In the practical application of atmospheric lidar, when there are clouds, smoke or hard targets with large backscatter coefficient on the laser detection path, and the abrupt signal is strong enough, the backward integration inversion of extinction coefficient by Fernald method will produce obvious inversion error in the backward data of abrupt signal. Based on the characteristics of Fernald backward integration and combined with the piecewise slope method, a new inversion method was proposed to deal with the above situation and optimize the inversion strategy of the traditional algorithm. That is, by judging the position of the abrupt signal, the reference distance and boundary value were constantly updated for iterative inversion, and more accurate inversion data were spliced and overwritten into the initial inversion data. The validity of the inversion method was verified by using the measured data obtained from vertical and horizontal detection, and the optimized data was compared with the corresponding data when there was no abrupt signal nearby. The results show that, compared with the traditional algorithm, the relative errors of extinction coefficients of vertical and horizontal detection data are optimized by the new inversion method, which is about 79% and 96%, respectively, verifying the feasibility of the new method.