福建电脑2025，Vol.41Issue(11)：15-20,6.DOI:10.16707/j.cnki.fjpc.2025.11.003
面向唐卡分类的全维度动态卷积ResNet模型
A ResNet-Based Model with Omni-Dimensional Dynamic Convolution for Thangka Classification
摘要
Abstract
To improve the accuracy of thangka image classification and promote its digital protection,this paper proposes a classification method that integrates Full Dimensional Dynamic Conv(ODConv)and ResNet,and constructs the ODConv-ResNet101 model.This model significantly improves the network feature extraction capability by introducing dynamic mechanisms in multiple dimensions such as spatial domain,input channel,and output channel.Experiments based on a self built Tangka dataset show that the proposed model has a classification accuracy of 92.32%,which is superior to improved ResNet models such as SE and CBAM,demonstrating stronger classification performance and generalization ability.Research has confirmed that ODConv ResNet can effectively improve the accuracy of thangka image recognition,providing a new technological path for the intelligent protection of ethnic art resources.关键词
唐卡分类/全维度动态卷积/残差网络/图像识别Key words
Tangka Classification/Full Dimensional Dynamic Convolution/Residual Network/Image Recognition分类
信息技术与安全科学引用本文复制引用
唐忠杰,潘春花..面向唐卡分类的全维度动态卷积ResNet模型[J].福建电脑,2025,41(11):15-20,6.基金项目
本文得到青海非物质文化遗产工艺品数字化追溯系统开发(No.2020-GX-113)、青海"非物质文化遗产"数字化保护、展示及综合应用项目建设(No.2021-Y-20)资助. （No.2020-GX-113）
