Efficient reinforcement learning training method for video coding optimization

Abstract

The invention discloses an efficient reinforcement learning training method for video coding optimization, which is used for solving the problems of low convergence rate and unstable training of reinforcement learning in the training process of video coding optimization. According to the method, firstly, a good initial strategy is obtained by adopting an evolutionary algorithm, and then reinforcement learning network parameters are initialized by adopting a supervised learning method, so that the problem of slow convergence of a training initial stage caused by random initialization is reduced. In the reinforcement learning training process, a good strategy is stored, the strategy is randomly sampled at a certain probability, the problem that too many bad strategies are caused by blindnessof reinforcement learning in the exploration process is reduced, and the convergence speed and stability of training are improved. After a certain stage of reinforcement learning training, the systemmay fall into local optimum, thereby causing limited improvement of video coding compression efficiency. According to the method, small-amplitude random disturbance is carried out on specific parameters of the strategy network at certain intervals, the system exploration range is widened, and the compression efficiency of video coding is further improved.

Description

technical field [0001] The present invention relates to video coding and reinforcement learning, in particular to an efficient reinforcement learning training method for video coding optimization Background technique [0002] With the continuous development of multimedia digital video applications and the continuous improvement of people's demand for video cloud computing, the data volume of the original video source makes the existing transmission network bandwidth and storage resources unbearable. Therefore, the compression of video signal has become one of the hot spots of academic research and industrial application at home and abroad. Video compression, also known as video coding, aims to eliminate redundant information between video signals. So far, domestic and foreign standardization organizations have successively formulated a variety of different video coding standards. Since the H.261 video coding standard, the mainstream video coding standards have adopted the ...

AI Technical Summary

Problems solved by technology

The invention solves the problem that the traditional method cannot obtain the global optimal solution, and the machine learning method is difficult to generate labels for training

Aiming at the problem that traditional reinforcement learning converges very slowly when there are many network parameters, this invention proposes a pre-training method to speed up the convergence of the algorithm

Images