Waveform recognition and process interpretation of microseismic monitoring based on an improved LeNet5 convolutional neural network

The development of high-precision and interpretable automatic waveform classification algorithms with strong adaptability is becoming increasingly significant under the background of the big data era of microseismicity. Considering the deficiency of the existing network in waveform recognition and classification, an improved model which is suitable for microseismic (MS) monitoring waveform recognition was proposed in this study based on the LeNet framework. The improved model was applied to investigate thirteen kinds of MS monitoring signals that appear within 8 months of the Hanjiang-to-Weihe River Diversion Project. The results show that the accuracy of the best framework in the improved model is 0.98, which is 0.1 higher than original model. The average precision, recall and F 1 values of all improved models increased by 0.11, 0.12 and 0.12, respectively. Meanwhile, the improved model can visualize the entire waveform recognition process. A novel observation is that in some signal categories, the improved model mainly classified by focusing on the background information instead of the waveforms. It provides a reference for the intelligent classification of signals in MS monitoring engineering.