摘要
电磁层析成像(electromagnetic tomography, EMT)传统算法由于物理模型的限制,导致重建数据缺失,使其逆问题存在着严重的不适定性和病态性。为解决重建图像普遍存在伪影多,质量差等问题,提出了一种基于MLP-AE的复合电磁层析成像算法。首先用待测物场信息作为输入送入自编码神经网络(AE)学习,得到编码数据;再对待测物场进行电磁激励获取电压数据;将电压数据作为输入,待测物场信息编码后的数据作为输出送入多层感知机神经网络(MLP)学习;最终解码实现端到端的图像重建。通过均方误差、结构相似性指数和成像时间评估所述算法的性能,并与线性反投影算法、Tikhonov正则化算法、Landweber迭代算法进行对比。实验结果表明,所述算法在单幅图像上:均方误差较以上传统算法分别降低了28.77%、22.57%、23.74%,结构相似性指数分别提升了17.54%、14.38%、15.44%,成像时间分别快了73.78%、98.63%、93.86%。所述重建算法对于待测物位置和形状的预测更为精确,时间大幅度减少,为之后进行实时精确成像提供了思路。
Due to the limitations of physical models,the traditional algorithm of electromagnetic tomography(EMT)leads to the lack of reconstruction data,which makes its inverse problem have serious discomfort and pathology.In order to solve the problems of many artifacts and poor quality in the reconstructed images,a composite electromagnetic tomography algorithm based on MLP-AE is proposed.Firstly,the field information of the object to be tested is sent to the self-coding neural network(AE)for learning as input to obtain the encoded data.Then the electromagnetic excitation of the measured object field is carried out to obtain voltage data.The voltage data is taken as input,and the data after encoding the field information of the DUT is sent to the MLP neural network for learning as output.Finally decoding enables end-to-end image reconstruction.The performance of the proposed algorithm is evaluated by mean squared error,structural similarity index and imaging time,and compared with the linear backprojection algorithm,Tikhonov regularization algorithm,and Landweber iterative algorithm.The experimental results show that the proposed algorithm reduces the mean squared error by 28.77%,22.57%and 23.74%compared with the above traditional algorithms on a single image,the structural similarity index is increased by 17.54%,14.38%and 15.44%,and the imaging time is 73.78%,98.63%and 93.86%faster,respectively.It provides an idea for real-time accurate imaging later.
作者
贾虎
王明泉
商奥雪
JIA Hu;WANG Mingquan;SHANG Aoxue(School of Information and Communication Engineering,North University of China,Taiyuan,Shanxi 030051,China)
出处
《计量学报》
CSCD
北大核心
2024年第8期1096-1102,共7页
Acta Metrologica Sinica
关键词
电磁计量
电磁层析成像
深度学习
图像重建
自编码
MLP
metrology
electromagnetism metrology
electromagnetic tomography imaging
deep learning
image reconstruction
self coding
MLP
