Progressive Extraction of Railway Tracks via Fused Attention and Residual Mechanisms

To address the challenges associated with the large data volume, extensive distribution, and inherent complexity of railway data, as well as the difficulty in removing objects adjacent to the railway tracks, we propose a progressive railway track extraction method. This method integrates a fused attention mechanism and residual architecture to enhance the accuracy and efficiency of track extraction. Our method is divided into two distinct modules: the ballast bed extraction module and the railway track extraction module. The approach begins with preprocessing the raw point cloud data to facilitate the initial extraction of the ballast bed region. Following this, vegetation points adjacent to the ballast bed are isolated to enhance the accuracy of extraction process. For the railway track extraction, we utilize a sampling and grouping strategy to capture local features from diverse regions using PointNet. The feature learning phase incorporates both residual and attention mechanism to effectively fuse spatial and channel-wise information, thereby enhancing feature representation. Finally, the extracted features are passed through a fully connected layer for the final railway track extraction. Experimental results demonstrate that the proposed method delivers robust performance in accurately extracting railway tracks across various railway environments, showcasing its effectiveness and adaptability in complex scenarios. The proposed method achieves an overall accuracy of 95.84%, with a mean intersection over union of 89.33% for railway track extraction. Additionally, the F1-score for ballast bed extraction exceeds 94%, highlighting the effectiveness and precision of the approach in both track and ballast bed extraction tasks. These results underscore the robustness and reliability of the method in handling complex railway environments.