Extraction of rockfall sources by AI based on digital data
-Proposal of DX for erosion control-

Takeharu SATO, Kotaro KANEMOTO and Tetsuya ODA

Abstract

In recent years, laser surveying using drones have made it easier to obtain orthophoto and fine grid data of steep slopes. Artificial Intelligence (AI) based risk assessment methods and extracting methods for microtopographic elements using these images and topographic analysis maps are being actively studied. On the other hand, these methods have challenges in terms of work efficiency and cost, as they require a large number of images to be trained by AI and resources to create input data. In this paper, we propose an AI based extraction method for rockfall sources that can improve these challenges. The proposed method can extract all the rockfall sources while ensuring an accuracy of more than 90%, although the recall is not as good as the accuracy. When the proposed method is applied to arbitrary slopes with different geological conditions, it achieved more than90%accuracy and extracted all rockfall sources. But the recall ranged from 66% to 89% due to the different geological conditions, making it necessary to improve the accuracy of the proposed method. In addition, the proposed method achieved the following three operational improvements effective for the Digital Transformation (DX): (1) The policy of the proposed method is to examine the grid data provided as the evaluation unit. The effect of this policy is that AI training data and test data can be created by increasing the column data in the grid data file, thus enabling analysis using AI with only two files; (2) By adopting AI software packages that match the file format of the grid data, work efficiency could be improved. In addition, we propose a method that requires no introduction costs by using free software with an AI development environment; (3) We proposed a flexible method that can address various issues by simply changing the data according to the objective. In this paper, the presence/absence of rockfall sources is considered as the objective variable; The operational improvements in (1) through (3) automate operation processes, and their effects go beyond mere efficiency gains to the realization of operation that could not be done before.

Key words

laser surveying, grid data, falling rock source, AI, DX