摘要
针对历史建筑多尺度保护中的原型识别问题,尝试将深度学习运用到其下采样的识别推导过程,利于价值信息的精炼与传导。通过采集历史建筑图像与基础信息,构建适配历史建筑功能分类的数据集与深度学习模型,评估模型性能并降维可视化模型分类结果,依据样本相似性结构与核密度估计值识别功能原型,并在中东铁路建筑群进行技术应用,分析沿线功能原型在空间分布、类型分化、要素表达三方面的特征差异,为多尺度保护传承提供技术支撑。
In response to the prototype identification problem in multi-scale conservation of historical buildings,deep learning was applied to its reduction process in down-sampling,contributing to information refinement and transfer with value.This paper built a function dataset and deep learning model adapted function classification by collecting images and basic information about historic buildings.Then the model was trained and evaluated performance.The functional prototypes were identified based on universal similarity structure by kernel density estimates,which visualized the classification results through dimensionality reduction,taking historic buildings along the Chinese Eastern Railway as an example for technical application and further analyzing the characteristic differences of functional prototypes along the route in terms of spatial distribution,typological division,and elemental representation,effectively providing technical support for multi-scale heritage protection and inheritance.
作者
李沛伦
赵志庆
谢佳育
陈玉玲
LI Peilun;ZHAO Zhiqing;XIE Jiayu;CHEN Yuling
出处
《建筑学报》
CSSCI
北大核心
2024年第S01期60-65,共6页
Architectural Journal
基金
国家自然科学基金项目(T2261139560,52278055)
中央高校基本科研业务费专项资金(HIT.DZJJ.2023081)
关键词
历史建筑
功能原型
深度学习
图像分类
中东铁路
historic building
functional prototype
deep learning
image classification
Chinese Eastern Railway
作者简介
通讯作者:陈玉玲
