<p>Recognition of oracle bone characters is essential for accurately interpreting Chinese cultural history. However, existing recognition techniques are significantly challenged by severe noise present in the original images, leading to low recognition accuracy. To address this issue, this paper proposes an oracle bone recognition method based on eigenvalue-based preprocessing and deep learning. The preprocessing method computes four tailored eigenvalues-equivalent connectivity domain diameter, minimum connectivity domain distance, point feature analysis matrix, and point eigenvalue-based on the unique structural characteristics of oracle bone characters, enabling more effective noise reduction. Subsequently, an improved model, DC-YOLOv8, is introduced, which replaces the standard convolution module with a deformable convolution module in the high-level layers of the backbone to enhance adaptability to highly deformed characters. Additionally, the Convolutional Block Attention Module is incorporated into the head component to improve the model’s ability to extract key features in critical regions. Experimental results demonstrate that YOLOv8 achieves accuracy rates of 90.14% and 90.97% on two datasets after applying the proposed preprocessing method. Furthermore, the accuracy of DC-YOLOv8 reaches 95.11%, 91.74%, and 93.75% across three datasets.</p>