Page 319 - Special Topic Session (STS) - Volume 4
P. 319

STS637 Ronald W. J.
            Examples of data science projects – (1) Mapping the urban forest
                In  September  2018,  data  scientists  of  the  ONS  Campus  published  the
            results  of  their  project  “Mapping  the  urban  forest  at  street  level”.  They
            developed an experimental method for estimating the density of trees and
            vegetation present at 10 meter intervals. The approach uses images sampled
            from Google Street View as the input to an image segmentation algorithm,
            which enables to derive a vegetation density map by percentage, for the road
            network  of  an  entire  city.  The  developed  system  is  built  on  recent
            advancements in the field of deep learning for semantic image segmentation.
            This system makes use of an automated tree detection procedure coupled
            with street-level image data. The result is a consistent methodology that can
            be used to add value to existing tree valuation approaches, with the capability
            to  assess  urban  vegetation  from  a  remote  location.  Such  benefits  are
            important for policy making and urban planning.
                Figure  2  depicts  the  segmentation  labels  produced  by  the  PSPNet
            (Pyramid Scene Parsing Network ) and the Random Forest machine learning
                                            8
                     9
            methods .  Challenges  in  the  learning  of  these  methods  are  that  green
            vegetation will be less prevalent in autumn and winter months, and that not
            all tree species are green and not all green objects are trees. The PSPNet model
            is robust to all of these.
















            Figure 2: Comparison of vegetation segmentation methods, from left to right: ground-truth
            (Mapillary data), PSPNet, Random Forest, a* threshold. Images copyright Mapillary

            Examples  of  data  science  projects  –  (2)  Using  mobile  phone  data  for
            tourism statistics
                Another example is the use of mobile phone data. Mobile phones send
            and receive signals from cell towers about 30 to 100 times a day with the only
            condition  that  the  phone  is  turned  on.  Those  signals  constitute  valuable
            information, which can be made into estimates about population density at
            any  time  of  the  day,  can  determine  commuting  partners,  determine

            8  See https://hszhao.github.io/projects/pspnet/
            9  See https://datasciencecampus.ons.gov.uk/wp-content/uploads/sites/10/2018/09/ons-dsc-
            mapping-the-urban-forest.pdf
                                                               308 | I S I   W S C   2 0 1 9
   314   315   316   317   318   319   320   321   322   323   324