基于深度信念网络的ECT图像重建算法

ECT Image Reconstruction Algorithm Based on Depth Belief Network

为提高图像重建质量,针对电容层析成像技术(ECT)中的电容数据复杂多样且与介电常数呈非线性关系的特点,提出一种基于深度信念网络(DBN)的重建算法,利用DBN的深层非线性网络结构来实现电容值与重建图像灰度值非线性关系。并对DBN进行了改进,将自适应步长(AS)引入到对比散度(CD)算法中,解决固定步长寻找全局最优困难的问题,改善图像质量;在微调阶段采用拟牛顿法加快收敛速度,减少训练时间。在COMSOL5.3软件上进行仿真试验,通过MATLAB2014a对图像进行重建。试验结果表明:DBN能够有效地重建图像,并且要优于传统算法;改进后的DBN训练时间缩短了5.51 s,图像误差低至0.0094,相关系数高达0.9973,是研究ECT图像重建的新方法和手段。

In order to improve the quality of image reconstruction,in view of the complexity and variety of capacitance data in electrical capacitance tomography(ECT)and the nonlinear relationship between capacitance data and dielectric constant,a deep belief network(DBN)was proposed.The reconstruction algorithm of DBN used the deep nonlinear network structure of DBN to realize the nonlinear relationship between the capacitance value and the gray value of the reconstructed image.The DBN was improved and the adaptive step size(AS)was introduced into the contrastive divergence(CD)algorithm to solve the problem of finding the global optimum with fixed step size and to improve the image quality.In the fine-tuning stage,quasi-Newton method was used to accelerate the convergence speed and reduce the training time.The simulation experiment was carried out on COMSOL 5.3 software,and the image was reconstructed by MATLAB 2014a.The experimental results showed that:DBN can effectively reconstruct the image,and is better than the traditional algorithm;the improved DBN training time is shortened by 5.51 s,the image error is as low as 0.0094,and the correlation coefficient is as high as 0.9973,which is a new method and means to study ECT image reconstruction.