基于DK-SVD的深度学习电阻抗块稀疏成像方法研究  

Study on the Electrical Impedance Block Sparse Imaging Method of Deep Learning Based on DK-SVD

作　　者：王琦[1] 杨雨晗 李秀艳[1] 段晓杰[1] 汪剑鸣[2] 孙玉宽 冯慧 WANG Qi;YANG Yuhan;LI Xiuyan;DUAN Xiaojie;WANG Jianming;SUN Yukuan;FENG Hui(Electronic and Information Engineering College,Tianjin Polytechnic University,Tianjin 300387,China;Computer Science College,Tianjin Polytechnic University,Tianjin 300387,China;Jiangning Hospital Affiliated to Nanjing Medical University,Nanjing,Jiangsu 211100,China)

机构地区：[1]天津工业大学电子与信息工程学院,天津300387 [2]天津工业大学计算机学院,天津300387 [3]南京医科大学附属江宁医院,江苏南京211100

出　　处：《计量学报》2024年第9期1370-1377,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(62072335,62071328)。

摘　　要：针对电阻抗层析成像逆问题的病态性和非线性,提出一种基于DK-SVD的电阻抗块稀疏图像重建方法。该算法通过多层感知器为每组测量数据提供最优的模型参数,以适应数据集的多样性,进一步提高成像质量,并在稀疏编码阶段采用迭代收缩阈值算法加快收敛速度。仿真实验结果表明DK-SVD算法重建图像的结构相似性可达到0.95以上,误差可控制在0.1左右,平均重建速度为0.034 s,有效地提高了电阻抗层析成像的质量和效率,且经进一步实验证明了该算法具有良好的噪声鲁棒性和实际应用价值。Aiming at the ill-posedness and nonlinearity of electrical impedance tomography inverse problem,a DK-SVD-based block sparse image reconstruction method is proposed.The multi-layer perceptron is introduced to finetune optimal model parameters for measurement data considering the complexity of datasets and improve the image quality.The iterative shrinkage threshold algorithm is used to accelerate convergence in the sparse coding stage.The simulation results show that the structural similarity of the reconstructed image by DK-SVD algorithm can reach more than 0.95,the error can be controlled at about 0.1,and the average reconstruction speed is 0.034 s,which effectively improves the quality and efficiency of electrical impedance tomography,and further experiments prove that the algorithm has good noise robustness and practical application value.

关 键 词：电学计量 电阻抗层析成像 块稀疏 DK-SVD 图像重建 深度学习 

分 类 号：TB973[一般工业技术—计量学] TB96[机械工程—测试计量技术及仪器]