Research On Compression Sampling Data Recovery Technology Based On Deep Learning

In recent years,unmanned intelligent devices such as drones and robots have ushered in rapid development,and the real-time sharing of airborne sensor data with receiving stations has become an urgent need in the industry.However,due to the limitation of the Nyquist sampling rate and the communication channels are extremely limited.Therefore,a new technology is urgently needed to break through the classic Nyquist sampling theory,and compression sampling is in line with this technology.Compression sampling technology combines the sampling and compression processes into one,which greatly reduces the storage space of the acquired signal.Next,use deep learning to perform data recovery on the compressed sample data.Since the parameters of deep learning convolutional network models are often as high as millions or more,it is necessary to perform parameter compression on deep convolutional networks.This thesis first introduces the background of the subject,the current state of research on compression sampling,neural network compression at home and abroad,and related concepts,and then mainly studies the use of deep convolutional networks to recover compressed SAR original data and compress the parameters of deep convolutional networks.The problem.The main work of this article is briefly described as follows:1)A data recovery algorithm for matrix completion of Cascaded Convolutional AutoEncoder Network(CCAE+Net)based on deep learning is proposed.The algorithm is based on a convolutional self-encoder network.In order to make the data sparse,a soft threshold is introduced between the encoder and the decoder.The CCAE+ Net network framework adopts a cascade connection method to realize the image quality of the reconstructed SAR raw data for imaging processing is gradually improved.On this basis,in order to prevent deep neural networks from overfitting,ResNet and DenseNet methods are introduced on the network.The experimental results show that compared with the traditional matrix completion reconstruction algorithm,the algorithm proposed in this paper has a significant improvement in the data recovery effect,and the application of CCAE+ network to the recovery of water resources data validates the algorithm proposed in this chapter in other data Effective.2)The algorithm uses the proposed basic network as the basic network.The basic network adopts a cascade method to realize the image quality of the reconstructed SAR raw data for imaging processing.The parameters in the basic network are compressed using the joint compression model method,which further reduces the network parameters.The experimental results show that the algorithm proposed in this paper can not only reduce the parameters of the deep neural network,but also significantly improve the image generated by the SAR original data for imaging processing.