基于卷积神经网络的聚落肌理表意型分类探究

The Classification of Settlement Texture Based on Convolutional Neural Network

该文选取两类极端的聚落形态:有机型和无机型聚落肌理,基于聚落肌理图片数据,通过人工分类与卷积神经网络图像识别相结合,在大量样本训练基础上,根据聚落肌理形态表意特征对聚落进行分类判定。阐明了两种聚落肌理数据集的分类整理方法以及判定模型的建立过程,结合实际聚落案例,对不同聚落的序列信息以及内部空间关系进行分类研究,利于探究聚落序列符号的共性及多样性,可作为后续聚落空间形态关联研究的理论基础。

In this paper, two types of extreme settlement patterns are selected: organic and inorganic settlement texture. Based on the image data of the settlement texture, through the combination of artificial classification and convolutional neural network image recognition, based on a large number of sample training, the settlement is classified and judged according to the characteristics of the settlement texture. We illustrate two settlements texture data sets of sorting method and determine the process of establishing model, combined with the actual settlement case, the sequence information of different settlement and the classification between internal space, we explore the commonality and diversity of settlement sequence symbols, which can be used as a theoretical basis for subsequent research on spatial morphology of settlements.