LASA: A Line Anchor Self-Assigned Network for Rail Detection

Rail detection is a critical component of railway anomaly detection systems. However, the visibility of rails in images can be significantly affected by factors such as lighting conditions, rail intersections, and vegetation coverage, making the task of accurate rail detection a considerable challenge. To solve the above problems, we propose a Line Anchor Self-Assigned Network for rail detection. This network achieves self-assigned of line anchors by adaptively adjusting their parameters multiple times, thereby improving detection performance. Specifically, our method designs a Layer Aggregation Module to align and aggregate multi-scale semantic features extracted from the backbone network and establish global long-range dependencies. The Self-Assigned Module then adaptively adjust line anchor parameters multiple times to allocate line anchors based on aggregation features. These allocated line anchors are used to accurately fit the rail instances. Additionally, to handle scenes with low visibility and occlusion, we propose Rail Contrastive Learning. This module combines locally extracted features with globally captured cross-similarity features to achieve accurate rail detection. To evaluate the effectiveness of our method, we conducted extensive experiments on a self-built database called RDD, which contains 15000 images of railway in different scenarios. The experimental results show that our method has achieved state-of-the-art performance.