数字图像是古文字最主要的保存方式，利用图像技术识别古文字是文字学研究领域的重要课题，也是助力汉字文化传播的关键核心技术。本项目针对古文字图像样本数量少且不均、识别模型泛化能力弱等问题，研究古文字的年代-地域-图像关联学习的联合识别算法，以实现先秦多种古文字的识别。研究内容包括：根据古文字考释的推理逻辑，研究构建古文字的年代-地域-图像相关联的多维度特征提取机制；研究基于多维度特征的联合识别算法，提高模型的泛化能力；研究元学习与古文字多维度特征相融合的特征学习方法，强化数据间区分特征的敏感度，建立针对小样本数据集的联合识别模型；构建多种古文字数据集，研究数据集、检测识别模型和模型优化方式的集成方法，开发先秦古文字识别与测试系统。项目研究将为古文字的研究和推广提供技术支撑。

Ancient characters are mostly in the form of digital images. The recognition of ancient characters based on images is an important subject in the field of philology and a key core technology for dissemination of ancient characters. In order to solve the problems of few samples and weak model generalization ability, we introduce Old Chinese charactors recognition model based on text-image correlation learning, which meets the needs of the recognition of various ancient characters in the pre-Qin period. Firstly, the multi-dimensional feature extraction mechanism that correlates age and geographic information with image features is proposed, which refers to the reasoning logic of ancient characters research. Then, algorithm based on multi-dimensional features is proposed to improve the generalization ability of the ancient character recognition model. With the advantage of enhancing the sensitivity of distinguishing features between data, a method is proposed to achieve feature learning that combines meta-learning and multi-dimensional features of ancient characters. On the basis of this method, a recognition model for small sample association learning is established. At last, the method of integrating the data set, the recognition model and the model optimization method is proposed. Based on a variety of ancient character data sets, a pre-Qin ancient character recognition and testing system is constructed by the project. The research content of the project will provide technical support for the research and dissemination of ancient characters.