Presentation Information
[2P0101-01-01]Integration of hyperspectral imaging and deep learning system for the evaluation of the degree of weathering of rocks
○Jaewon Kim1, Youhei Kawamura1, Sinaice Bino Brian1, Natsuo Okada1 (1. Akita University)
Keywords:
Deep learning,Weathering degree classification,Hyperspectral Image,CNN
The purpose of this study is the evaluation of the degree of weathering in rocks using hyperspectral imaging and a Convolutional Neural Network (CNN). Up until recently, geologists and petrologists evaluate the degree of weathering in rocks using relatively simple visual guides and, their own experience. In civil or mining engineering projects such as dams, tunnels, and mine sites, it is important to evaluate the degree of weathering because of its effect on the strength of the surrounding rock. Due to some circumstances or constraints, it might not be always possible to consult experts on judging the degree of weathering in rocks. Therefore, a new methodology using machine learning for evaluating the degree of weathering in rock could be developed as an alternative. In this study, deep learning is used in the form of CNN that learns using data from hyperspectral imaging from rocks in varying degrees of weathering. Hyperspectral imaging system (Specim, Spectral Imaging Ltd., Oulu, Finland) that is used in this study is included utilizes 204 bands (400nm-1000nm). The CNN algorithm is based on modified a VGG-16 architecture. The total amount of data amounts to 19,456 CSV files. 90% of the total data will be used as learning data and the remaining 10% will be used for testing. Preliminary results show that the system has an average accuracy of over 90%. The study, therefore, concludes that the combination of hyperspectral imaging and CNN is a feasible process for evaluation of the degree of weathering rock without the need for human expertise or bias.