利用深度卷积神经网络算法识别传统篆体书法的应用研究

Application Research on Recognizing Traditional Seal Calligraphy Using Deep Convolutional Neural Network Algorithm

利用先进的数字化和智能化技术对我国古代文化遗产进行数字化应用和有效保护,以更快捷、更高效的方式解决凭借人力无法解决的问题,具有重要的现实意义。通过研究深度卷积神经网络在书法文字检测识别任务上的应用,设计了一个篆体书法文字检测识别的完整系统,包括数据采集、数据扩充、算法训练与测试和算法模型部署等流程。整个系统以YOLOv4目标检测算法为基础,根据篆体书法图像数据特征对采集得到的数据进行有效地扩充,进行多次训练和验证测试,最终获得了89.7%的平均精度、92.3%的准确率和94.7%的召回率,同时达到45张/s的识别速度;最终将识别检测模型部署至服务器端,并提供了接口供外部调用。实验证明设计的识别系统可以利用深度卷积神经网络自动、快速、准确地对篆体文字进行定位和识别,并且可以方便地调用训练和部署完成的模型。

It is of great practical significance to use advanced digital and intelligent technology to digitize and protect the ancient cultural heritage effectively and solve the problems that cannot be solved by the workforce in a faster and more efficient way.This paper mainly studies the application of deep convolutional neural network in calligraphy character detection and recognition task and designs a complete seal script calligraphy character detection and recognition system,including data acquisition,data expansion,algorithm training and testing,algorithm model deployment,and other processes.The whole system is based on the YOLOv4 object detection algorithm and effectively expands the collected data according to the seal calligraphy image data characteristics.After many times of training and verification tests,the 89.7%average presion,92.3%accuracies,and 94.7%recall rate were achieved,and the recognition speed of 45 images per second was completed.Finally,identification detection model is deployed to the server-side and provide an interface for external invocation.The experiment proves that the recognition system can use a deep convolutional neural network to locate and recognize the seal script automatically,quickly and accurately.The model of training and deployment can be conveniently called easily.