多模融合的陶瓷图像中文描述生成方法研究  

Study On The Chinese Description And Generation Method Of Multimodal Fusion Ceramic Image

作　　者：胡智猛 彭永康 张秀娟 HU Zhimeng;PENG Yongkang;ZHANG Xiujuan(School of Information Engineering,Jingdezhen Ceramic University,Jingdezhen,China,333403)

机构地区：[1]景德镇陶瓷大学信息工程学院,江西景德镇333403

出　　处：《福建电脑》2025年第1期11-17,共7页Journal of Fujian Computer

基　　金：景德镇市级科技计划项目(No.2023GY001-01);江西省03专项及5G项目(No.20232ABC03A29);高等学校大学生创新创业训练计划项目(No.202310408016)资助。

摘　　要：早期的陶瓷图像描述生成方法存在识别和描述准确性不足的问题。针对这些问题,本文提出一种基于深度残差网络和特征金字塔网络的多尺度图像特征提取方法,并利用带有加性注意力机制的长短期记忆网络生成中文描述的Res-FL模型。实验结果表明,在描述准确性和细节捕捉方面,Res-FL模型显著优于传统神经网络方法,在提高陶瓷图像描述的一致性和精确性方面具有较高的应用价值。Early methods for generating ceramic image descriptions had issues with insufficient accuracy in recognition and description.To address these issues,this paper proposes a multi-scale image feature extraction method based on deep residual networks and feature pyramid networks,and utilizes a long short-term memory network with additive attention mechanism to generate a Res FL model for Chinese descriptions.The experimental results show that the Res FL model is significantly superior to traditional neural network methods in terms of description accuracy and detail capture,and has high application value in improving the consistency and accuracy of ceramic image description.

关 键 词：陶瓷图像 图像描述 图像特征提取 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]