Developing an integrated approach based on geographic object-based image analysis and convolutional neural network...

Area of Science:

• Geosciences
• Remote Sensing
• Artificial Intelligence

Background:

• Traditional landform modeling is labor-intensive and time-consuming.
• Digital approaches are increasingly vital for landform analysis.
• Accurate landform mapping is crucial for understanding earth surface processes.

Purpose of the Study:

• To analyze the effectiveness of combining Convolutional Neural Networks (CNN) with Geographic Object-Based Image Analysis (GEOBIA) for mapping volcanic and glacial landforms.
• To develop an automated workflow for rapid and accurate landform detection.
• To assess the performance of this integrated approach for regional and large-scale applications.

Main Methods:

• Utilized Sentinel-2 imagery and Digital Elevation Model (DEM) derivatives (slope, aspect, curvature, flow accumulation).
• Applied multi-resolution segmentation and feature selection for landform categorization.
• Developed object-based features (spectral, geometrical, textural) for CNN training and testing.
• Trained and validated landform models using GEOBIA-generated objects and ground control points.

Main Results:

• The integrated GEOBIA and CNN approach achieved high accuracy (ACC > 0.9600) for various volcanic and glacial landforms.
• Specific landforms like dacite lava, caldera, andesite lava, volcanic cone, volcanic tuff, glacial circus, glacial valley, and suspended valley were mapped with high precision.
• Cross-validation accuracy exceeded 0.9400, confirming the robustness of the model.

Conclusions:

• The integrated GEOBIA and CNN method offers a fast and efficient solution for landform mapping.
• This approach is highly recommended for regional and large-scale mapping of volcanic and glacial features.
• The study highlights the potential of AI and remote sensing for advancing earth surface management.