Research On Image Super-resolution Reconstruction Based On Multi-level Degradation And Attention Mechanism

Image super-resolution reconstruction has attracted a lot of researchers,which is an important branch of deep learning.Now,that the image super-resolution method is based on in-depth learning has made a breakthrough in the qualitative and quantitative image of the image.In order to improve the quality of image reconstruction,the existing image super-resolution methods based on in-depth learning generally enhance the quality of picture reconstruction by increasing the depth of the network model.However,experiments also show that with the increase of the depth of the network model,the image transition smoothing phenomenon will become more and more serious.At the same time,it is difficult to describe the real distribution of low resolution images in natural scenes by down sampling because of the influence of illumination,noise and ambiguity.Therefore,the article proposes an image super-resolution reconstruction algorithm based on multi-level degradation channel attention mechanism.Firstly,a spatial attention image super-resolution reconstruction network based on multi-level degradation is constructed.Because the natural low-resolution image is greatly affected by the weather,noise and weather and other factors,the distribution is more complex,so this paper uses multi-level degradation to simulate the generation of real low-resolution image,that is,adding the ambiguity and noise to the high-resolution image to generate a near real low-resolution image.At the same time,the spatial attention mechanism is used to construct the spatial attention module,and the lightweight spatial attention module is used to obtain the self similarity in the image.Secondly,the spatial attention mechanism only uses the spatial relationship to constrain feature extraction,and fails to express and utilize the perceptual information of different filters.Therefore,the channel attention mechanism is designed in the network model.Through the combination of spatial attention mechanism and channel attention mechanism,the effectiveness of the network model for the extraction and expression of key features in the image is improved.Finally,on the basis of the previous two experiments,in order to better learn high-frequency information,wavelet loss function is introduced to verify its role in channel attention mechanism and channel spatial attention mechanism.The model of this subject is configured and run in the environment of Python version 1.4.0.The data set used is div2 k,in which there are 800 training data sets and100 verification data sets respectively.The evaluation index mainly uses structural similarity and peak signal to noise ratio to evaluate the reconstructed image.Proved by experiment,the result indicates that the model in this paper has a significant visual improvement.Therefore,the experiment proves that the combination of multilevel degradation with spatial attention mechanism,channel attention mechanism and wavelet loss function can improve the visual reconstruction quality of images.