• Lightweight Design: The model structure is designed to be more compact, reducing computational resource requirements.

• Low Power Consumption: The introduction of spiking neural networks for initial classification tasks significantly lowers power consumption.

• Low Data Dependency: By incorporating a contrastive learning framework, our method effectively handles small sample datasets, minimizing the reliance on large-scale datasets.

The dataset depicted in the image is one of the most comprehensive collections in the field of rail defect detection. Captured in real-world rail scenarios, this dataset includes samples under varying lighting conditions and track surface states. The presence of background elements in the rail images adds complexity, posing additional challenges for segmentation models.