一种改进的代数神经网络电阻层析成像图像重建算法

Improved image reconstruction algorithm based on the neural network for Electrical resistance tomography

本文以16电极的ERT系统为背景,从图像重建的稳定性和速度两方面对两相流场的图像重建算法优化进行实验室研究。针对电阻层析成像系统存在的软场特性、强非线性和不适定性,重建的图像质量差、计算时间长等问题。采用了一种将基于类支集函数的代数神经网络算法,将图像重建转变为一个严格对角占优的线性方程组的求解问题,以达到图像快速、准确的重建目的,该算法的求解过程稳定并具有良好的计算性能。同时针对大规模神经网络算法训练速度较慢的问题提出了分区域求解的改进方法。通过实验仿真分析,改进后的算法具有简化神经网络结构,比大规模神经网络运算速度快,误差小等优点,为电阻层析成像系统图像重建提供了新的有效方法。

Aiming at improving the image reconstruction algorithm in 16-Electrical resistance tomography system,this thesis conducts an experimental study of image reconstruction algorithm on two phase flow,which mainly depend on the stability and the speed.For the soft-field characteristics in electrical resistance tomography system,strongly nonlinear and ill-posedness,the poor reconstructed image quality,long time computation problems.Using neural network algorithm based on the class-supported function, changing the image reconstruction to a strictly diagonally dominant linear equations problems in order to make the process fastly and accurately,keeping process stability of solving and the computational performance.At the same time,put forward an improved method of dividing processes to overcome the long time of training large-scale neural network.Through experimental simulation,the improved algorithm has the advantages of simplified neural network structure,faster than a large-scale neural network and the low error.Providing a new effective method for electrical resistance tomography system.