Service / Application

Deep Learning and Automation of Landslide Topography Reading Technique Using Artificial Intelligence (AI)

AI Application to Slope Disaster Prevention Technology

NEEDS

The Need for Advanced and Efficient Survey Technology in Disaster Prevention and Mitigation

With the intensification of climate change in recent years, Japan has been experiencing a lot of damages caused by landslides.
Landslides and mudslides are phenomena that cause the movement of sediment on slopes, and these phenomena leave traces of movement on the topography before they occur. In the past, engineers have developed topographic reading techniques by precisely deciphering the traces left on the terrain to predict where landslides are likely to occur.
However, there is a limit to the number of places where landslides are likely to occur that can be predicted manually, and there is a growing need to improve the efficiency and automation of terrain reading using AI.
At Nippon Koei, we are developing a technology to improve the prediction accuracy of landslide occurrence locations using deep learning. By using the topographic maps prepared by topographic interpretation engineers based on the numerical topographic information available in Japan as the teaching data, the AI can learn the traces of sediment movement that the engineers read during topographic interpretation, and extract landslide and erosion topography at a certain rate through deep learning.

Figure-1 Examples of Topographic Interpretation for Slope Disaster Prevention

SOLUTION

AI-based Automatic Reading of Topographic Precursors of Sediment Transport Phenomena

Landslides and mudslides happen in some forms before they occur.
In the past, experts have taken time to read the topography to predict landslides and mudslides and conduct preventive surveys to decipher these phenomena. In order to read the topography, a high level of technical skills such as expertise and experience in slope disaster prevention is required, and various types of information such as aerial photographs, topographic maps, and 3D representations are used for reading.
Nippon Koei has developed a technology to streamline and automate the reading of such information using AI deep learning.
By using deep learning technology, it is possible to automatically extract the color characteristics of an image. By applying this technology, the AI can learn the topographical features that engineers focus on, and use them to improve the sophistication and efficiency of the reading process.

Figure-2 AI-inferred Image of Landslide
Figure-3 AI-inferred Image of Eroded Terrain

POINT

Automatic Reading of Erosion Topography and Its Application to Volcanic Disaster Prevention

This is an introduction to the use of deep learning technology for volcanic disaster prevention efforts. In Sakurajima, Kagoshima Prefecture, mudslides frequently occur due to the mix of persistent rains and ashfall from nearby volcanic activity. It is known that the erosion (rills and gullies) topography formed in the sediment production area is a major precursor to the occurrence of mudslides, and by understanding this erosion topography, it may be possible to detect the risk of mudslides in advance.
As with landslide topography, erosion topography has been determined by experts from aerial photographs and topographic maps through its formation and development. In this paper, we introduce the possibility of learning erosion landforms by deep learning. Deep learning takes a day to learn, but inference (image output) can be done in a few hours. By taking advantage of this expediency, we are developing applications for disaster risk reduction, such as deep learning of erosion topography in advance, and having AI inference from aerial photographs taken immediately after a volcanic eruption, ashfall, or rainfall, to quickly predict where erosion topography will occur, is vital to preventive measures and post-event responses.

Figure -4 AI Applications in Volcanic Disaster Prevention and Future Prospects

Installation Results

  • Study on runoff analysis of debris flow after ashfall in 2020
  • Study on investigation method of seepage capacity and analysis of runoff during volcanic eruption in 2018