A one-dimensional inversion method for the vertical Z component of any measuring point in the frame of the large loop source transient electromagnetic method is proposed in this paper. The algorithm uses the CMD regularization factor optimization scheme, smoothness function and natural boundary conditions to constrain the inversion process, reduce the initial value dependence, and ensure the autonomy of the inversion process and the stability of the inversion results. In the inversion process, while ensuring that the model residual decreases along the gradient, the maximum value of the model correction is limited, so that the inversion model is not easy to fall into the local minimum, which ensures the global convergence of the inversion results. The sensitivity matrix is updated by Newton's method, thus reducing the workload of the entire inversion calculation. Finally, the reliability of the algorithm is guaranteed by five-layer double low-resistance layer model, double high-resistance layer model and measured data.