“…So far the effective inversion algorithms mainly include: Cao [10] proposed self-incentive simultaneous algebraic reconstruction algorithm(SASIRT) which was suitable for ray distribution was not uniform or measurement error was larger, however, it had slow calculation speed and low accuracy; Saad and Schultz [11] proposed generalized minimal residual method(GMRES) which could be used for solving non symmetric linear equations, the calculation process would not be interrupted generally until obtaining the exact solution, nevertheless, there may be not converge; Van and Vorst [12] proposed the double stable conjugate gradient method (BICGSTAB), which could be used for solving linear equations whose coefficient matrix was asymmetric, it used short recursive method to reduce residual progressively, so the advantage was it occupied less memory, but the convergence was irregular, the convergence rate may be amplified severely under the condition of finite precision; LSQR with damping factor method (DLSQR) proposed by Yang [13], it improved the inversion precision effectively, avoided numerical instability of LSQR algorithm when the measurement error was large, it was especially suitable for solving the equations whose coefficient matrix was large and sparse, compared with other iteration method, it could obtain faster convergence rate and better acceptable results in solving singular or illconditioned problems, currently it is practical inversion method which was most commonly used, however, the occupation of computer memory was large and accuracy should be further improved. In recent years, artificial intelligence method was also applied to the inversion algorithm [14], such as Simulated Annealing(SA) or Genetic Algorithm(GA), but it had strong dependence on the initial model, easily influenced by random disturbance, the distribution reconstruction effect on slightly more complex medium was poor.…”