The Research Of Image Restoration Method Based On Neural Network

In the process of optical imaging of the target,it will be affected by factors such as camera shake,object motion and atmospheric turbulence,resulting in blurred images.At present,the field of image restoration is relatively mature in traditional algorithms,but traditional algorithms require a lot of iterative calculations,which will lead to slower problem solving,so it is very important to explore more efficient methods.With the development of computer vision,deep learning has been used more and more in the field of image processing,but in some specific fields,such as solving the problem of blurring caused by atmospheric turbulence effects,there is a lack of data sets required by the network,so research is better Earth image restoration methods are of great significance.The main research contents of this paper are reflected in the following aspects:This paper analyzes the model-based optimization method and the image restoration method based on discriminant learning,and uses the half-quadratic splitting method to combine the advantages of the two,and proposes a deep learning-based convolutional neural network.In the model-based optimization method,the point spread function of the image is accurately estimated,and this method is used to solve the motion blur in the image restoration problem.In the verification of the experiment,different representative experiments are selected for comparison,and it can be seen from the vision that the method in this paper can effectively eliminate the artifacts of blurred images.In the restoration of astronomical images,the blurring caused by atmospheric turbulence is complicated.This paper proposes to solve such problems based on generative adversarial neural network.Since the network needs a lot of data for training,this paper proposes a telescope optical imaging system model,simulates a set of astronomical image data sets,and introduces an attention mechanism into the network model.This in turn enhances the resilience of the network.The experimental results are verified from different aspects.The model proposed in this paper can effectively remove the blurred astronomical image caused by atmospheric turbulence,and can remove the noise phenomenon.After restoring the image with lower resolution,it can reach the diffraction limit,and can effectively restore the actual astronomical images.In the restoration process of astronomical aberration image restoration,this paper constructs the optical imaging aberration model,and then constructs the astronomical aberration image dataset.This paper proposes a generative adversarial neural network with supervised attention mechanism,which can improve the learning ability of the network.The effectiveness of the method proposed in this paper is verified by the simulation experiments and the restoration results of actual astronomical images,and the restoration results can also remove the noise phenomenon.This work has good application prospects.