Abstract: Aiming at low description and stability of 3D feature descriptors in complex interference scenes, this paper proposes a feature description method based on voxel partition in spherical neighborhood space. This method consists of a stable local reference frame (LRF) and a feature descriptor based on voxel expression. For LRF, the Z-axis is calculated through the weighted covariance matrix, the sum of the weighted point cloud projection vectors is taken as the X-axis, and the Y-axis is obtained by the cross multiplication of the two axes. For feature descriptors, the spherical neighborhood is spatially divided, and the voxel label value is determined by judging whether there are points in each spatial voxel. Finally, the feature information of the key points is encoded according to the voxel index. The experiments show that, compared with other descriptors, this method has excellent performance against noise, uneven distribution of point cloud surface, scattered occlusion and other interference, and has good generalization. The registration experiments further verify the effectiveness of this descriptor.