Abstract：Aiming at the problem of the loss of the details of the reconstructed image caused by excessive smoothness of the Tiknonov regularization in the image reconstruction of electrical capacitance tomography(ECT) system, an image reconstruction model with norm l2,p(0<p≤1), which combines the smoothness of Euclid normand the sparsity of norm lp(0<p≤1) matrix norm is presented, which is non-convex and non-Lipschitz continuous problem . The extensive experiments have showed, the l2,p-minimization, which has stronger adaptability, higher resolution and better image quality, is better than the Newton iterative method, conjugate gradient method and traditional singular value decomposition algorithm.
［1］ Yang Y J, Peng L L. Data Pattern With ECT Sensor and Its Impact on Image Reconstruction［J］. IEEE Sensors Journal, 2013, 13(5):1582-1593.
［2］ Yang W Q, Liu S. Electrical capacitance tomography with square sensor［J］. Electronic letters,1999,35(4):295-296.
［3］ Liu C, Xu L J, Cao Z. Measurement of nonuniform temperature distribution by combining line-of-sight TDLAS with regularization methods［J］. Instrumentation, 2014, (3):43-57.
［4］ 马敏, 高振福, 王化祥. 基于BP神经网络的ECT图像重建算法［J］.计量学报, 2013, 36(6):524-528.
［5］ Nie F P, Huang H, Cai X, et al. Efficient and robust feature selection via joint l2,1- norms minimization［C］// Advances in Neural Information Processing Systems 23, Conference on Neural Information Processing Systems 2010, Vancouver, British Columbia, Canada, 2010:1813-1821.
［6］ Wang L P, Chen S C, Wang Y P. A unified algorithm for mixed-norms minimization and its application in feature selection［J］.Computational Optimization and Applications,2014, 58(2) : 409-421.
［7］ 王园萍, 殷洪友. 基于矩阵分数范数的人脸识别方法［J］. 计算机技术与发展, 2015, 25(4):22-25.
［8］ Rakotomamonjy A, Flamary R, Gasso G, et al. lp-lq Penalty for sparse linear and Sparse multiple kernel multitask learning［J］. IEEE Transaction on Neural Networks, 2011, 22(8):1307-1320.
2017, Vol. 38 
