基于极端学习机的正则化电容层析图像重建算法

Image Reconstruction Algorithm Based on Extreme Learning Machine for Electrical Capacitance Tomography

针对传统ECT对于复杂情况下成像精度不高的问题,提出一种基于深度学习的反演方法。通过对传统极端学习机的改进和优化,采用重建图像方法获得的图像特征信息作为训练数据,并将数据输入预测模型得到的结果作为先验信息。通过成本函数封装先验信息和领域的专业知识,并引入空间正则器和时间正则器以增强稀疏性,利用分离的Bregman(SB)算法和迭代收缩阈值(FIST)方法求解规定的成本函数,以获得最终的成像结果。仿真实验结果表明,该方法重建的图像与原流型相比,误差小于10%,并且减少了伪影和变形,提高了重建图像质量。

Aiming at the problem that the traditional ECT is not accurate in complex situations, this paper proposes a depth learning based inversion method. Through the improvement and optimization of the traditional extreme learning machine, the image feature information obtained by the reconstructed image method is used as the training data, and the result obtained by inputting the data into the predictive model is used as the prior information. The cost function is used to encapsulate the prior knowledge and domain expertise, and spatial regularizers and time regularizers are introduced to enhance sparsity. The separated Bregman(SB) algorithm and the iterative shrinkage threshold(FIST) method are used to solve the specified cost function. The final imaging result is obtained. The simulation results show that the image reconstructed by this method has less than 10% error compared with the original flow pattern, and reduces artifacts and distortion, which improves the reconstructed image quality.