The invention belongs to the field of
nuclear magnetic resonance data processing, and particularly relates to a deep neural network magnetic
resonance signal denoising method based on discrete cosinetransform. The method includes the following steps of: firstly, adopting the
discrete cosine transform to transform a noisy
signal and a simulated
signal as the input and ideal output of a deep neuralnetwork; and then, performing layer-wise greedy pre-training on the deep neural network by adopting an
unsupervised learning mode to realize network weight initialization; finely adjusting global parameters by using an error back propagation method; and finally, inputting a
test set to the trained deep neural network, performing inverse normalization on the
network output, and then performing inverse
discrete cosine transform to obtain the denoised
nuclear magnetic resonance signal. According to the method, the nonlinear mapping from the noisy signal to a clean signal can be realized, and one-time
elimination of all types of
noise in the
nuclear magnetic resonance signal can be realized; various complex and variable detection environments and
noise interference can be adapted, the signal-to-
noise ratio can be significantly improved, and the accuracy of the subsequent inversion and interpretation extraction parameters can be improved; and moreover, the introduction of
discrete cosine transform and
restricted Boltzmann machine pre-training greatly shortens the
training time of the deep neural network, improves the training efficiency of the deep neural network, and makes a method for eliminating nuclear magnetic
resonance noise in the deep neural network be practical.