Research On Urban Land Use Type Recognition Based On Taxi Trajectory Data

The rapid development of communication technology,the increasing popularity of mobile devices and the increasing function of devices have led to an exponential growth of digital footprints in urban environment.The ubiquitous big data(such as taxi trajectory data,mobile phone data and social media data)provide new data sources for scientific research,urban planning,government decision-making and resource allocation.And its data validity has been widely recognized.In urban studies,researchers can quickly grasp the urban dynamics and identify land use types from the perspective of social functions,because big data records the interaction between human and urban environment.Considering the temporal variations of human activities in different city areas,many scholars identified urban land use types by constructing time series of human activities.However,most time series in current researches were constructed based on one human activity feature(such as the time series based on hourly calling volume).Integrating multiple human activity features can help to better describe human activities and provide more information for land use type recognition.Therefore,this study proposed a method of urban land use type recognition based on fuzzy c-means clustering algorithm and fuzzy comprehensive evaluation model.It integrated the outflow,inflow,net flow and net flow ratio features extracted from the taxi trajectory data of Nanjing and carried out land use type recognition.And the effectiveness of this method had been verified.The main contents and conclusions of this paper are as follows:(1)Extract human activity features and construct time series.Extracting human activity features is the basis of constructing time series,and the selection of features will directly affect the results of land use classification.By referring to the studies of urban land use type identification based on taxi trajectory data,four commonly used features(the outflow,inflow,net flow and net flow ratio features)were extracted.The construction of time series involves the selection of data aggregation methods and the determination of time intervals.By referring to relevant literatures,the weekday-weekend aggregation method and the 1-hour interval were used to construct time series for each human activity feature.(2)Cluster time series and construct the membership matrix.The method of land use type recognition proposed in this study introduced the fuzzy theory.The time series of each feature were input into the Fuzzy C-Means Clustering Method(FCM)to obtain the clustering centers and membership degree.By matching the clustering centers with land use type centers,the membership degree in the clustering results can represent the membership of the unclassified region to different land use types.Using the membership degree in the clustering results of different human activity features,a membership matrix can be constructed for each region.(3)Determine land use types and weights of features.The membership matrix was used as the fuzzy comprehensive evaluation matrix to obtain the evaluation results based on the fuzzy comprehensive evaluation model,and the land use type of each unclassified region was determined according to the principle of maximum membership degree.The weight of each feature in this model was determined by the training process,and the objective function was to minimize the error rate of the training set.The results of land use type recognition in Nanjing showed that the framework based on the fuzzy c-means clustering algorithm and the fuzzy comprehensive evaluation model can integrate human activity characteristics and effectively improve the accuracy of land use type recognition.By combining different features and compared the results of land use type recognition,the combination of the outflow,inflow,net flow and net flow ratio features was found to have the highest accuracy(the overall accuracy is 0.858,the Kappa coefficient is 0.810),and the confusion of land use types was the lowest.Compared with the method of combining inflow and outflow time series,the overall accuracy of our method was improved by 0.138,the Kappa coefficient was increased by0.178.The producer's accuracy of each land use type was higher than 0.735,and the user's accuracy was higher than 0.700.Therefore,this study proposed an effective land use type recognition method,which provided a new idea for land use type recognition based on big data.