Li Brocade Pattern Based On Deep Learning Research On Super-Resolution Reconstruction Technology

China is a unified multi-ethnic country with rich ethnic cultures.The Li ethnic group in Hainan is a member of China’s ethnic minorities.The art of Hainan Li brocade has been included in the UNESCO list of "urgently needed intangible cultural heritage",and Hainan Li brocade is an indispensable part of national culture.The Li brocade pattern is the most artistic feature of Li culture and a symbol of Hainan culture.The patterns of Hainan Li brocade contain distinct ethnic characteristics and rich ethnic history and culture.However,with the development of modern society,the traditional culture of Lijin has been impacted by modern culture,so it is urgent to inherit and protect the Lijin culture in Hainan.During the process of collecting Li brocade original pattern images,factors such as shooting tools and external environment can affect them,resulting in low resolution of the obtained Li brocade pattern.At present,with the development of multimedia technology and the rise of various image processing algorithms,more and more scholars are starting to combine Li brocade patterns with computer methods for research.Therefore,the demand for high-definition Li brocade patterns is becoming increasingly strong;In addition,considering the cost and limitations of changing hardware,it becomes more meaningful to use software tools to obtain high-resolution Li brocade patterns while the collection hardware remains unchanged,and apply the processed high-definition Li brocade patterns to the design of actual artworks,daily necessities,clothing design,decoration,etc.Image super-resolution reconstruction technology requires the use of relevant theories in the fields of computer vision and digital image processing,with low resolution images as the processing object.Through relevant processing techniques and algorithms,the reconstruction is completed to achieve the acquisition of high-resolution images.Because both the environment itself and the image acquisition system have certain limitations,which lead to blurring or low resolution of the image,that is,the collected image has quality issues such as low clarity or blurriness.The research in this article aims to provide solutions to existing problems in this area,in order to effectively solve such problems.With the development of multimedia technology and the rapid development of algorithms such as deep learning in image processing,scholars have an increasing demand for the resolution of Li brocade patterns.In practical applications,there is also a strong demand for high-resolution Li brocade patterns.Therefore,improving image quality has become an urgent problem to be solved.However,after consulting a large amount of data,it was found that there have been no relevant reports at home and abroad on the use of super-resolution reconstruction technology to study the reconstruction of Li brocade patterns.This article studies and designs a network model for super-resolution reconstruction of Li brocade patterns,and the research on super-resolution reconstruction of Li brocade patterns is an innovative work from scratch.The main research content of this article is divided into two parts:The first part is to design an SR network based on convolutional neural network structure to perform reconstruction work on Li Jin patterns.It is a top-down encoding and decoding network structure,and the encoding and decoding network structure used for image detection and segmentation is used for the super-resolution work of Li Jin patterns.In the process of encoding and decoding,in the face of Li brocade patterns with rich Semantic information and more texture details,multi-scale feature cross level fusion is proposed.High level Semantic information and low level feature information in Li brocade patterns are retained to the greatest extent on multi-scale feature maps,and they are fused to decode and reconstruct in a bottom-up manner,which not only retains multi-level information,but also reduces the amount of computation and improves the speed of the model.Establish a new parallel branch in the face of extracted multi-scale features to supplement the information loss caused by reduced channel count;Add a mechanism to stimulate and squeeze attention,squeeze feature maps,and stimulate the allocation of different attention weights to make the model pay more attention to important features.Finally,an asymmetric convolutional feature extraction layer was designed to enhance the model’s ability to extract image features without additional inference losses.The design of a CNN based Lijin SR reconstruction network was completed.The second part is to design a Li Jin SR network based on generative discrimination.Designed to address the issue of using CNN based SR networks to reconstruct SR images with less realistic texture details,the reconstructed Li Jin pattern is more in line with the visual perception of the human eye.Chapter 4 conducts Li Jin pattern reconstruction work on SR networks based on GAN network structure.Using SRGAN as the foundation network,RMGAN is designed for Li Jin pattern super-resolution reconstruction,followed by network generation.A residual multi head self attention generator is designed to increase network depth and reduce network parameters,enabling the generation network to pay attention to global features and generate better SR Li Jin images.In the discriminative network,a parameter free self attention discriminator was designed to improve the performance of the baseline model,accelerate model inference speed,and add energy weight allocation to important features without increasing network parameters.Finally,the discriminant function of the network is designed with focus frequency loss and Swin Generator Loss,which enables the SR network to have better reconstruction effects on the Li brocade pattern.The visual texture is closer to the real HR Li brocade pattern,which is more in line with the human visual effect,while also retaining good objective evaluation indicators.This article investigates the method of using Li brocade patterns in super-resolution reconstruction networks with different basic structures.Although SR networks with different structures focus on different results,they all enable the obtained Li brocade patterns to have higher resolution.This article studies and designs a super-resolution reconstruction algorithm for Li brocade patterns from the perspective of deep learning theory.Practice has shown that the effect is good.The results of this study will be helpful for future research in related fields and provide relevant support for research on the super-resolution of Li brocade patterns.On the other hand,this study also solves the problem of unclear patterns caused by hardware conditions,cost,and other limitations.