The autonomous movement of shotcrete robot in coal mine roadways requires an accurate environment map， and the weight degradation and particle dilution in large scenes are easy to happen in the Gmapping algorithm， which leads to excessive pose estimation error of robot and poor consistency such as map overlap and layering. Therefore a classification recovery resampling algorithm is proposed. When resampling， it modifies and recovers the low-weight particles in proportion， which makes full use of the existing information and tries to protect the particle diversity while restraining the weight degradation. The experimental results show that when mapping ACES building and MIT Killian Court data sets， using the traditional algorithm to locate and map the number of particles with extremely poor effect， the improved Gmapping algorithm can still maintain the translation error and rotation error of the robot at a low level， and can obtain a clear and accurate environment map .The particle distribution in the later stage of the classification recovery resampling algorithm is more in line with the requirements of the particle filter， which verifies the effectiveness of the classification recovery resampling algorithm.