Presentation Title

Unsupervised Classification of Earth Surface for Landslide Detection

Faculty Mentor

Omar Mora

Start Date

18-11-2017 9:15 AM

End Date

18-11-2017 9:30 AM

Location

9-243

Session

Engineering/CS 1

Type of Presentation

Oral Talk

Subject Area

engineering_computer_science

Abstract

Landslides are geological events caused by slope instability and degradation, leading to the sliding of large masses of rock and soil down a mountain or hillside. These events are influenced by topography, geology, weather and human activity, and can cause extensive damage to the environment and infrastructure, such as the destruction of transportation networks, homes, and businesses. It is therefore imperative to detect early-warning signs of landslide hazards as a means of mitigation and disaster prevention. Traditional landslide surveillance consists of field mapping, but the process is expensive and time consuming. This study uses Light Detection and Ranging (LiDAR) derived Digital Elevation Models (DEMs) and k-means clustering and Gaussian Mixture Model (GMM) to analyze surface roughness and extract spatial features and patterns of landslides and landslide-prone areas. The methodology based on several feature extractors employs an unsupervised classifier on the Carlyon Beach Peninsula in the state of Washington to attempt to identify slide potential terrain. When compared with the independently compiled landslide inventory map, the proposed algorithm correctly classifies up to 87% of the terrain. These results suggest that the proposed methods and LiDAR-derived DEMs can provide important surface information and be used as efficient tools for digital terrain analysis to create accurate landslide maps.

Summary of research results to be presented

Landslides in the Carlyon Beach Peninsula have caused a major impact on human life, infrastructure and economy. As a result, landslide-mapping technology has undergone innovation and improvement within the last few decades to help mitigate the damages caused by such geological disasters. Traditional mapping methods such as aerial photographic analysis and field inspection are still employed internationally to detect landslide prone areas, but these methods have proven to be time consuming, costly, and unable to detect small scale failures. Meanwhile, modern technologies employ the use of DEMs and automated algorithms that are able to successfully detect landslide and non-landslide terrain with highly accurate results in a cost-effective manner. The methods performed feature extraction on the DEM of the study area by identifying stable terrain with smooth features, and terrain with rough surfaces as landslide prone. The classification results from k-means and GMM clustering, are able to attain an accuracy up to 87% when compared to the landslide inventory map. As a result, modern techniques are able to detect landslide terrain using DEMs in a time-effective manner that is affordable and accessible. Future recommendations include the use of a larger dataset and fusing several feature extractors to perform the classification. This may help determine if the results can be improved and if the techniques can be applied to other regions.

This document is currently not available here.

Share

COinS
 
Nov 18th, 9:15 AM Nov 18th, 9:30 AM

Unsupervised Classification of Earth Surface for Landslide Detection

9-243

Landslides are geological events caused by slope instability and degradation, leading to the sliding of large masses of rock and soil down a mountain or hillside. These events are influenced by topography, geology, weather and human activity, and can cause extensive damage to the environment and infrastructure, such as the destruction of transportation networks, homes, and businesses. It is therefore imperative to detect early-warning signs of landslide hazards as a means of mitigation and disaster prevention. Traditional landslide surveillance consists of field mapping, but the process is expensive and time consuming. This study uses Light Detection and Ranging (LiDAR) derived Digital Elevation Models (DEMs) and k-means clustering and Gaussian Mixture Model (GMM) to analyze surface roughness and extract spatial features and patterns of landslides and landslide-prone areas. The methodology based on several feature extractors employs an unsupervised classifier on the Carlyon Beach Peninsula in the state of Washington to attempt to identify slide potential terrain. When compared with the independently compiled landslide inventory map, the proposed algorithm correctly classifies up to 87% of the terrain. These results suggest that the proposed methods and LiDAR-derived DEMs can provide important surface information and be used as efficient tools for digital terrain analysis to create accurate landslide maps.