Low-dose CT multi-target image reconstruction method of unsupervised learning based on ADMM

The invention relates to an unsupervised learning low-dose CT multi-target image reconstruction method based on ADMM, and belongs to the field of image reconstruction. According to the method, an ADMM iteration format is mapped into a deep learning network through an expansion technology, and the network comprises a reconstruction layer, a noise reduction layer and a Lagrange multiplier updating layer. And noise reduction is carried out in a noise reduction layer by adopting a classic self-encoding-decoding residual convolutional neural network. And then, generating a large number of images of different doses as labels by using a cyclic generation network cyclGAN, and adding Gaussian noise and Poisson noise into the images of different doses as inputs to solve the problems of insufficient sample data and lack of labels. And then an iterative data expansion technology is adopted, and iterative training is performed on the convolutional neural network according to the expanded data, so that data bias caused by