顾及多模态数据特征关联的城市功能区识别方法——以武汉市为例

A method for identifying urban functional areas considering the correlation of multimodal data feature:A case study of Wuhan

城市功能区在城市化建设当中扮演着重要的角色,城市功能区建模与分析对于规划城市土地布局和优化土地利用模式有着重要意义。目前在城市功能区建模与分析的研究中,仍存在多模态数据利用率低、关联特征损失大的问题,这导致难以精确捕捉城市空间结构信息。因此,本研究以武汉市中心区域为例,基于街景影像数据和兴趣点数据构建城市功能区识别方法。首先,将研究区域划分为500 m×500 m的格网单元,运用自然语言模型、深度学习、空间插值等方法来提取多维度的城市功能特征;其次,利用核典型相关分析方法提取多模态数据的关联特征,分析不同共享空间维度和相关性的关系;最后,基于建成环境特征、社会经济语义特征、拼接特征及关联特征的多维度特征构建随机森林功能区识别模型,并分别与两种常用数据模态进行对比评价。结果表明,本方法相较于没有顾及关联特征的模型准确度提高了3.4%。研究成果可为政府决策规划、城市发展评估提供帮助,推动城市化建设。

Urban functional zones play an important role in urbanization,and modeling and analyzing them are vital for planning urban land layout and optimizing land use patterns.However,traditional research often relies on historical planning data and administrative units,hindering the accurate representation of modern urban information and functional structure.Multimodal data,a frontier in geographic information science,offers advantages like large volume,strong present situation,multimodal,low cost,covering detailed and comprehensive urban spatial-temporal information.Despite these benefits,challenges like low utilization and significant loss of associated features persist.This paper addresses these challenges by proposing a method for identifying urban functional areas that considers the correlation of multimodal data features.This paper takes Wuhan city center area as an example and constructs an urban functional area recognition framework based on street view image data and interest point data.First,the study area is divided into 500×500m grid cells,and methods such as FCN image semantic segmentation,Place2Vec natural language model,IDW spatial interpolation,and others are used to extract multidimensional urban functional features.Validation is done through methods like POI richness,cluster analysis,DC coefficient curves,and spatial distribution of scene elements.Second,nuclear typical correlation analysis is used to extract correlation features of multimodal data and analyze the relationships between different shared spatial dimensions and correlation.Finally,a random forest functional area identification model is constructed,incorporating built environment features,socio-economic semantic features,splicing features,and correlation features.The results are analyzed using methods such as confusion matrices,feature importance score calculations,and real maps.In the experiment of extracting multi-dimensional urban functional features in this paper,the clustering results and POI richness are compared and analyzed with the real map,and found to be basically consistent with the actual situation.through the DC coefficient curves and the spatial distribution of scene elements,it can be seen that the spatial interpolation results in this paper are good,and the results of the semantic segmentation basically conform to the actual situation.In the experiment of extraction and analysis of associated features,through the visual display of related variables,it can be found that there are obvious differences before and after the fusion of the two kinds of data.Through the study of shared spatial dimensions and correlation,30 dimensions are finally chosen as the shared feature spatial dimension.In the functional area identification model experiment,through the comparative analysis of multiple models constructed by different features,it can be clearly found that multimodal data has a driving effect on the improvement of identification accuracy compared with single data.On this basis,the input of associated features further improves the recognition accuracy by 3.4%and achieves the highest score in the feature importance score calculation.In the confusion matrix of the recognition results,it is found that the accuracies of the five site types are all above 70%.The full exploitation and use of multimodal data are of great significance for the modeling and analysis of urban functional zones.In this paper,we constructed an urban functional area identification method based on the KCCA data fusion method,which takes into account the feature association of multimodal data.The main conclusions from the experimental analysis are as follows:(1)The semantic features extracted in this paper can accurately express urban functions.The results of multi-method validation experiments fully prove the validity of the socio-economic semantic feature extraction method in this paper.It also shows the feasibility of extracting socio-economic semantic features using the Place2vec model,and its potential semantics have some correlation with the urban functions.In addition to verifying the validity of the semantic feature extraction method of the urban scene in this paper,it is found that in the distribution of the elements,the semantic features of human-induced and human-induced scenes account for the highest proportion of the six elements.Among the six highest elements,the proportion of man-made features and natural features is balanced,which also shows the importance of natural features in the development of urbanization.(2)The KCCA data fusion method can improve the performance of the urban functional area identification framework.The results show that data splicing can improve the recognition accuracy compared with single feature,while multi-feature data fusion,i.e.,the method in this paper,can further improve the recognition accuracy compared with data splicing.This fully proves that the KCCA data fusion method has better performance than pure data splicing and also has a better ability to express the functional characteristics of the city.In addition,the data samples used in this paper have the problem of uneven distribution,but the recognition accuracy of the five land use types is more consistent,indicating that the framework of this paper has enough adaptability to the problem of uneven distribution of data samples.