基于多级残差注意力的X-ray图像超分辨率重建  

X-RAY IMAGE SUPER-RESOLUTION RECONSTRUCTION BASED ON MULTISTAGE RESIDUAL ATTENTION

作　　者：张建波 杨璐[1,2] 刘佳琦[1,2] 张祯[3] Zhang Jianbo;Yang Lu;Liu Jiaqi;Zhang Zhen(Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control,Tianjin University of Technology,Tianjin 300384,China;National Demonstration Center for Experimental Mechanical and Electrical Engineering Education(Tianjin University of Technology),Tianjin 300384,China;Intensive Care Unit,General Hospital of Tianjin Medical University,Tianjin 300052,China)

机构地区：[1]天津理工大学天津市先进机电系统设计与智能控制重点实验室,天津300384 [2]机电工程国家级实验教学示范中心(天津理工大学),天津300384 [3]天津医科大学总医院重症医学科,天津300052

出　　处：《计算机应用与软件》2023年第10期235-241,共7页Computer Applications and Software

基　　金：国家自然科学基金项目(61873188)。

摘　　要：针对肺部X-ray图像在超分辨率重建过程中出现的肺部边缘不清晰以及器官纹理模糊等问题,提出一种基于多级残差注意力的X-ray图像超分辨率重建方法。将注意力机制嵌入残差块中构建网络基本块,在加速网络收敛的同时,使网络更加关注图像的边缘纹理特征;设计多尺度特征融合模块进行特征提取,保证结构信息的完整性;通过多级残差学习加速网络训练,并允许构建更深层次的网络;融合上采样图像与Bicubic图像完成最终重建,弥补特征提取过程中的特征损失。实验结果表明,所提出模型的PSNR、SSIM均高于现有算法,且重建出的图像具备更加丰富的细节和清晰的边缘。In order to solve the problems such as unclear lung edges and fuzzy organ textures in the super-resolution reconstruction of lung X-ray images,a super-resolution reconstruction method based on multilevel residual attention is proposed.We embedded the attention mechanism into the residual block to construct the basic block of the network,which accelerated the convergence of the network and made the network pay more attention to the edge texture feature of the image.A multi-scale feature fusion module was designed to ensure the integrity of feature structure information.Through the multistage residual structure,the network sped up the training and allowed the construction of a deeper network.The up-sampling image and Bicubic image were fused to complete the final reconstruction to make up for the feature loss during feature extraction.The experimental results illustrate that the proposed model's PSNR and SSIM are higher than the existing algorithms.The reconstructed images have richer details and clearer edges.

关 键 词：X-RAY 通道注意力机制 特征融合 多级残差学习 超分辨率重建 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]