Page 19 - JCAU-6-1

Journal of Chinese
Architecture and Urbanism Spatial exploration through image semantic segmentation

These streets feature an array of restaurants, shops, and has been constructed. Simultaneously, every effort has been
entertainment venues, attracting people for shopping, made to retain the distinctive characteristics and imagery of
socializing, and experiencing urban life (Hui, 2021). the original area (Fan, 2010). Within the process of preserving
In contrast, historic streets hold significant historical, and utilizing these cultural heritage sites, it becomes crucial
cultural, or architectural values. They often preserve to quantitatively assess and preserve the spatial vitality of Lu
ancient buildings, street scenes, and cultural heritage, Xun’s hometown (Fan, 2010). Traditional assessment methods
retaining architectural styles, layouts, and streetscapes that such as mental maps and cognitive image analysis primarily
bear witness to the historical evolution of cities or regions. involve qualitative analysis and struggle to comprehensively
Historic streets attract tourists and residents to explore and accurately depict the unique value and spatial experience
the cultural background of the past, offering different of Lu Xun’s hometown (Zeng et al., 2021). Therefore, more
spatial atmospheres and distinct spatial experiences. advanced techniques are needed for precise analysis (Liu,
These areas are conceived in the two street types, 2021).
influenced by a range of factors such as the distribution In this paper, image semantic segmentation is
and proportion of different elements within the streets. employed to quantitatively analyze the differences
Semantic segmentation technology can conveniently and between historic streets and modern streets, providing a
intuitively obtain various information about each element clearer representation on various dimensions. The paper
in the image from the image level, including proportion, commences with a comprehensive overview of semantic
distribution status, and mutual connections. The current segmentation and its application in urban design, followed
popular application of semantic segmentation involves by data collection and pre-processing for applying
the recognition of building outlines in satellite images, semantic segmentation to street-view images of Lu Xun’s
followed by tasks such as 3D reconstruction and area hometown. Subsequent sections detail the training models
estimation statistics. However, there is currently a limited and frameworks used for image processing through
quantity of analysis, and the existing studies are somewhat Python. This case study offers a comprehensive visual
superficial in exploring the semantic segmentation-based reproduction and analysis of the spatial characteristics
spatial quality aspects of historic neighborhoods (Zeng and historical context of this cultural area in Lu Xun’s
et al., 2021; Cao, 2021; Cheng, 2022). hometown. Quantitative evaluation of the distribution
Lu Xun’s hometown is located in Shaoxing, Zhejiang and features of different elements in streetscapes enables
Province, China. This unique historical and cultural area, a better understanding of the spatial conditions of various
being the birthplace of Lu Xun – a prominent figure in types of streets in Lu Xun’s hometown. This, in turn,
modern Chinese literature – is rich in cultural heritage provides a scientific basis and decision-making support
and historical memories. Renowned for its distinctive for its conservation and restoration efforts, as well as for
historical districts, traditional architecture, and intriguing similar heritage sites.
alleyways, this locale provides a unique backdrop to Lu
Xun’s formative years. Notable sites in Lu Xun’s hometown 2. Methods
include Shaoxing Lu Xun Residence, Baicaoyuan (Hundred 2.1. Basic concepts and approaches of semantic
Grass Garden), Sanwei Shuwu (Three Flavor Study), Lu segmentation
Xun’s Ancestral Residence, Tugu Temple, Changqing Semantic segmentation is a task within the field of computer
Temple, Lu Xun’s Literary Theme Park, and Shaoxing Lu vision that aims to assign each pixel in an image to different
Xun Memorial Hall (designed by Academician Cheng semantic categories, achieving pixel-level segmentation of the
Taining), among many other cultural relics associated with image. The objective is to segment different objects and regions
Lu Xun (Fan, 2010).
in the image and assign each pixel to its corresponding semantic
As the city undergoes development and modernization, the category, including categories such as person, car, tree, and
spatial environment of Lu Xun’s hometown is also experiencing others. Common methods for semantic segmentation involve
changes. Given Mr. Lu Xun’s significant role in modern the use of convolutional neural networks (CNNs), which are
Chinese history, preservation and research efforts concerning deep learning models widely applied in image analysis tasks
his former residence have been systematically carried out (Shi, 2022). Specifically, for tasks demanding pixel-level detail,
since the mid-20 century. Emphasis has been placed on the computer programs such as fully convolutional networks
th
restoration and renovation of his former residence, ancestral proved effective (Deng et al., 2021; Fang et al., 2021). These
home, and school, along with the transformation of nearby networks excel at capturing local features and retaining
ordinary dwellings. Additionally, a Lu Xun Memorial Hall, spatial information, making them well suited for pixel-level
designed under the guidance of Academician Cheng Taining, classification (An, 2021; Cao, 2017; Dong, 2022).

Volume 6 Issue 1 (2024) 2 https://doi.org/10.36922/jcau.1736