基于CBAM-YOLOv4的东巴象形文识别方法研究  

DONGBA PICTOGRAPH RECOGNITION METHOD BASED ON CBAM-YOLOV4

作　　者：黄颢 吴国新 徐小力 赵西伟 Huang Hao;Wu Guoxin;Xu Xiaoli;Zhao Xiwei(Key Laboratory of Modern Measurement&Control Technology,Ministry of Education,Beijing Information Science&Technology University,Beijing 100912,China)

机构地区：[1]北京信息科技大学机电学院现代测控技术教育部重点实验室,北京100912

出　　处：《计算机应用与软件》2025年第2期196-201,226,共7页Computer Applications and Software

基　　金：国家重点研发计划项目(2020YFB1713200);北京市教委科研计划项目(KM202011232001)。

摘　　要：东巴象形文字是中国早期使用的一种象形文字,对该文字的识别和数字化保护等方面的研究对传承我国文化具有深远意义。针对从东巴古籍提取的象形文字结构复杂、存在异体字、记录该文字的特殊东巴纸的纹理特征干扰识别的情况,提出一种基于CBAM-YOLOv4的图像识别改进算法,该算法添加注意力机制模块CBAM(Convolutional Block Attention Module)和特征融合模块,通过CBAM中的通道和空间注意力子模块依次对图像推断出注意力图,并结合特征融模块对输入的东巴象形文字图片进行更深的特征提取,从而实现对YOLOv4图像检测识别算法的优化。将改进后的CBAM-YOLOv4算法应用于东巴象形文字识别,相比YOLOv4算法mAP值提高了4.42百分点,表明该算法具有较好的东巴文字识别性能。Dongba pictograph is a kind of pictograph used in early China.The research on the recognition and digital protection of Dongba pictograph is of far-reaching significance to the inheritance of Chinese culture.In view of the complex structure of the pictographs extracted from ancient Dongba books,the presence of heterogeneous characters,and the interference of texture features of the special Dongba paper recording the characters,an improved image recognition algorithm based on CBAM-YOLOV4 is proposed.The attention mechanism module CBAM and feature fusion module were added to the algorithm.The channel and spatial attention sub-modules in CBAM were used to deduce the attention diagram of the images in turn,and the feature fusion module was used to further extract the features of the Dongba pictograph images,so as to realize the optimization of YOLOv4 image detection and recognition algorithm.The improved CBAM-YOLOV4 algorithm was applied to Dongba pictograph recognition.Compared with the YOLOv4 algorithm,the mAP value of the improved algorithm is improved by 4.42 percentage points,indicating that the algorithm has better Dongba character recognition performance.

关 键 词：东巴文识别 YOLOv4 CBAM 特征提取 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]