Research on the Extraction of Hazard Sources along High-Speed Railways from High-Resolution Remote Sensing Images...

Area of Science:

• Remote Sensing
• Artificial Intelligence
• Railway Engineering

Background:

• High-speed railway safety is threatened by numerous potential hazard sources.
• Traditional ground search methods are insufficient for efficient and safe hazard investigation.
• Accurate and timely identification of railway hazards is crucial for operational safety.

Purpose of the Study:

• To propose an advanced deep learning model for accurate extraction of railway hazard sources from high-resolution remote sensing images.
• To enhance the detection capabilities for small targets and boundary details of hazard sources.
• To address the challenge of class imbalance in remote sensing datasets for hazard identification.

Main Methods:

• Development of a texture-enhanced ResUNet (TE-ResUNet) model incorporating texture enhancement modules.
• Integration of a multi-scale Lovász loss function to manage class imbalance and optimize model parameters.
• Comparative analysis of TE-ResUNet against established methods like FCN8s, PSPNet, DeepLabv3, and AEUNet using the GF-2 railway hazard source dataset.

Main Results:

• TE-ResUNet demonstrated superior performance in overall accuracy, F1-score, and recall compared to existing methods.
• The texture enhancement modules effectively improved the extraction accuracy of hazard source boundaries and small targets.
• The multi-scale Lovász loss function successfully mitigated the class imbalance problem, leading to better parameter learning.

Conclusions:

• The proposed TE-ResUNet model offers an accurate and efficient solution for extracting railway hazard sources from remote sensing imagery.
• The model's ability to maintain high recall for small-area targets is critical for comprehensive railway safety monitoring.
• This research contributes a robust deep learning approach to enhance the safety and efficiency of high-speed railway operations.