基于街道图像与深度学习的城市景观研究

USING STREET-LEVEL IMAGES AND DEEP LEARNING FOR URBAN LANDSCAPE STUDIES

城市街道不仅是人类活动的集聚地,也是居民与城市建成环境发生社会交互的主要界面。因此,加深对于城市街道景观的了解在城市研究工作中至关重要。街道图像获取性的大大提高为城市景观研究提供了新的机遇,也提高了街道景观研究与分析的准确性与多样性。本研究基于街道图像,呈现了新近研发的深度卷积神经网络在景观分析中的应用。利用经过训练的深度卷积神经网络模型,我们能够准确地从街道图像中识别出不同的城市特征。根据图像分割技术处理结果,我们进一步测算出了马萨诸塞州剑桥市的街道绿化空间分布情况,并对街谷开阔程度进行了量化分析。诸如上述人工智能与大规模采集的街道图像的结合,将为世界范围内的城市景观研究提供全新的视角。

Streets are a focal point of human activities and a major interface of the social interaction between urban dwellers and urban built environment. A better understanding of the urban landscapes along streets is thus important in urban studies.The increasing availability of street-level images provides new opportunities for urban landscape studies to study and analyze streetscapes at a fine level and from a different perspective.In this study,we presented an application of a recently developed deep Convolutional Neural Network on landscape analysis based on street-level images.different urban features were identified from street-level images accurately using a trained deep Convolutional Neural Network model.Based on the image segmentation results,we further measured the spatial distribution of the street greenery and quantitatively analyzed the openness of street canyons in Cambridge,Massachusetts. The proposed combination of Artificial Intelligence and the massively collected street-level images provides a new sight for urban landscape studies for cities around the world.