CNN-POT: predicting oceanic turbulence phase variations with convolutional neural networks

Abstract Accurately predicting turbulence phase information from distorted beam intensity is crucial for investigating anti-turbulence interference in wireless optical communications. In this study, we investigated ocean turbulence and developed a predictor of the phase of ocean turbulence (POT) based on a convolutional neural network (CNN) (called the CNN-POT predictor). Furthermore, the influence of different temperature dissipation rates and radial indices on the POT prediction effect was analyzed. Meanwhile, the CNN is optimized by adding efficient attention mechanism and mixed loss function. When orders p = 3, the average test loss rate of the mixing test dataset with varying temperature dissipation rates was 0.0576. This demonstrated the reliability of the CNN-POT predictor predictions in various ocean turbulence environments. We innovatively demonstrated that high-order radial LG beams can improve the accuracy of the CNN-POT predictor in predicting POT. These results not only provide a new method for accurately predicting POT, but also provide a theoretical basis for the study of anti-turbulence interference in applications such as quantum information and wireless optical communications.