Yoga in New York. Image via Creative Commons

人工智能通过分析建筑物和绿地评估肥胖问题
Artificial Intelligence Estimates Obesity by Analyzing Buildings and Green Space

由专筑网邢子，李韧编译

华盛顿大学研究人员开发了一种人工智能算法，即通过分析城市的基础设施来评估肥胖问题，城市福祉和健康追踪科技又向前迈进了一步。研究人员在JAMA网络公开赛上发表的报告解释了该算法如何通过使用谷歌卫星和街景图像来表达各种城市关系。正如Quartz报道的那样，拥有更多绿色空间的区域以及建筑密度角度的区域的人群肥胖率较低。

Urban well-being and health tracking has taken another step forward, as researchers from the University of Washington have created an artificial intelligence algorithm that estimates obesity levels by analyzing a city's infrastructure. Published in JAMA Network Open, the researchers' report explains how the algorithm uncovers urban relationships by using satellite and Street View images from Google. As Quartz reports, the project correlated areas with more green spaces and areas between buildings with lower obesity rates.


Yoga in New York. Image via Creative Commons

该算法在六个城市使用超过150,000个卫星图像进行实验，使用深度学习来了解城市规划及其对肥胖问题的影响。该研究旨在说明卷积神经网络如何协助研究建筑环境与肥胖率之间的关系，同时分析和改善城市人群的健康状况，并且塑造新建筑。项目中包含了96类兴趣点，其中结合思考了城市便利设施对邻里活动的影响。

研究人员明确表示他们觉得该算法可能会受到经济状况的影响。考虑这一情况，该项目还可以表明富裕社区和居民肥胖之间的相关性。通过进行一系列的验证测试，研究人员发现该算法确实将绿色空间、建筑数量与肥胖联系了起来。正如该论文所述，“美国超过三分之一的成年人口都有肥胖问题。肥胖与遗传、饮食、身体活动和环境等因素有关”，研究人员希望他们的工作能够展示卷积神经网络（CNN）对建筑环境特征所进行的量化处理结果。

虽然这项研究是基于美国的数据，但研究人员希望该算法能够适用于分析世界各个城市。然而，该项目确实能够证明CNN在肥胖患病率与物理环境之间存在着关联。

Trained using more than 150,000 satellite images across six cities, the algorithm uses deep learning to understand city planning and its affect on obesity. The study looked to answer how convolutional neural networks can assist in the study of the association between the built environment and obesity prevalence. Aiming to analyze and improve a city's health, the project hopes to shape new construction. 96 categories of points of interest were included in the work, accounting for the effect urban amenities can have on the the activity of a neighborhood.
Researchers have explicitly stated their understanding that the algorithm can be skewed by income and wealth. Recognizing this condition, the project can also draw correlations between wealthier neighborhoods and resident obesity. By conducting a series of validation tests, researchers found that the algorithm does link green space and the number of buildings to obesity, not just wealth. As the paper states, "more than one-third of the adult population in the United States is obese. Obesity has been linked to factors such as genetics, diet, physical activity, and the environment." Researchers hope their work can show how convolutional neural networks (CNN) can allow for consistent quantification of a built environment's features.
While the study was based on US data, researchers hope the algorithm can be adapted to analyze cities around the world. However, the project does begin to provide evidence of the efficacy of CNNs at associating obesity prevalence with significant physical environment features.