To address the issue of ineffective copyright protection for high-dimensional images using existing deep learning-based watermarking algorithms, this paper proposes a multi-scale knowledge learning-based image watermarking algorithm for diffused-weighted image copyright protection. Firstly, a watermark embedding network based on multi-scale knowledge learning is proposed to embed the watermark. The semantic, texture, edge, and frequency-domain information of the diffused-weighted image are extracted as multi-scale knowledge features using a fine-tuned pre-trained network. Then, these multi-scale knowledge features are combined to reconstruct the diffused-weighted image while redundantly embedding the watermark. This results in a visually similar watermark-containing diffused-weighted image to the original image. Finally, a watermark extraction network based on pyramid feature learning is proposed to improve the algorithm's robustness by learning the distribution correlation of the watermark signal in different scales of the diffused-weighted image's context. Experimental results demonstrate that the proposed algorithm achieves a high average peak signal-to-noise ratio for the reconstructed watermarked images. In the face of common noise attacks, the watermark accuracy of this algorithm exceeds 95% and effectively protects the copyright information of the diffused-weighted image.