Research On Image Super-Resolution Method Based On Deep Learning

Image super-resolution technology is to restore low-resolution images to high-resolution images with clear texture and bright edges through specific algorithms.The existing works still have some limitations,including insufficient use of characteristic information,difficult training and so on.With the development and growth of computer technology in recent years,deep learning algorithms have been applied to the field of image super-resolution reconstruction and have achieved good recovery results.In this thesis,based on the knowledge of deep learning,the image super-resolution reconstruction model is constructed,and its research focuses on the following points:(1)In this paper,a new image super-resolution reconstruction network model is proposed,which is developed based on deep learning to fully obtain the effective information of image features.In addition,the learning-efficiency of the deep model is enhanced by rational application of residual structure and jump link.(2)This paper proposes a multi-scale attention mechanism,which is based on the deep learning model.The mechanism obtains and fuses the spatial information of each channel through multi-convolutional structure,then establishing the dependence relationship between channels by obtaining the attention weight of each channel,and finally establishing the interactive information between global and local features by optimizing the weight value.Multi-scale attention mechanism is proposed to enhance the ability of image feature representation.(3)This paper proposes a feedback-supervision mechanism,aiming to solve the difficulty of deep learning model training.Specifically,the feedback-supervision strategy includes two closed-loop supervisory tasks,which reduces the difficulty of model training by using posterior-information.Through experiments on different types of data sets,the experimental results show that the proposed method has advantages over traditional algorithms.