Research On Dynamic Convolution And Adaptive Learning Rate Based Image Super Resolution Reconstruction

The image super-resolution reconstruction technique aims to recover highresolution images from low-resolution images and provide more detailed information.The details of the information in the image not only enhance the visual effect of the image,but also help people to identify,analyze and process the image.Convolution neural network is a feedforward neural network,which has excellent performance for image processing.Convolution neural network utilizes the advantages of its own network structure to reduce the complexity of the network model and reduce the number of network weights,and the image can be directly used as the input of the network,avoiding the complicated feature extraction and data reconstruction process.The convolution neural network has been shown to be able to learn the mapping of lowresolution images to high-resolution images,and to achieve an improved end-to-end learning framework.Therefore,based on the advantages of convolution neural network,there are two contributions as following:(1)Dynamic convolution layer based Image super-resolution reconstruction algorithm: In the convolution neural network,it is very important to choice of network structure and the selection of parameters are very important.Based on the convolution neural network,In order to maximize the use of the prior knowledge of the image,the use of convolution neural network weight sharing network structure and back propagation algorithm,in the convolution layer is not artificially set the filter parameters,but through a convolution network Training the image learning to get the filter,the learning filter will be added to the image reconstruction network,improve the convolution network structure,and further improve the quality of image reconstruction.(2)Adaptive learning rate based Image super-resolution reconstruction algorithm: The convergence rate in the convolutional neural networks based image reconstruction algorithm is very slow.Based on the study of traditional gradient optimization and learning rate algorithm,an adaptive learning rate optimization algorithm is proposed.By setting parameters of the adaptive algorithm,the learning rate could be set in the training process through the loss function and the self-adjust.It can reach an optimal value.