An intelligent method for rapid identification of large landslide events from continuous seismic recordings
Large landslide disasters generate seismic signals distinct from tectonic earthquake activity,and the rapid and accurate identification of such seismic signals from continuous seismic recordings can provide crucial support for rapid reporting or early warning of large-scale landslide disasters.However,achieving such identification poses a significant challenge.In this paper,we collected 1431 seismic signals generated by 150 landslides worldwide.Using basic data augmentation method,we augmented the dataset to 8351 seismic signals related to landslides.We trained a deep learning network,named the landslide recognition network(LRNet),based on the ResNet architecture.The LRNet was enhanced with skip and connection structures to expand network width and feature depth,thereby further enhancing its generalization capability.Using labelled seismic data associated with landslides,LRNet,AlexNet,VGGNet,and ResNet networks were compared and analyzed,with LRNet achieving a recognition accuracy of 98.14％.The LRNet's accurate identification of seismic signals related to landslides was further validated through continuous seismic recordings of the Sedongpu landslide disaster in Linzhi,Tibet,demonstrating reliable generalization performance.Additionally,Green's function inversion was employed to reconstruct the motion trajectory of the main event,providing reliable technical support for rapid reporting and rescue efforts of landslide disasters.
Landslide monitoringSeismic signals associated with landslidesDeep learningSeismic signalsGreen's function
万方数据
维普
