人工神经网络在建筑风格判定中的应用——以大连近代历史建筑为例

The Application of Artificial Neural Networks in the Determination of Architectural Style: The Case of Dalian’s Modern Historical Architectures

卷积人工神经网络可用于图片分类研究,根据这一手段建立了基于建筑图片的建筑风格判别方法,其结果可量化不同建筑风格在同一幢建筑表皮上的表现程度。近代建筑通常在一栋建筑中混杂不同的建筑风格,通过量化其不同风格的表现程度可以为其历史走向定位提供辅助手段。以大连地区近代历史建筑为例,简要介绍了卷积神经网络的基本原理与Inception-v3模型,说明了八种建筑风格图像数据集的搜集、整理方法以及建筑风格判定模型建立过程,并结合实际历史建筑案例,分析了该模型作为辅助工具区分量化建筑风格的使用方法,比较了两栋近代建筑中不同建筑风格的分析结果,简述了该方法在近代建筑风格判别中的作用。

CNN can be used for image classification research. According to CNN, we establish a method of architectural style discrimination based on architectural pictures, which can quantify the performance of different architectural styles in the skin of the same building. Taking the modern historical buildings in Dalian as an example, we introduce the principles of CNN and the Inception-v3 model, then we explain the collection and sorting methods of eight architectural style image data sets and the establishment process of architectural style decision models. Combined with actual historical building case, we analyze the use of this model to distinguish quantitative architectural styles,finally we compare the analysis results of different architectural styles in two modern buildings and describe the role of this method in discriminating modern architectural styles.