Research On Image Super-resolution Lightweight Algorithm Based On Deep Learning

With the rapid development of the diversified information era and the remarkable improvement of computer hardware level,people pay more and more attention to the quality of received images,and high-resolution images are favored due to their high quality.Image super-resolution reconstruction has received wide attention as an important means to improve image resolution.Deep learning techniques are widely used in the field of computer vision and image reconstruction by extracting features of low-resolution images in super-resolution reconstruction tasks to obtain a corresponding high-resolution image.For the existing deep learning-based image super-resolution algorithms on the reconstruction task have too many model parameters and more complex computation,which are not conducive to the application in realistic scenarios,two lightweight superresolution algorithms are proposed in this paper.(1)In response to the fact that most existing image super-resolution reconstruction algorithms extract more feature details by extending the depth and width of the convolutional neural network,which will lead to an increase in the computational complexity of the algorithm and an increase in the number of model parameters,this paper proposes a lightweight image super-resolution reconstruction algorithm with adaptive residual attention.The algorithm as a whole adopts a combination of global and local residual connectivity,and generates an attention feature map focusing on high-frequency location information by improving the coordinate attention network;then the improved adaptive residual attention information extraction module and the coordinate attention module are connected in dual branches in parallel,so that the output feature information contains more image details,which greatly reduces the number of parameters while ensuring the model performance.It is experimentally demonstrated that the algorithm has better reconstructed image quality on the public test set and higher objective evaluation index compared with other algorithms,while the number of model parameters is less.(2)To address the problem that existing image super-resolution reconstruction algorithms have insufficient processing of feature details,this paper proposes a lightweight image super-resolution reconstruction algorithm based on improved generative adversarial networks.The algorithm adopts the idea of generative adversarial network as a whole,and the improvement is mainly carried out in the generative network part.Firstly,the edge-oriented convolution module is used in the deep feature extraction module of the generative network to better extract the image edge information;secondly,in the upsampling reconstruction module,a combination of pixel attention mechanism and upsampling is used to better recover the image details;finally,in the inference stage,the edgeoriented convolution module is further integrated into the regular convolution using the idea of heavy parameters to achieve the goal of reducing the model parameters.purpose.It is experimentally demonstrated that the algorithm performs better in terms of feature details for the images reconstructed using generative adversarial networks,and the number of model parameters is greatly reduced by using the reparameterization technique.