Abstract: Aiming at  the problem that marine controlled-source electromagnetic (MCSEM) signals were prone to be interfered by various noises in exploration, which  affected the accuracy of later inversion and data processing, we proposed an attention mechanism-guided convolutional autoencoder marine controlled-source electromagnetic data denoising method. Firstly, based on  the  autoencoder, we constructed a noise suppression network based on convolutional autoencoder for marine controlled-source electromagnetic data. Secondly, we opimized it according to the characteristics of noise in the data, deepened the depth of the network, introduced attention mechanism to make the network pay more attention to the effective signal features in the data, enhanced the feature extraction ability, constructed the network model, and realized the noise suppression of marine controlled-source electromagnetic data. The experimental results show that this method has higher signal-to-noise ratio and lower mean square error than the db8 wavelet noise suppression method and the variational mode decomposition noise suppression method. Meanwhile, it can still retain the signal features and increase the interpretable range of offset distance in the measured data, which proves the effectiveness of this method in the noise suppression of marine controlled\|source electromagnetic data.

Key words: marine controlled-source electromagnetic method, deep learning, convolutional autoencoder, attention mechanism, noise suppression