Research On Eit Imaging Algorithm Based On Deep Learning

Electrical impedance tomography(EIT)collects electrical signal data by placing measurement electrodes on the boundary of the measured object field,thereby it can reconstruct the electrical conductivity distribution of the target object in the sensing area.EIT technology has the advantages of non-invasiveness,high security,good real-time performance,and simple equipment.It has widely potential applications in the fields of geological prospecting,biomedical imaging,and human-computer interaction.Due to inherent problems of EIT technology,the imaging quality and reconstruction speed are not satisfactory for most practical applications.In order to improve the imaging quality and increase the imaging reconstruction speed,this thesis proposes a method based on deep learning scheme for the EIT imaging algorithm combining with simulation and water tank imaging reconstruction experiments to demonstrate its efficiency.The main research contents of this thesis are as follows:1.The development status and problems in the fields of biomedicine,robotic electronic skin,and gesture recognition are introduced.The feasibility of introducing EIT technology into the above-mentioned fields is analyzed.The current research status of EIT technology at home and abroad are briefly described and some valuable improvements in imaging algorithms are introduced.Some problems of EIT technology itself in detail are analyzed,the research goals and main content of this article based on this are put forward.2.The theoretical basis for the design of the EIT imaging system is explained from the three perspectives of the detection principle of EIT technology,the forward problem and the inverse problem.Then the simulation and physical imaging experimental platforms are built respectively.3.The basic principles of deep learning are introduced and the inverse problem EIT imaging network EITDNN is built.The dataset acquisition,training process and experimental results of the inverse problem simulation experiment are elaborated in detail.The experimental results show that the EITDNN built in this paper can solve some problems of traditional imaging algorithms,such as unclear contours and the inability to image sharp and complex shape targets.It can effectively distinguish target objects of different shapes,but the imaging shadow is still existing.4.The basic knowledge of image semantic segmentation algorithms are introduced.The EIT post-processing semantic segmentation network is built.The post-processing simulation experiment and water tank imaging experiment are elaborated in detail.In the water tank imaging experiment,the post-processing of EIT imaging based on the PSPNet semantic segmentation model of Mobile Net V2+ network has been used as an example to elaborate and analyze,and the size error(SE)and position error(PE)have been used as evaluation indicators.The experimental results of water tank imaging show that the PSPNet semantic segmentation model based on Mobile Net V2+ network built in this paper can remove imaging artifacts,thereby it can significantly improve imaging quality.Compared with the traditional imaging algorithm NOSER,the SE index pixel value is reduced by an average of 290.17,and the area of the conductive plate is reduced by an average of 9.02%,and the PE index is reduced by an average of 1.59%.