Error-resilient coding by convolutional neural networks for underwater video transmission

Due to an increased number of transmission errors, the real-time transmission needs for seafloor videos are imposing severe challenges on underwater acoustic networks. In this work, we propose an error-resilient coding method based on convolutional neural networks and multiple descriptions to combat packet losses for underwater video transmission. By exploiting the inter-frame motion information, our convolutional neural networks propagate the regions of interest, providing extra protection for multiple description coding. To achieve a good tradeoff between coding efficiency and error resiliency, video sequences are split into two kinds of descriptions that are encoded under a bit-rate constraint condition. Simulation experiments with underwater video datasets are conducted to verify the effectiveness of our approach at different packet loss rates, compared to state-of-the-art video coding schemes. (C) 2021 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.