Reinforcement Learning Based Efficient Underwater Image And Video Communication

Nowadays,people are exploring and utilizing the ocean resources more and more.As a key technology in the field of underwater detection and monitoring,the communication technology of underwater images and videos is getting more and more attention.The underwater acoustic channel,which is currently the main communication channel for underwater long-and medium-range communication,is severely constrained by its time-varying channel,limited bandwidth,and huge time delay for efficient communication of underwater data.These problems are particularly significant for information such as underwater images and videos,which have large data volumes.Therefore,this thesis combines the techniques of underwater adaptive technology and reinforcement learning to study and design systems for underwater images and videos respectively,and proposes two algorithms that can significantly improve the performance of underwater communication.The main research of this thesis is as follows.For underwater image communication:1.An efficient underwater image acoustic communication algorithm based on reinforcement learning is proposed,which can improve the image quality while reduce the energy consumption and time delay in fast time variant UWA channels.In the proposed algorithm,the received image quality and other communication performance parameters are estimated at the sink continuously and then feedback to the transmitter by an independent channel.The transmitter uses a reinforcement learning algorithm to process the obtained feedback information and then adaptively selects the most appropriate modulation coding method to achieve the efficient underwater image communication.Experiments were designed to compare with traditional underwater adaptive algorithms,and the reinforcement learning based algorithms can obtain huge performance gains after convergence.2.A watermarking-based image quality evaluation method is designed to obtain the images quality evaluation results during transmission.The method obtains a relatively accurate quality of image chunking without significant bandwidth loss and uses it as a parameter for reinforcement learning return calculation.In experiments compared with reinforcement learning algorithms that do not take image quality into account,the design that take image quality into account can substantially improve the quality of the received images.For underwater video communication:1.The different characteristics of video and image transmission in underwater are analysed.Based on the system for underwater images,an underwater video communication algorithm that adapts the video code rate and the modulation and coding of the transmitted data according to the state of the underwater channel is proposed.The algorithm implements the optimal decision of the modulation and coding method,video bitrate in a dynamic game without the need for a specific underwater channel model.We collected data from a large number of pool experiments for algorithm simulation and proved that the algorithm can effectively reduce the unframing failure rate of video communication,improve the video quality under bandwidth constraint and reduce the time and energy consumption of video transmission.2.Parameters describing the state of underwater video communication are defined,these parameters include frame quality,energy consumption ratio,delay ratio and frame loss rate.The H.265 video coding structure was studied and a segmentation scheme was designed to efficiently segment and transmit the H.265 video while maintaining the H.265 coding compression rate and resolving the feedback in sequence at the receiver.Each valid frame is evaluated for quality using the SSEQ non-reference quality evaluation method to obtain frame quality parameters.A reward function is defined in combination with other parameters,using deep reinforcement learning for adaptive design.In comparison with the Q learning algorithm,the deep reinforcement learning algorithm outperformed in terms of convergence speed and performance in the simulation experiments.