable residual pyramid aggregation module (SRPAM) and deformable convolutional network (DCN) are introduced into the backbone network to optimize its multi-scale feature extraction capability and adaptability to multi-scale targets. Among them, dense connections, deformable convolution modules, residual structures, and attention mechanisms are referenced in the atrous spatial pyramid pooling (ASPP) module to ensure that it can enrich feature information and expand receptive fields without significantly increasing computational complexity. Secondly, a cross layer feature balancing module is introduced in the path aggregation feature pyramid network (PA-FPN) to optimize the cross-layer feature fusion effect, weaken background information, and improve the detection performance for small-scale object detection. Finally, deformable convolution is introduced into FCOS Head to further optimize the comprehensive detection performance of the model. Comparative experiments have shown that the proposed fusion pyramid focus network is superior to most comparison methods. In addition, the proposed detection framework achieved a detection accuracy (mAP) of 48.6% on the CSCs dataset, while maintaining a low computational complexity (FLOPs=38.6). Therefore, the method proposed in this article can be effectively applied to the detection of CSCs.