Research On Electrical Impedance Tomography Of Flexible Sensor Array Based On Deep Learning

Physical pressure measurement is very important in wearable electronics,robots and other fields.The flexible pressure sensor array can be used to monitor the physical pressure between the human body and the external environment,and can also be used as the skin of the flexible robot to sense the physical pressure between the machine and the external environment.Pressure measurement technology based on flexible sensor array has been widely concerned.At present,there are some limitations in pressure measurement:(1)flexible sensor and its connection are easy to be damaged: at present,researchers usually use dot matrix sensor and area array sensor for data acquisition,which will be directly squeezed and impacted by external force,resulting in damage to the internal structure of the sensor and fracture of the connection between sensors,resulting in the use time of sensor array is short;and point array sensor and area array sensor are both used The point pressure is used to replace the local surface pressure,which leads to the failure to obtain the complete information of pressure surface distribution.(2)Limitations of image reconstruction algorithm: most of the reconstruction algorithms used in electrical impedance tomography are traditional Newton iterative algorithms,which are sensitive to the initial value.For different pressure regions,the initial value needs to be adjusted many times,so the repeatability is poor.Moreover,the reconstruction accuracy of the algorithm for pressure image with noise is poor,and it can't remove noise in the algorithm reconstruction Finally,the approximate solution method of iterative algorithm leads to unclear boundary of imaging object.(3)The types of neural network imaging reconstruction area are single: Recently,researchers take neural network algorithm as the reconstruction algorithm of electrical impedance tomography technology,but most of the neural networks used are shallow network structure.The shallow neural network can't describe the effective boundary information in the pressure area,and can't image the pressure area with sharp shape(such as triangle).In this paper,the following works are carried out for these key problems:(1)two kinds of flexible sensor arrays,pressure-sensitive conductive foam and pressure-sensitive conductive fabric,are designed as data acquisition carriers,and their piezoresistance curves are analyzed to verify the resistance distribution uniformity of the flexible sensor array;a multi-channel data acquisition system is designed to describe the gating mechanism of data gating module and control module.The adjacent driving mode is used to collect the voltage data.(2)Combined with electrical impedance imaging technology,the Gauss Newton imaging algorithm based on Tikhonov regularization is realized.The preprocessing of the collected data and the measurement standard of the imaging results are explored.Finally,the voltage data is reconstructed into the pressure distribution image of the imaging object,and the mean square error(MSE)in the pressure image is 0.35 on average.The image correlation coefficient between the real image and the reconstructed image is achieved.The average ICC is 82.8%,the SSIM between real image and reconstructed image is 85%,and the pressure center and reconstructed image center are 0.43.The results show that the Gauss Newton imaging algorithm based on Tikhonov regularization is effective and accurate in the measurement of pressure distribution with noise.(3)In view of the shortcomings of traditional iterative algorithm and shallow neural network algorithm,this paper designs a U-shaped structure composed of full convolution layer.The mean square error(MSE)of the depth learning algorithm in pressure image is 0.236,the image correlation coefficient(ICC)between real image and reconstructed image is 85.8%,and the SSIM between real image and reconstructed image is 90%.The results show that the network structure is feasible and superior in the reconstruction algorithm of electrical impedance imaging technology.This algorithm enriches the shape of the pressure area,improves the accuracy of the pressure reconstruction image,reconstructs the sharp shape of the stress area,distinguishes the shape boundary of the image,and improves the diversity and accuracy of the pressure imaging area.In this paper,the end-to-end pressure distribution imaging based on flexible sensor array is realized by combining depth learning and electrical impedance imaging technology,which can effectively improve the physical pressure measurement mode and the accuracy of voltage distribution imaging,and provide a new idea for promoting the development of pressure measurement technology based on flexible sensor array.