Research On Mural Image Super-Resolution Reconstruction Based On Convolutional Neural Network

With the development of super-resolution reconstruction technology,super-resolution reconstruction of mural images has become a key research topic.In recent years,deep learning based on convolutional neural networks as a new image processing technology has been widely used in many fields such as pattern recognition,computer vision,target classification and detection.The use of convolutional neural networks to reconstruct images in super-resolution has become a very active topic in the field of image restoration.As the number of layers of the convolutional neural network model continues to deepen,its self-learning ability to reconstruct images continues to improve.However,the current image super-resolution reconstruction model based on convolutional neural network still has certain problems when it comes to super-resolution reconstruction of mural images with rich structure details and complex textures and colors.Therefore,based on the texture and structural characteristics of the mural image and the design ideas of the convolutional neural network,this paper studies the problem of super-resolution reconstruction of the mural image.1.Super-resolution reconstruction method of mural image based on multi-scale residual attention networkIn order to expand the width and depth of the network,this paper uses Inception Block in GoogleNet and Residual Block in ResNet in the feature extraction stage.First,through the multi-scale mapping unit,the convolution kernels of different scales are used to directly extract the features of the low-resolution mural image;then,the fused feature map is input to the residual channel attention block,and each of the convolution features is modeled.The functional relationship between the channels enables the network to optimize the weights of each feature map based on global information and enhance the depth mapping ability of the network model;finally,a sub-pixel convolution layer is introduced at the end of the network to rearrange the pixels to be reconstructed High resolution mural image.2.Super-resolution reconstruction method of mural image based on recursive residual attention networkIn order to avoid the parameter problems caused by the deepening of the network layers,this paper proposes a recursive residual attention network structure,which constructs a benchmark model by removing the local residual blocks of the BN layer and embedding them in the local residual blocks The channel attention mechanism enhances the network's expressive ability by automatically calibrating the weight of the feature channel,ensuring the rapid circulation and integrity of the mural image feature information.Then the introduced recursive structure leads to a bypass after every two local residual blocks to reduce the amount of parameters by sharing network parameters.Finally,the deconvolution layer is used to reconstruct the mural image under different magnifications.The two improved algorithms proposed in this paper are tested on the mural image experimental data set.The experimental results show that these two improved algorithms can reduce the reconstruction error and enhance the edge,texture and structure information of the reconstructed mural image,making the reconstructed The details of the mural image are more abundant.