Abstract: Rock and mineral samples contain a large amount of high-density metal substances, which often lead to metal artifacts in CT images and seriously affect the parameters analysis accuracy of rock and mineral samples. In order to improve the quality of CT images and suppress metal artifacts, in this paper we propose a dual-domain adaptive network-based metal artifact reduction algorithm for CT images of rock samples (DDA-CNN-MAR). The metal artifacts are suppressed by the projection domain network and the image domain network successively, and the dual domain processing results are adaptively fused to realize the end-to-end mapping from images with artifacts to artifact-free images. The algorithm is based on the residual encoder-decoder network model (RED-CNN), which is easy to extract features and restore image details. The dual-domain structure can adaptively adjust the weights of the projection domain (artifact suppression) and image domain (detail inpainting) to obtain optimal reduction results. The experimental results show that, compared with the RED-CNN metal artifact denoising method in the image domain, the MSE of the image corrected by the DDA-CNN-MAR method is reduced by 2.570, while the PSNR and SSIM are increased by 1.218 dB and 0.018 respectively, which effectively improves the CT image quality.