SEGJ Technical Conference

Application of automatic firstbreak-picking using fully Convolution Neural Networks for evaluating geological conditions ahead of the tunnel face

Abstract We have developed SSRT (Shallow Seismic Reflection survey for Tunnels) so far to evaluate geological conditions ahead of a tunnel face. By using tunnel excavation blast as a seismic source, we can avoid to interrupt tunnel construction work. This method only requires a single geophone recording system after all by introducing seismic interferometry. Additionally, we have introduced the automatic first-break picking method using fully convolutional neural networks (CNN). It has a lot of advantages over the other picking methods, because it works properly, effectively and stably even for data with low signal-to-noise ratios. We apply our method to a new tunnel under construction in Japan.