Research On The Key Technology Of Image Reconstruction Based On Brain Signal

Reconstruction of visual stimulus images from brain signals expands the application field of brain science and accelerates the exploration of the working mechanism of the brain.In this paper,the key technology of image reconstruction based on brain signals is taken as the research object,aiming at the improvement of functional magnetic resonance imaging(fMRI)reconstruction image model,the design of image reconstruction scheme based on electroencephalography(EEG)signals,and the introduction of super-resolution image reconstruction technology to carry out secondary image reconstruction.The details are as follows:(1)An improved multiview depth generative model is proposed to address insufficient attention to the neural network structure and tuning strategy based on the functional MRI model.Firstly,the advantages and disadvantages of the multi view depth generative model are outlined;secondly,coders and decoders with different network structures are designed to address the shortcomings of the multiview depth generative model;finally,the different degrees of reconstructed image quality under different tuning strategies are compared.The experimental results showed that the improved model improved the average PCC and SSIM values by up to 3.61%and 3.10%,respectively,compared with the original model;the average MSE value decreased by-13.51%.(2)To solve the problems of the complex process and high cost of the reconstruction experiment based on fMRI,an image reconstruction scheme based on EEG is proposed.Firstly,the idea of scheme design is expounded.Secondly,the original EEG data are collected and preprocessed by traditional pretreatment methods such as filtering and principal component analysis and statistical methods such as outlier check,outlier processing,data distribution check,and correlation measurement.Finally,the image reconstruction model based on brain signals is used to reconstruct the processed data,and the image reconstruction experiment based on EEG is carried out.Experimental results demonstrate the effectiveness of the image reconstruction scheme.(3)An enhanced deep plug-and-play super-resolution(DPSR)algorithm based on the residual channel attention network is designed to reconstruct brain signal-based images with blurred and low resolution.Firstly,the advantages and disadvantages of the DPSR algorithm are described;secondly,the DPSR is improved with the residual channel attention network;finally,the performance of the improved model is tested on four datasets.The experimental results show that the enhanced algorithm’s average PSNR and SSIM values are improved by 0.31dB and 0.55%under different blur kernels.The above two values are improved by 0.26 dB and 0.51%under different noise levels.In this paper,based on fMRI,EEG,and super-resolution image enhancement technologies,the quality of reconstructed images based on EEG signals is further improved by designing image reconstruction schemes,self-building data sets,and improving image reconstruction models.In general,the research results of this paper can promote the early completion of the brain visual information decoding task.