Measuring human-scale urban form and its determinants of inequality.

我们提出了综合和发展机器学习和统计学的方法，利用图像来定量评估城市空间不平等。通过衡量建成环境的相似性，我们从街道的人类尺度角度处理不平等问题，考察城市规划作为经济、环境和社会包容或排斥的潜在机制所起的作用。使用伦敦市作为我们的第一个城市布局，我们将通过以下方式做到这一点：-利用谷歌街景图像像素分布中编码的信息来定义地理位置图像之间的相似性度量。对相似图像进行群集，从而创建跨越空间的视觉上同质的图像集，并且还在空间上受限制以定义邻居。-使用与诸如卫星图像和道路网络的附加数据源相结合的群集图像来创建邻居的地理空间地图。-制定在空间流行病学中使用图像的方法，不仅界定新的邻里边界，而且将图像用作空间模型中的协变量。-包括在整个分析过程中重点关注图像特征的可解释性，以便就建筑环境中不平等的决定因素得出结论。我们认为，通过发展利用图像分析城市不平等的方法，我们将首先对人类规模的城市形式的空间分布进行无监督的量化，制定可转移到世界各地城市的方法；其次，制定在贝叶斯模型中纳入图像协变量的方法，这些方法可转移到广泛的领域应用；最后，将城市形式作为城市不平等的决定因素进行研究，有助于我们更好地理解人类规模的城市形式及其在经济、环境和社会三个维度的表现。

We propose synthesising and developing methods from machine learning and statistics to quantitatively assess spatial urban inequalities with the use of imagery. Through measuring similarity of the built environment, we approach inequalities from the human-scale view of the street, examining the role of urban planning as a potential mechanism for economic, environmental and social inclusion or exclusion. Using the city of London as our first urban layout, we will do this by:- harnessing information encoded in distribution of pixels from Google Street View images to define similarity metric between geo-located images. Clustering similar images and thus creating visually homogeneous sets of images, across space, and also spatially conditioned to define neighbourhoods.- creating geospatial map of neighbourhoods using clustered images combined with additional data sources such as satellite images and road networks. - developing methods for the use of imagery in spatial epidemiology, not only by defining new neighbourhood boundaries, but also, the use of imagery as covariates in spatial models. - including throughout our analysis a focus on interpretability of image features, in order to draw conclusions about determinants of inequality in the built environment. We believe that by developing methods for analysing urban inequalities using imagery, we will first; quantify the spatial distribution of human-scale urban form in an unsupervised manner, developing methods which are transferable to cities worldwide, second; develop methodologies for incorporating image covariates in Bayesian models, which are transferable to a wide range of domain applications, and lastly; examine urban form as a determinant of urban inequality, contributing to our broader understanding of human-scale urban form and its performance in three dimensions: economic, environmental and social.