何宛余 李春 聂广洋 杨良崧 王楚裕

He Wanyu, Li Chun, Nie Guangyang, Jackie Yong Leong Shong, Wang Chuyu

关键词：人工智能；深度学习；卷积神经网络；图像判别；城市感知

Keywords：Artificial Intelligence; Deep Learning; Convolutional Neural Network; Image Discrimination; Urban Perception

摘要：

作为人工智能领域的研究重点，机器学习近年衍生出了各式各样的智能化应用，例如图像判别、语音助手和智能翻译等。尤其是图像判别技术已在各行业进行了大量的研究和实践，城市领域也不例外，这很大程度上是因为深度学习的卷积神经网络在计算机视觉领域取得了令人瞩目的成果。这也使得训练计算机判别建筑风格、城市肌理等城市特征的准确率大幅提升。本研究立足于深度学习图像判别技术，探索卷积神经网络在城市感知方面的应用。鉴于直接利用现成开源的带标签图像数据集训练个性化图像判别模型可能带来局限性和误差，本研究探索了从收集数据到自定义训练数据集，到搭建满足特定需求的图像判别模型的整体流程，并通过三个实验案例：城市风貌分析、城市问题侦测和城市肌理评估，阐明深度学习在城市感知和城市规划中的应用可能性及潜力。

Abstract：

Nowadays, machine learning attracts intense attention from artificial intelligence researches and extends a variety of applications such as image discrimination, voice assistant and smart translator. In particular, image discrimination has been extensively studied and practiced in various industries, including urban field. Thanks to Convolutional Neural Network (CNN) based on Deep Learning (DL) that has made remarkable achievements in computer vision, it is more efficient to train computer to discriminate architecture styles, urban texture and other urban features. Based on image discrimination by DL, this research focuses on exploring the applications of CNN in the field of urban perception. In consideration of limits and errors brought by training customized image discrimination model with the existing open source labeled image dataset, this paper explores a whole process from collecting data, self-constructing training dataset to building a customized image discrimination model which satisfies specific requirements. The latent application of DL in urban scale are discussed through three experiment cases: the cityscape analysis, urban problem detection and urban pattern evaluation.

何宛余，小库科技，创始人兼首席执行官

李春，小库科技，联合创始人兼首席技术官

聂广洋，小库科技，人工智能学家

杨良崧，小库科技，高级研究员

王楚裕，小库科技，智慧城市高级研究员

• [1] 尼格尔· 泰勒. 1945 年后西方城市规划理论的流变[M]. 李白玉, 陈贞, 译. 北京: 中国建筑工业出版社, 2006.

  [2] LUSK E, BOYLE J, WOS L, et al. Automated reasoning: introductions and applications[M]. New Jersey: Prentice Hall, 1984: 4.

  [3] FLASI?SKI M. Introduction to artificial intelligence[M]. Switzerland: Springer Nature, 2016: 5.

  [4] 朱玮, 王德. 大尺度城市模型与城市规划[J]. 城市规划, 2003, 27(5): 47-54.

  [5] 陈苏雅. 5 月1 日起深圳交警试点“刷脸”执法( 附电子警察位置)[N/OL]. (2018-04-19)[2018-10-22]. http://www.sznews.com/news/content/2018-04/19/content_18921886.htm.

  [6] ARIETTA S M, EFROS A A, RAMAMOORTHI R, et al. City forensics: using visual elements to predict non-visual city attributes[J]. IEEE Transactions on Visualization and Computer Graphics, 2014, 20(12): 2624-2633.

  [7] COURVILLE A, GOODFELLOW I, BENGIO Y. Deep learning[M]. Cambridge: MIT Press, 2016: 9.

  [8] KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[C] // PEREIRA F, BURGES C J C, BOTTOU L, et al. Proceedings of the 25th International Conference on Neural Information Processing Systems – Volume 1. New York: Curran Associates Inc., 2012: 1097-1105.

  [9] GILL J K. Automatic log analysis using deep learning and AI for microservices[EB/OL]. (2017-07-21)[2018-10-20]. https://www.xenonstack.com/blog/data-science/log-analytics-deep-machine-learning-ai/.

  [10] NG A. Machine learning yearning (draft version)[M/OL]. (2018-09-29)[2018-09-28]. https://www.deeplearning.ai/machine-learning-yearning/.

  [11] PATEL S, PINGEL J. Introduction to deep learning: what are convolutional neural networks?[EB/OL]. (2017-04-24)[2018-09-28]. https://www.mathworks.com/videos/introduction-to-deep-learning-what-areconvolutional-neural-networks--1489512765771.html.

  [12] NAIK N, KOMINERS S D, RASKAR R, et al. Computer vision uncovers urban change predictors of physical urban change[J]. Proceedings of the National Academy of Sciences, 2017, 114(29): 7571-7576.

  [13] UBA B K. Land use and land cover classification using deep learning techniques[D/OL]. Arizona: Arizona State University, 2016. [2018-11-20]. https://repository.asu.edu/attachments/170740/content/Uba_asu_0010N_15901.pdf.

  [14] PAPADOMANOLAKI M, VAKALOPOULOU M, ZAGORUYKO S, et al. Benchmarking deep learning frameworks for the classification of very high resolution satellite multispectral data[C]. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences Volume III-7. G?ttingen: Copernicus Publications, 2016: 83–88.

  [15] PENATTI O A B, NOGUEIRA K, SANTOS J A D. Do deep features generalize from everyday objects to remote sensing and aerial scenes domains?[C]. 2015 IEEE Conference on Computer Vision and Pattern Recognition Workshops. New York: Curran Associates Inc., 2015: 44-51.

  [16] ROMERO A, GATTA C, CAMPS-VALLS G. Unsupervised deep feature extraction for remote sensing image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2016, 54(3): 1349-1362.

  [17] SHEN Q, ZENG W, YE Y, et al. StreetVizor: visual exploration of human-scale urban forms based on street views[J]. IEEE Transactions on Visualization and Computer Graphics, 2018, 24(1): 1004-1013.

  [18] YE Y, RICHARDS D, LU Y, et al. Measuring daily accessed street greenery: a human-scale approach for informing better urban planning practices[J]. Landscape and Urban Planning, 2018. https://doi.org/10.1016/j.landurbplan.2018.08.028.

  [19] DUBEY A, NAIK N, PARIKH D, et al. Deep learning the city: quantifying urban perception at a global scale[C]. LEIBE B, MATAS J, SEBE N, et al, eds. Computer Vision – ECCV 2016. Cham: Springer, 2016: 196-212.

  [20] NADAI M D, VIERIU R L, ZEN G, et al. Are safer looking neighborhoods more lively? a multimodal investigation into urban life[C] // HANJALIC A, SNOEK C. Proceedings of the ACM Multimedia Conference 2016. New York: ACM, 2016: 1127-1135.

  [21] SERESINHE C I, PREIS T, MOAT H S. Using deep learning to quantify the beauty of outdoor places[J/OL]. London: Royal Society Open Science, 2017, 4(7). (2017-07-01)[2019-09-28]. https://royalsocietypublishing.org/doi/full/10.1098/rsos.170170.

  [22] DOERSCH C, SINGH S, GUPTA A, et al. What makes Paris look like Paris?[J]. Communications. New York: ACM, 2015, 58(12): 103–110.

  [23] SHALUNTS G, HAXHIMUSA Y, SABLATNIG R. Architectural style classification of building facade windows[C] // BEBIS G, BOYLE R, PARVIN B, et al. ISVC 2011: Advances in Visual Computing. Berlin: Springer, 2011: 288.

  [24] ESLAMI S M A, REZENDE D J, BESSE F, et al. Neural scene representation and rendering[J]. Science, 2018, 360(6394): 1204–1210.

  [25] SCOTT G J, ENGLAND M R, STARMS W A, et al. Training deep convolutional neural networks for land-cover classification of high-resolution imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(4): 549-553.

  [26] JEAN N, BURKE M, XIE M, et al. Combining satellite imagery and machine learning to predict poverty[J]. Science, 2016, 353(6301): 790-794.

  
  

• 收起