基于深度学习的地震与爆破事件自动识别研究

Automatic Recognition of Earthquake and Blasting Events Based on Deep Learning

针对天然地震事件、爆破事件分类问题,使用甘肃及周边地区80个天然地震事件和20个爆破事件建立数据集,采取深度学习卷积神经网络(convolutional neural network,CNN)方法搭建两个不同结构的模型进行训练,并用500条训练集之外的天然地震事件与爆破事件波形作为测试数据集,其训练和测试准确率均达到90%以上。结果表明,本文设计的两种模型均具有一定的泛化能力,尤其是Inception V1模型在天然地震事件与爆破事件分类识别中效果良好。

Aiming at the classification of natural earthquake events and blasting events,we use 80 natural earthquake events and 20 blasting events in Gansu and its surrounding areas to establish datasets,and apply deep learning convolutional neural network(CNN)method to build two models with different structures for training,and use 500 waveforms of natural earthquakes events and blasting events out of the training sets as test datasets.The accuracy of training and testing is more than 90%.The results show that two training models designed in this paper have a certain generalization ability;especially the Inception V1 model has good effect in the classification and recognition of natural earthquake events and blasting events.