Research on Railway Track Extraction Method Based on Edge Detection and Attention Mechanism

The accurate extraction of railway tracks is crucial for the development of digital railway systems. However, traditional manual methods for track extraction are both time-consuming and tedious. At the same time, current deep learning neural networks often suffer from issues such as missed detections and false positives when it comes to identifying and detecting railway track edges. To address these problems, this paper proposes an improved d-linknet convolutional neural network that integrates a specially designed edge detection module to fuse multi-level features, thereby enhancing the model's segmentation and extraction of target edges. Additionally, the network introduces a channel-spatial dual-attention mechanism to expand its perceptual field, strengthen foreground responses in the target region, and further reduce missed detections. Experimental results demonstrate that the proposed method, when tested on a railway track dataset, outperforms the original d-linknet model with an accuracy improvement of over 2% and an average intersection over union improvement of over 5%. Furthermore, this method excels in terms of classification accuracy and visual interpretation on two different datasets compared to other comparative methods.