基于CA-FasterR-CNN的甲骨文原始拓片单字分割方法

CA-Faster R-CNN-based Single-word Segmentation Method for Oracle Bone Original Topographies

甲骨文拓片经过长时间的埋藏和侵蚀,变得形态复杂,字体模糊,单字之间缺乏明确的分隔,这给甲骨文识别带来了极大的困难。基于此,本文提出了一种基于坐标注意力机制的快速区域卷积神经网络(Coordinate Attention Mechanism-based Faster Region Convolutional Neural Network,CA-Faster R-CNN)模型以实现对甲骨文拓片图像中的单字分割。通过坐标通道注意力机制的引入,模型能够更加关注甲骨文字形特征,从而提升了对甲骨文图像细节的捕捉能力,最后训练结果框线与标准框线基本重合,证明模型分割效果良好。

After a long time of burial and erosion,the oracle bone topographies have become complex in form,with blurred fonts,and lack of clear separation between single characters,which brings great difficulties to the recognition of oracle bones.Based on this,a Coordinate Attention Mechanism-based Faster Region Convolutional Neural Network(CA-Faster R-CNN)model based on the coordinate attention mechanism is proposed to realize the segmentation of single characters in oracle bone tablet images.Through the introduction of the coordinate channel attention mechanism,the model can pay more attention to the character shape features of oracle bones,thus improving the ability to capture the details of the oracle bone image,and the final training result box lines largely overlap with the standard box lines,which proves that the model segmentation effect is good.