基于强化学习的能量采集物联网视频编码技术

Reinforcement-learning-based video coding for energy harvested IoT devices

物联网设备的视频压缩码率和量化参数等编码方案决定着视频质量和物联网设备的使用寿命.针对未知视频传输模型和能量采集模型的物联网设备,提出一种基于强化学习的视频编码方案,能够动态优化视频编码码率和量化参数.该技术根据物联网设备的无线信道带宽、电池水平和采集的能量,结合反馈的视频质量和时延等信息,采用强化学习算法优化选择视频编码码率和量化参数.在动态的网络环境下,物联网设备不需预知视频传输模型就可以综合优化视频质量和设备能量损耗.仿真结果表明,该方案可以提高视频质量,降低设备能量损耗和时延,改善物联网设备效益.

Video coding technology can improve the video quality and save the energy consumption of Internet of Things(IoT)devices,in which the optimal video coding policy depends on the coding parameter selection.In this paper,we propose a reinforcement-learning-based video coding scheme for IoT devices with energy harvesting without knowing the video transmission model.The IoT device selects the encoding bit rate and quantization parameter according to the measured wireless channel bandwidth,battery level of IoT device,harvested energy,transmission distance and the previous video quality and delay.This scheme can achieve a trade-off between the video quality and energy consumption via trials and errors in the dynamic IoT network.Simulation results show that the proposed scheme can improve the video quality and the utility of IoT devices as well as reduce the energy consumption and delay.