Rubbing Oracle Bone Character Recognition Based on Improved YOLOv8 Network

Aiming at the problems of low recognition accuracy caused by the background noise of oracle bone characters and irregular font morphology, as well as the deployment requirements of application scenarios such as research and education, exhibitions, etc., the paper proposes a lightweight improved YOLOv8n oracle bone character recognition algorithm. First, the incorporation of the Deformable Convolutional Networks Version 2 module in the backbone network aims to improve the model's ability to extract irregular target features. Second, the redundant small target detection layer is removed to reduce the model complexity to decrease the computation. Finally, a lightweight CA attention mechanism is added in front of the detection head to effectively capture the relationship between channel and position and improve the feature representation ability of the model. The experimental results show that the improved model improves the accuracy by 2.9%, reduces the computation by 30%, and the number of parameters is only 3.2M to meet the deployment requirements of the application scenarios.