In order to achieve the measurement of the 3D environment, a LiDAR SLAM 3D mapping algorithm for weak geometric road environments oriented to sparse features and fusing semantic features is built. The extraction of ground point cloud is optimized for the distortion-corrected point cloud data. The characteristics of columnar structures and roadside structures in sparse feature environments are analyzed, and the corresponding semantic feature extraction scheme is designed to implement the LiDAR odometry based on the above features. A certain detection strategy is proposed to construct loop closure constraints to avoid invalid and frequent loop closure detection. A global descriptor-based loop closure detection method is introduced for loop closure drift in large scale range. In the back-end of the SLAM algorithm, keyframes are matched with local maps, and the LiDAR, IMU, and loop closure constraints constructed above are added into the factor map to optimize the algorithm, and a SLAM 3D mapping algorithm is finally constructed for the fusion of semantic features in sparse feature environments.