Study on the noise reduction method for semiaerial transient electromagnetic data based on LSTM
With the increasing number of factors interfering with the electromagnetic environment,it is difficult to correct-ly process and analyze the characteristics of noise contained in semi-aerial transient electromagnetic data,and denoising is one of the critical means to improve the accuracy of transient electromagnetic signal processing and analysis.Compared to traditional denoising methods,which mostly rely on frequency characteristics,take a long time,and are challenging to determine compu-tational parameters,LSTM(Long Short-Term Memory)neural networks can handle nonlinear time series well while not relying on frequency characteristics.Therefore,this article applies LSTM to semi-autonautical transient electromagnetic data process-ing.Based on adding various types of noise to one-dimensional forward modeling data,it is divided into training and testing.The training uses unsupervised learning methods to capture signal characteristics and structural information.In the testing phase,supervised learning methods are used to optimize the network using labeled training data to improve the noise removal effect further.Finally,a model with good test results is applied to practical data,and the results show that this method can ef-fectively eliminate noise at the end of the channel and is a feasible method that can be used for noise removal in semi-aerial tran-sient electromagnetic data.
国家科技期刊平台
NSTL
万方数据
