Optimization of Nadam algorithm for image denoising based on convolutional neural network

By studying the structure of feedforward denoising convolutional neural network (DnCNN), image denoising technology further integrates deep neural network structure, learning algorithm and regularization method, at the same time, in order to spee
¹
d up the network training process and improve the denoising of the model Performance, the paper imports the residual learning and batch normalization. This paper uses the Nadam optimization algorithm that is different from the common gradient descent algorithm SGD and Adam algorithm. The DnCNN-N model can solve the Gaussian denoising with a certain degree of noise. This paper analyzes the three noise levels: $\sigma =$15, 25 and 50. DnCNN-N uses residual learning methods to remove potentially clean images in the hidden layer. The experiments show that this model can show high efficiency in conventional image denoising tasks, and can be efficiently implemented through GPU computing.