Compressive Sampling,Reconstruction And Prediction Of Flexible Pressure Array Information Based On Deep Learning

With the development of information technology,more and more data are required to be sampled and transmitted in networks and industrial processes.In the process of sampling,transmitting and storing large amounts of data,information systems are increasingly loaded.Compressed sensing theory can use data far below the Nyquist sampling frequency to acquire data and accurately reconstruct the data,greatly reducing the burden on the information system to process large amounts of data.Deep learning forms a more abstract high-level representation attribute through lowlevel features to discover the characteristic representation of data.Its purpose is to simulate the ability of the human brain to analyze learning,and to construct a neural network structure of brain-like biological mechanisms that mimic the biological mechanisms of the human brain to process various types of data.The human brain is an extremely complex biological tissue made up of hundreds different types and several hundred billion nerve cells,including many biological mechanisms such as sensing ability,memory and nerve attention.Therefore,using the neural structure,biological mechanism and information processing mechanism of the human brain for reference,designing and improving the existing deep neural network model can increase the ability of the information system to sense and process large amounts of data.Based on the theory of intellisense and deep learning,this research presents a method of compressive sampling,data reconstruction and predictive data for flexible pressure arrays.The proposed method is verified by human body tiny pressure information processing.The main contents are as follows:(1)In order to improve the efficiency of flexible array body tiny pressure data collection based on compressed sensing theory,a compressed sensing model based on sparse autoencoder is established.The sparse representation and measurement vector in compressed sensing theory are studied,and the idea of sparse autoencoder model in deep learning is integrated into compressed sensing theory,which provides a bridge between deep learning and compressed sensing theory.Using the proposed model,the process of compressive sampling in compressed sensing theory is improved,and the error between the reconstructed data and the original data is calculated.The proposed model can continuously adjust the sparsity value and the length of the measurement vector according to the set reconstruction error limit,so that the output reconstruction data satisfies the error requirement.The effectiveness of the proposed method is demonstrated by the human tiny pressure data sampling experiment of the flexible pressure array.(2)For the problem that the sampling length is small and the reconstruction accuracy is not high,a new neural network model Compressed Sensing Network(ComsensNet)is proposed.First,the compressive sampling process in compressed sensing theory is modeled as a neural network,and a stacked long short-term memory network based on deep learning is proposed as a compressed sensing reconstruction algorithm.Furthermore,according to the idea of sparse autoencoder in deep learning,the proposed neural network for compressive sampling process and the stacked long short-term memory network reconstruction algorithm are integrated into a new neural network model-Compressed Sensing Network.Finally,the human body model tiny pressure information acquisition experiment of the flexible pressure array verifies the superiority of the proposed method.(3)Aiming at the problem of low precision in reconstruction of densely sampled tiny information,a compressed sensing reconstruction algorithm based on human brain memory mechanism and neural attention mechanism is proposed.Firstly,according to the idea of human brain memory mechanism,a multilayer long short-term memory neural network model is proposed.Further,the neural attention mechanism of the human brain is combined with the proposed multilayer long short-term memory network model.Furthermore,a new compressed sensing reconstruction algorithm is designed by combining the two biological mechanisms of human brain.Finally,the validity of the method is verified by the reconstruction of pressure distribution in the human body model.(4)Aiming at the problem that the pressure information of different parts with the human body needs to be measured and sampled several times,a deep gated recurrent unit neural network structure is proposed.The pressure information of a certain part of the human body is used to predict the pressure information distribution of the key points in the upper body part of the human body.The correlation of pressure information between different parts of the human body is obtained through deep gated recurrent unit neural network learning,so as to improve the efficiency of sampling and measuring the human body pressure information and shorten the measurement time.Predictive experiments on tiny pressure information in different parts of the human body have verified the effectiveness of the proposed method.At the end,the content of this research is summarized and prospected.The deficiencies in the research are pointed out,and the follow-up research content and related work are prospected.