摘要
为了解决电容层析成像(ECT)图像重建中电容值与介电常数这一非线性病态问题,提出一种自适应加权的多特征融合(AWMF) ECT图像重建算法,利用网络模型拟合电容张量和介电常数的非线性映射关系。首先,在网络模型中采用密集卷积网络(DenseNet),不仅缓解梯度消失现象,还融合不同通道的特征信息;添加挤压激励网络(SENet)自适应调整特征通道的权重,用以提取不同通道的关键特征,提高重建图像的精度。其次,构建树形聚合结构(TASN)网络模块,扩大感受野并提取丰富的多尺度特征信息,消除普通卷积所带来的伪影现象。在COMSOL5.3软件上建模仿真后,通过MATLAB2014a对图像进行重建。实验结果表明,重建图像误差系数降低至0.025 6,相关系数提高至0.971 7,与传统算法和CNN算法相比,具有更高的图像重建质量。
In order to solve the nonlinear ill-conditioned problem of capacitance and permittivity in electrical cpacitance tomography(ECT) image reconstruction, An adaptive weighted multi-feature fusion(AWMF)ECT image reconstruction algorithm is proposed to realize the nonlinear mapping between capacitance value and dielectric constant is fitted by network model. Firstly, dense convolutional network(Densenet) is used in the network model, which not only alleviates the phenomenon of gradient disappearance, but also integrates the characteristic information of different channels. The weights of the feature channels are adjusted adaptively by squeeze excitation network(SENet) to extract the key features of the different channels to improve the accuracy of the image reconstruction. Secondly, the tree aggregation structure network(TASN) Network module is constructed to expand the receptive field and extract rich multi-scale characteristic information to eliminate artifacts brought by ordinary convolution. After modeling and simulation on COMSOL5.3, the image was reconstructed by MATLAB2014 a. Experimental results show that the reconstructed image error coefficient is reduced to 0.025 6, and the correlation coefficient is up to 0.974 7. Compared with the traditional algorithm and CNN algorithm, the reconstructed image has higher quality.
作者
马敏
高晓波
Ma Min;Gao Xiaobo(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China)
出处
《电子测量技术》
北大核心
2022年第21期130-135,共6页
Electronic Measurement Technology
基金
国家自然科学基金面上项目(61871379)资助。
