Landslide susceptibility assessment using lightweight dense residual network with emphasis on deep spatial features

Area of Science:

• Geosciences
• Geological Engineering
• Artificial Intelligence

Background:

• Landslides pose significant global risks to societal development.
• Accurate landslide susceptibility assessment is crucial for mitigation.
• Existing methods face challenges with limited data, feature utilization, and computational cost.

Purpose of the Study:

• To propose an efficient and accurate landslide susceptibility assessment method.
• To address limitations of existing deep learning approaches in landslide prediction.
• To develop a model that minimizes computational expense while maximizing feature extraction.

Main Methods:

• A novel lightweight dense residual network (DS-DRN) was developed.
• Depthwise separable convolutions were used to reduce computational load.
• Dense connections were implemented for enhanced feature extraction and gradient flow.
• Softmax classification was employed for final susceptibility prediction.

Main Results:

• The DS-DRN method demonstrated superior prediction accuracy compared to CNN, CPCNN-RF, and U-net.
• The proposed model achieved significant savings in computational costs.
• DS-DRN effectively captured complex nonlinear relationships in landslide data.
• Deep spatial features were successfully mined and utilized.

Conclusions:

• The DS-DRN model offers a highly accurate and computationally efficient solution for landslide susceptibility assessment.
• This method effectively addresses the limitations of traditional deep learning models in this field.
• The findings support the use of DS-DRN for improved geological hazard management and sustainable development planning.