| Functional zone reflects city’s spatial structures,and as a carrier of social and economic activities,it is of critical significance to urban management,resource allocation and planning.However,most researches on functional zone division are based on a large spatial scale such as blocks or other scales larger than it.Aiming at a subtle fine functional result,the concept of Super Image Object(SIO)was stressed proposed,along with a Super Image Object-Convolutional Neural Network(SIO-CNN)based urban functional zone fine division method applied to very high resolution(VHR)remote sensing image.Besides,the related and similar concepts of SIO also have been explained.Fractal network evolution approach was used to segment SIO.In order to prove the feasibility and effectiveness of using SIO as the basic functional zone unit,not only the segmented SIO was visual evaluated,but the statistic method was also used to evaluate the inner homogeneity and external heterogeneity of basic functional zone units(including SIO,Block,Grids with different sizes).Object based CNN was used to decided the functional attributes of SIOs.A random point generation algorithm was used to generate the voting points of the research area,besides,one center point will be generated in every SIO to ensure every SIO at least has one voting point.Then a trained CNN model was used to assign functional attributes to those voting points.Then a statistical method was involved to count the frequency of the classified voting points of different functional attributes in each basic functional zone units.The functional attribute with the highest frequency is assigned to the basic functional zone units.The voting process has corrected the misclassification results of CNN to some extent.Urban functional zone classification inevitably involves the modifiable area unit problem(MAUP).Different spatial will be demanded for different purposes and different researches,and the dealing process of urban functional zone division also contains different process parameters.This paper explored the scale effect of the SIO on the final functional zone classification result from two aspects,spatial scale of SIO and the sampling window size of CNN model.In this paper,scaling problems also have been discussed,and it has been proved that coarse urban functional zone division results can be obtained by scaling-up fine scale division result.What’s more,the scale effects of different urban functional zones have also been discussed along with the optimal spatial resolution of experimental data.One conclusion has been drawn from this research,that is the spatial resolution of experimental data need to be higher than 2 meters per pixel.The complexity of functional area determines that the functional area contains many types of objects.So the accuracy evaluation method for urban functional zone is different from the traditional object-oriented image analysis method(OBIA).Aiming at this problem,this paper explores the performance of two different evaluation methods.They are point sample based and polygon sample based evaluation methods.And the experiment shows that the point sample based method is much more suitable for functional zone division evaluation.What’s more,the experiment also shows that the boundary of SIO-CNN method is relatively accurate,and the overall functional zone division accuracy reaches 91.09%. |
