CAS Landslide Dataset: A Large-Scale and Multisensor Dataset for Deep Learning-Based Landslide Detection.

Area of Science:

• Geosciences
• Artificial Intelligence
• Remote Sensing

Background:

• Landslide occurrences are increasing due to climate change and seismic activity.
• Accurate and efficient landslide recognition is crucial for disaster management.
• Existing datasets have limitations in size, coverage, sensor diversity, and resolution.

Purpose of the Study:

• To introduce the CAS Landslide Dataset, a large-scale, multisensor resource for deep learning-based landslide detection.
• To address the limitations of existing landslide datasets.
• To provide a benchmark for developing and evaluating advanced landslide identification models.

Main Methods:

• Compilation of 20,865 images from satellite and unmanned aerial vehicle (UAV) data.
• Integration of data from nine distinct geographical regions.
• Establishment of a robust quality evaluation methodology for dataset reliability.

Main Results:

• The CAS Landslide Dataset provides a comprehensive and high-resolution resource exceeding the scope of previous datasets.
• The dataset integrates diverse data sources, enhancing its applicability for various landslide detection scenarios.
• A rigorous quality assessment ensures the dataset's reliability for research and development.

Conclusions:

• The CAS Landslide Dataset serves as a valuable benchmark for advancing deep learning techniques in landslide identification.
• It enables enhanced prediction, monitoring, and analysis capabilities for researchers.
• The dataset facilitates the development of automated landslide detection systems, contributing to improved hazard mitigation.