Abstract: Chinese Spelling Correction is a task aimed at detecting and correcting misspelled characters in Chinese texts. Due to the current challenge of lacking large-scale annotated sentences, this paper proposes a data augmentation method based on dynamic masking from easy to hard. This method dynamically adds masks to source sentences from easy to hard during the training process, generating a more diversified set of correction sentence pairs. Additionally, this paper introduces an adaptive masking loss function that assigns different levels of attention to samples with different difficulty masks. On the SIGHAN 13, SIGHAN 14, and SIGHAN 15 datasets, experiments demonstrate the proposed method out-perform baselines with F₁ scores of 85.1%, 70.7%, and 81.0%, respectively.