A method for removing ring artifacts of absorption signals in grating-based X-ray phase contrast tomography is proposed. In the method, a processing algorithm combining sinogram-domain with reconstruction image-domain is adopted. For weak artifacts, sorting and filtering are performed in the sinogram-domain. While for strong artifacts, the residual image is calculated firstly according to the performance of ring artifacts in the polar coordinate system, which is converted to Cartesian coordinate system to obtain the artifacts and the sample boundaries. Then an image segmentation method based on machine learning is used to obtain the distribution of each type of samples, and in order to protect the boundary information, the internal region of the samples is obtained through morphology operation. Finally, the distribution features of the residual images are used to locate the artifact pixels, and the mean value of adjacent non-artifact pixels are used to replace them. The experimental results show that this method can effectively remove the ring artifacts in the image without destroying the sample boundaries.