Prediction of reverberation intensity based on convolutional cyclic self-coding model

Abstract The formation of oceanic reverberation is a complex physical process characterized by significant complexity and randomness. Developing a generalized physical model that can accurately predicting oceanic reverberation under various environmental conditions is a challenging task. This paper proposes a deep learning method based on a convolutional recurrent autoencoder, which combines the spatial feature extraction capability of convolutional neural networks with the temporal sequence analysis capability of long short-term memory networks. This method not only enables the reduction of physical data dimensionality for learning but also demonstrates the model’s ability to learn reverberation distributions in two-dimensional ocean environments that depends on depth. The research results indicate that the model can still achieve high-precision predictions of reverberation effects at different source depths, despite the presence of random phase and amplitude noise in the reverberation data. The predictive accuracy of the model was quantitatively evaluated using the mean square error (MSE) and the structural similarity index (SSIM), verifying its potential for real-time reverberation prediction in actual oceanic environments.