Single Image Super-resolution Reconstruction Based On Convolutional Neural Network

With the in-depth popularization of informationzation and intelligence,super resolution technology is becoming more and more mature and has begun to play an increasingly important role in daily life.The research has broad application prospects in the fields of medical imaging,remote sensing imaging,and public security.In recent years,with the wide application of deep learning,more and more super-division algorithms use convolutional neural networks to achieve reconstruction tasks.The existing advanced super-resolution methods with deepening or widening network demand high computational resources and memory consumption.It is difficult to directly apply them in practice.Secondly,it takes a long time for the algorithm to reconstruct image,and the reconstruction effect does not accord with human visual perception visual perception.Based on the above problems,this paper proposes two deep learning algorithms for single-image super-resolution reconstruction.This paper proposes a lightweight super-resolution reconstruction algorithm based on convolutional neural network for single image super resolution.The model consists of a 29-layer deep channel,the main learns the high-frequency texture information of the image.And the structure combines the dense block and residual connection.The dense block increases the data flow of the network and reduces the number of parameters,while the residual connection speeds up the convergence of the network.The network is evaluated on four different public datasets,the calculation amount and parameter quantity of the model,and Show the visual effect,outperforming the same period algorithm in accuracy and visual effect.Secondly,this paper proposes a fast and lightweight two-channel end-to-end superresolution reconstruction algorithm(FLSR)with fewer parameters and low computational complexity.The structure consists of a 3-layer shallow channel and a 29-layer deep channel,and a convolutional layer is used at the end of the structure to fuse the shallow and deep channel outputs.The shallow channel mainly restores the overall contour of the image,retaining the original information of the image;while the deep channel mainly learns the high-frequency texture information.Moreover,the algorithm is optimized and an enhanced network(FLSR-G)proposed using group convolution which significantly reduces the parameters and computational complexity with slight performance loss.Extensive quantitative and qualitative evaluations on benchmark datasets show that the proposed algorithm outperforms the current representative methods in terms of run-time and visual effect,with much fewer parameters and operations.