Enhanced SAR-Optical Image Fusion Based on RFN-Nest for Railway Intelligent Perception

As railway networks expand and high-speed trains increase in speed, ensuring train safety has become increasingly critical. Addressing significant challenges in the intelligent detection of anomalies over extended distances and in advance is essential for maintaining safe operations. Perception of the external environment plays a critical role in the safety monitoring of train operations, particularly for slowly evolving emergencies such as slope deformations and railway subsidence. However, human senses are often inadequate in responding to sudden hazards and fail to detect gradual changes. Consequently, integrating a variety of sensors to develop a new railway safety monitoring system is essential. This paper proposes a multi-source information fusion system for railway hazard perception, which enhances the perception ability of emergency situations, leveraging the advantages of improved information reliability and enhanced accuracy in target recognition by fusing Synthetic Aperture Radar (SAR) and optical image data. Additionally, based on RFN-Nest an image fusion network architecture is proposed, which is incorporating an attention mechanism to improve the model's fusion performance. Finally, the proposed method is validated using the WHU-OPT-SAR dataset, demonstrating its effectiveness.