Foreign object detection (FOD) in railway catenary systems is crucial for ensuring operational safety and preventing catastrophic failures. However, current detection frameworks encounter two significant challenges. First, the infrequency of fault events leads to severe data scarcity, hampering the training and validation of robust detection models. Second, although lightweight networks (e.g., MobileNet, YOLO) achieve compactness by compressing channel factors, they struggle to balance local feature extraction with global dependency modeling. To address these challenges, we propose a solution that includes the following: 1) RailFOD23, a publicly available dataset created using generative AI to mitigate data scarcity; and 2) EPRepSADet, a compact detection framework that utilizes a re-parameterizable bottleneck (Re-bottleneck) and lightweight self-attention (LSA) module for efficient FOD. The Re-bottleneck consolidates multi-branch structures into a single-path representation, whereas LSA facilitates element-wise attention modeling to effectively reduce computational complexity. In addition, the efficient detection head further minimizes model complexity through hierarchical semantic modeling. Extensive experiments demonstrate that EPRepSADet achieves a mean Average Precision (mAP) of 92.5% on the RailFOD23 test set, requiring only 1.7G FLOPs, thus outperforming several state-of-the-art baseline models.

The computational power required for training deep learning models has been skyrocketing in the past decade as they scale with big data, and has become a very expensive and scarce resource. Therefore, distributed training, which can leverage distributed available computational power, is vital for efficient large-scale model training. However, most previous distributed training frameworks like DDP and DeepSpeed are primarily designed for co-located clusters under homogeneous computing and communication conditions, and hence cannot account for geo-distributed clusters with both computing and communication heterogeneity. To address this challenge, we develop a new data parallel based distributed training framework called Co-Adaptive Data Parallelism (C-ADP). First, we consider a data owner and parameter server that distributes data to and coordinates the collaborative learning across all the computing devices. We employ local training and delayed parameter synchronization to reduce communication costs. Second, we formulate a data parallel scheduling optimization problem to minimize the training time by optimizing data distribution. Third, we devise an efficient algorithm to solve this scheduling problem, and formally prove that the obtained solution is optimal in the asymptotic sense. Experiments on the ImageNet100 dataset demonstrate that C-ADP achieves fast convergence in heterogeneous distributed training environments. Compared to Distributed Data Parallel (DDP) and DeepSpeed, C-ADP achieves 21.6 times and 26.3 times improvements in FLOPS, respectively, and a reduction in training time of about 72% and 47%, respectively.

In this paper, we introduce a novel image-goal navigation approach, named RFSG. Our focus lies in leveraging the fine-grained connections between goals, observations, and the environment within limited image data, all the while keeping the navigation architecture simple and lightweight. To this end, we propose the spatial-channel attention mechanism, enabling the network to learn the importance of multi-dimensional features to fuse the goal and observation features. In addition, a self-distillation mechanism is incorporated to further enhance the feature representation capabilities. Given that the navigation task needs surrounding environmental information for more efficient navigation, we propose an image scene graph to establish feature associations at both the image and object levels, effectively encoding the surrounding scene information. Cross-scene performance validation was conducted on the Gibson and HM3D datasets, and the proposed method achieved state-of-the-art results among mainstream methods, with a speed of up to 53.5 frames per second on an RTX3080. This contributes to the realization of end-to-end image-goal navigation in real-world scenarios. The implementation and model of our method have been released at: https://github.com/nubot-nudt/RFSG.

3D perception in train operating environments presents significant challenges, as it must ensure both precise distance estimation and computational efficiency to meet stringent braking requirements. To date, existing 3D detection architectures, which employ dense voxel or pillar representations, encounter challenges of computational inefficiency and accuracy degradation when processing large-scale railway Light Detection And Ranging (LiDAR) data. To address this challenge, we propose RailVoxelDet, a railway-optimized 3D detector integrating the Multi-factor Dynamic Voxel Feature Encoder (MDVFE) and efficient backbone. Specifically, MDVFE converts point clouds to 2D sparse voxels, reducing computational complexity. The backbone employs residual bottlenecks with shared full connected layers and sparse convolutions, enhanced by the SimAM-Point module. Additionally, the Feature Query and Matching Module (FQMM) is proposed to establish a bottom-up multi-level feature fusion architecture. Experimental results show RailVoxelDet reaches 71.29% mAP on OSDaR23 and 61.94% mAP on AirR24, with 6.42G FLOPs and a 71.42ms inference time. It outperforms 12 comparison models, delivering state-of-the-art results.