一种新的卷积神经网络的ECT图像重建算法  

A New ECT Image Reconstruction Algorithm Based on Convolutional Neural Network

作　　者：李兰英[1] 孔银[1] 陈德运[1] 

机构地区：[1]哈尔滨理工大学计算机科学与技术学院,黑龙江哈尔滨150080

出　　处：《哈尔滨理工大学学报》2017年第4期28-33,共6页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金(60572153;60972127)

摘　　要：针对电容层析成像技术(electrical capacitance tomography,ECT)反问题中图像重建困难的问题,研究了将卷积神经网络(convolutional neural network,CNN)应用于ECT图像重建的可行性,在对卷积神经网络中较耗时的深层结构和训练过程问题进行深入研究的基础上,对结构中的卷积层和训练中的子采样方法进行了改进,提出了一种加速收敛卷积神经网络(fast convergent convolutional neural network,FCCNN)的图像重建方法,并通过Matlab在计算机上建立了ECT实验仿真系统,与传统算法的仿真实验结果进行了对比和分析。实验结果表明,改进后的算法对常见管道流型的图像重建效率和质量都有一定的提高。In response to the problem of image reconstruction in electrical capacitance tomography （ECT） technology, the feasibility of applying convolutional neural network （ called CNN） to ECT image reconstruction is studied. On the basis of in-depth research for convolution neural network for the more time-consuming process of deep structure and training issues, the convolution layer and the training of the structure of the sub sampling method is improved, and a fast convergence convolution neural network （called FCCNN） image reconstruction method is proposed. Finally, the ECT simulation system is built by Matlab on the computer. For each algorithm the simulation results were compared and analyzed. The experimental results show that the improved algorithm has definitely improved on the image reconstruction efficiency and quality of the common flow pattern.

关 键 词：电容层析成像技术 卷积神经网络 图像重建 子采样 MATLAB 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]