Rural Road Extraction in Xiong'an New Area of China Based on the RC-MSFNet Network Model.

Area of Science:

• Remote Sensing
• Geographic Information Systems
• Computer Vision

Background:

• Rural road extraction from high-resolution imagery is challenging due to narrow road widths, blurred boundaries, and similar textures to surrounding environments.
• Existing methods often result in incomplete extraction and low accuracy for rural roads.
• The Xiong'an New Area presents complex rural terrain, necessitating improved road extraction techniques for development planning.

Purpose of the Study:

• To develop and evaluate a novel deep learning model, RC-MSFNet, for enhanced rural road extraction.
• To address the limitations of existing models in accurately identifying narrow, elongated, and boundary-obscured rural roads.
• To construct a dedicated rural road dataset (XARoads) for model training and validation.

Main Methods:

• The RC-MSFNet model, based on the U-Net architecture, incorporates residual neural networks to mitigate vanishing gradients and a connectivity attention mechanism for improved road completeness.
• A multi-scale fusion atrous convolution module is employed in the bottleneck to capture features at various scales.
• The model was trained and tested on the XARoads dataset and the DeepGlobe dataset, with comparisons against U-Net, FCN, SegNet, DeeplabV3+, R-Net, and RC-Net.

Main Results:

• RC-MSFNet achieved precision (P) of 0.8350, intersection over union (IOU) of 0.6523, and completeness (COM) of 0.7489 on the XARoads dataset.
• The proposed method demonstrated significant precision improvements over benchmark models, ranging from 0.58% to 7.85%.
• The model showed superior performance in extracting narrow, muddy, and boundary-indistinct roads, with reduced omission and false extraction errors.

Conclusions:

• The RC-MSFNet model offers a robust solution for accurate rural road extraction, particularly in challenging environments.
• The model's architecture effectively captures road connectivity and multi-scale features, leading to improved extraction performance.
• Accurate rural road data derived from this method can support urban development and planning initiatives, such as those in the Xiong'an New Area.