Unsupervised Deep Learning for Landslide Detection from Multispectral Sentinel-2 Imagery

Abstract

This paper proposes a new approach based on an unsupervised deep learning (DL) model for landslide detection. Recently, supervised DL models using convolutional neural networks (CNN) have been widely studied for landslide detection. Even though these models provide robust performance and reliable results, they depend highly on a large labeled dataset for their training step. As an alternative, in this paper, we developed an unsupervised learning model by employing a convolutional auto-encoder (CAE) to deal with the problem of limited labeled data for training. The CAE was used to learn and extract the abstract and high-level features without using training data. To assess the performance of the proposed approach, we used Sentinel-2 imagery and a digital elevation model (DEM) to map landslides in three different case studies in India, China, and Taiwan. Using minimum noise fraction (MNF) transformation, we reduced the multispectral dimension to three features containing more than 80% of scene information. Next, these features were stacked with slope data and NDVI as inputs to the CAE model. The Huber reconstruction loss was used to evaluate the inputs. We achieved reconstruction losses ranging from 0.10 to 0.147 for the MNF features, slope, and NDVI stack for all three study areas. The mini-batch K-means clustering method was used to cluster the features into two to five classes. To evaluate the impact of deep features on landslide detection, we first clustered a stack of MNF features, slope, and NDVI, then the same ones plus with the deep features. For all cases, clustering based on deep features provided the highest precision, recall, F1-score, and mean intersection over the union in landslide detection.

Landslide4Sense: Reference Benchmark Data and Deep Learning Models for Landslide Detection

“…Over one month, the total rainfall of more than 1200 mm triggered severe landslides and flash floods in this rural district. With this rainfall, several landslides occurred, many of them were debris flows, claiming 16 lives and damaged around 200 villages [65]. In this study area, the dominant geomorphological setting consists of highly dissected and sloping structural hill ranges.…”

“…The whole or some parts of landslide-affected areas in this district have already been evaluated for landslide detection by different researchers. For example, Shahabi et al [65] developed an unsupervised learning approach to detect landslides in the Kodagu district without using any inventory dataset.…”

“…slides, while the deep-seated ones were mainly the larger ones triggered by the typhoon. Some studies have evaluated supervised and unsupervised DL solutions for detecting the landslides triggered by this typhoon in some parts of Western Taitung County (see, for example, Ghorbanzadeh et al [5] and Shahabi et al [65]). Our landslide inventory dataset for this study area is compiled from previous studies and visual interpretation of Google Earth's archive images from 2011 to 2013.…”

A Sustainable Approach to Geomatic Methods in Ravine Monitoring and Associated Natural Disaster Prevention

“…However, the mentioned approaches are limited because of the accessibility to remote areas for field surveys. Moreover, these approaches depend on the visual interpretation of expert experience and knowledge" [10] GNSS is used to define all the navigation systems that are based on satellites. Topographers use GNSS receivers to determine the coordinates of one or more points located on Earth's surface with precision, in a given coordinate system.…”

Landslide Susceptibility Prediction Based on High-Trust Non-Landslide Point Selection

“…Semi-supervised learning has been widely used in sample data analysis and evaluation [26][27][28]. In landslide susceptibility prediction and landslide detection, supervised learning frameworks, semi-supervised learning frameworks, and unsupervised learning frameworks have also demonstrated their superiority [29][30][31]. This paper selects Fu'an City, Fujian Province, China, as the research area.…”