Research On Image Super-resolution Reconstruction Technology Based On Deep Learning

The development of science and technology has made people have higher requirements for the resolution of images.High-resolution images contain more information and have a better visual experience than low-resolution images.Therefore,improving image resolution through image super-resolution reconstruction technology is an important research topic in the field of computer vision.In recent years,the development of deep learning technology has greatly improved the performance of image super-resolution methods compared with traditional methods.However,many current super-resolution networks based on deep learning learn image features in a feed-forward manner,ignoring the feedback mechanism.The feedback mechanism can enable the network to learn a more complex mapping relationship between high-resolution and low-resolution images.Therefore,this paper conducts research on image super-resolution reconstruction based on the feedback mechanism in deep neural networks,which mainly includes the following:Firstly,aiming at the problems of the existing image super-resolution reconstruction methods that have weak ability to restore high-frequency details and insufficient feature utilization,a multi-scale feature fusion back-projection network is proposed for image super-resolution reconstruction.The network first uses multi-scale convolution kernels in the shallow feature extraction layer to extract feature information of different dimensions to enhance cross-channel information fusion;then builds a multi-scale back projection module to perform feature mapping through recursive learning to enhance the interdependence between the low-resolution images and high-resolution images,which improves the early reconstruction ability of the network;finally,the local residual feedback is combined with the global residual learning to promote the dissemination and utilization of features,thereby fusing feature information of different depths for image reconstruction.The experimental results of ×2 to ×8 super-resolution on the test images show that the quality of the SR images of this method is better than the existing image super-resolution methods in subjective perception and objective evaluation index,especially when dealing with large scaling factors(×8),the reconstruction performance is relatively better.Secondly,to solve the problem that many current depth models lack feedback mechanisms and treat feature channels equally,which hinders the expressive ability of super-resolution networks,an image super-resolution method based on attention mechanism feedback networks is proposed,using the hidden state in Recurrent Neural Network to make the network have a feedback mechanism.First,a multi-scale separable convolutional layer is designed to improve the utilization efficiency of parameters and extract more dimensional image features.Then the feedback module is designed to process the feedback connection between the sub-networks,and generate richer high-level representations through densely connected up and down projection units,providing more contextual information for reconstructing high-resolution images from input images;at the same time,the attention mechanism is introduced into the feedback module to adaptively allocates channel attention resources for image features,which enhances the interdependence between channels and promotes the network to mine richer high-frequency information.Finally,the final image is gradually generated through the iterative network,so that the network has a strong early image reconstruction ability.Experimental results show that,compared with many current advanced reconstruction methods,the super-resolution images generated by this method not only have higher objective evaluation indicators,but also have better visual effects,and still maintain excellent reconstruction performance at ×8 super-resolution.