Unpairde Data Conversion And Its Application In ECG

In recent years,the field of artificial intelligence has been greatly developed thanks to the generation of massive data,the development of algorithms and the increase in computing speed,making many tasks attempt to be solved by deep learning methods.This thesis attempts to use the method of deep learning to solve the task of denoising ECG data.The purpose of the denoising ECG data is to convert the noisy ECG data to the noiseless ECG data.Artificial intelligence is good at supervised learning tasks.In order to learn the mapping relationship between source domain and target domain,supervised learning needs paired data that exists in the source domain and target domain.However,the difficulty of denoising ECG data is that,we cannot obtain the paired noisy ECG data and noiseless ECG data and can only obtain noiseless ECG data sets.So,the denoising ECG data was converted into how to use the noiseless ECG data sets to denoise ECG data in the absence of paired ECG data,it was actually the conversion of unpaired data.This thesis first introduces a method for learning the mapping relationship between the source domain X and the target domain Y in the absence of paired data,and then applies this method to the task of denoising ECG data.In this thesis,the research on unpaired data conversion is first analyzed theoretically,and then solve the problems raised in the theoretical analysis using deep learning,so as to design a complete solution to unpaired data conversion.The approach includes feature extraction module and data generation module.(1)The feature extraction module based on the residual network transforms the source domain X and the target domain Y into the same feature space Z,the two domains have similar features in the feature space Z.(2)The data generation module based on conditional generative adversarial network transforms feature space Z into target domain Y.The two module are combined to realize the conversion from the source domain X to the target domain Y in the absence of paired data.Furthermore,this thesis applies the method of unpaired data conversion to denoise ECG data,the models of the two modules were instantiated in the unpaired data conversion method,which introduce the network structure,objective function and optimization algorithm.According to the experimental results and experimental evaluation,it can be concluded that the method of unpaired data conversion can complete the task of denoising the ECG data and the experimental results are good.