Abstract
Microtremer survey has played important roles in regional disaster prevention. However, in the survey, because observation data inevitably contains nonstationary noise, time-series data blocks appropriate for analysis are detected manually. It is highly expected that automation of this manual process promotes high-densely and long-time observation which have been major concern currently. In this research, we developed a method for auto detection of analysis blocks in microtremer records. The problem is formulized as binary classification problem, and it was solved using deep learning. Multilayer perceptron and convolutional neural network were applied and they showed about 95% accuracy in maximum.
