基于RBF神经网络的ECT图像重建

ECT image reconstruction based on RBF neural networks

线性反投影算法是最常用的ECT图像重建算法,该算法将极板电容测量值与成像区域介电常数间的非线性关系作线性化近似.由于神经网络的非线性映射能力可用来避免这种线性化近似,为此探讨了基于RBF神经网络的16极板ECT系统的图像重建方法.采用最大矩阵法确定RBF神经网络隐层神经元数目,用最小邻聚类方法确定径向基函数的宽度和中心,建立了极板电容测量值与成像区域介电常数间的RBF神经网络映射.仿真实验结果表明,基于RBF神经网络的ECT图像重建方法重建速度与线性反投影法相当,重建质量优于线性反投影法.

The algorithm most commonly used for ECT image reconstruction is the linear back-projection （LBP）, where the non-linear relationship between the permittivity distribution and capacitance measurements is usually approximated to the linear. Because the non-linear mapping ability of neural networks can avoid such linear approximation, the image reconstruction of 16-electrode ECT system based on RBF neural networks was discussed. RBF neural networks were trained to convert the electrode capacitance measurements to the permittivity distributions in image region. The number of hidden nodes in RBF neural networks was determined using maximal matrix element method, and the center and width of RBF function were determined using the nearest neighbor-clustering algorithm. Simulation results indicate that this image reconstruction algorithm can provide images superior to those obtained by the LBP algorithm within a similar reconstruction time.