Adaptive And Efficient Image/Video Transmission Based On Machine Learning

With the rapid development of mobile communication technology and the contin-uous upgrading of intelligent terminal hardware,the application of image and video has become an indispensable part of daily production and life.It is predicted that by 2021,the mobile image and video services will account for 78%of all mobile data services.With the explosive growth of image and video information and the increasing user de-mand in diversified scenarios,more and more service providers transmit image and video information based on wired and wireless heterogeneous networks.In this way,after the signal is sent to the sender,it is transmitted through the wired core network and the wireless edge network,and finally reaches the receiver.Diversified deployment and more flexible business model make the heterogeneous network model widely used in the existing transmission system.Due to the dramatic increase in data volume of image and video services and the in-creasingly diversified business scenarios,the limitations of wireless network resources and computing power are becoming more and more obvious.Traditional image and video transmission schemes are facing great challenges.On the one hand,the environ-ment of end-to-end transmission is constantly changing,and is vulnerable to external interference.On the other hand,wireless transmission is a very important part of image and video transmission,and the explosive growth of image and video data is making wireless bandwidth resources increasingly scarce.Therefore,it is of great significance to study image and video transmission technology with better adaptive performance and higher efficient.Online video playback is an important scene in image and video services,and adaptive bitrate algorithm has been proved to be an effective end-to-end performance improvement means.This method chooses the rate of the next video block of online video,so as to improve the overall transmission performance of the system.Most of the existing rate adaptive transmission is based on the prediction of the rate or the pre-supposition of the buffer occupancy.However,the prediction effect of these methods is generally conservative,and can not adapt to the increasingly complex and diverse environment.The first research of this paper is to propose an adaptive bitrate algorithm with better performance.In recent years,machine learning methods have been proved to be efficient and low-complexity in many fields.Therefore,based on enhancing the prediction ability of learning accurately,this paper carries out online learning process for various state variables of the client network environment,so as to achieve better video experience performance.At the same time,in order to meet the real-time require-ment in online video scene,the scheme can not only reduce the influence of external input on the algorithm itself,but also reduce the complexity of the algorithm itself by designing network adaptation algorithm and simple neural network structure based on multiplexing convolution filter,thus jointly accelerating the convergence process of the neural network to meet the end-to-end playback of online video demand.On the other hand,the increasing image and video data makes the wireless band-width resources more and more scarce.In recent years,efficient image transmission based on compressed sensing has been widely studied.The method directly sampled the sparse signal at the transmitter,and then reconstructed the signal perfectly through the reconstruction algorithm deployed at the receiver after transmission through the chan-nel.Although the existing methods have better transmission performance,there are many problems such as high complexity and easy to receive wireless channel interfer-ence.In order to solve this problem,the second part of this paper proposes a compres-sive sensing image transmission scheme based on deep learning,which can alleviate the shortage of wireless bandwidth resources and achieve efficient image transmission in wireless channels.The method partially sampled the image signal at the base station of the transmitter,and then sent the sampled signal to the wireless transmission channel.When it arrived at the receiver,it was reconstructed by deploying the reconstruction algorithm at the receiver.Compared with the existing methods,a two-stage algorithm is deployed at the receiving end.Firstly,the signal is transformed from transform do-main to pixel domain by Fast algorithm,and the initial coarse-grained reconstruction is carried out Then,the convolution neural network is used to further improve the quality of the signal.This method effectively reduces the complexity of reconstruction algo-rithm,and can be applied to any sampling rate,and can input any size of images.At the same time,this method trains a large number of image signals with different channel SNR,which can effectively resist the influence of channel noise on the reconstruction of wireless channel.