The Study Of Data Quality Assessment For OpenStreetMap City Buildings

With the rapid development and popularization of geographic information technology,reliablity of spatial data is important for decision-making and applications.However,crowd-sourced geographic information data are mainly provided by some nonprofessional contributors.They can't make sure the quality of provided data.OpenStreetMap(OSM)is one of the most representative projects in crowd-source geographic data,which has a lot of users.Building is a common spatial element in OSM,and its expression is complex.Therefore,this paper takes OSM building data as the research object to conduct in-depth research on crowd-source geographic data quality.This paper evaluates OSM building data quality quantitatively.The method researches on how to evaluate the OSM based on official vector data and how to evaluate the OSM based on high resolution remote sensing data,separately.The main contents include the following three aspects:1.Building Polygon Similarity Measurement Model Considering Spatial Distribution and Geometric TransformationBuildings are usually represented as polygons in maps.To calculate the shape accuracy during assessing the quality of OSM,different similarity measurement models are established according to different polygons.First,for simple polygons,the farthest point describing function is proposed,and geometric invariant descriptors are extracted by Fourier transform;Then,for polygons with holes,based on similarity measurement of simple polygons,the concept of position graph is introduced to describe the spatial distribution of holes,and various translations are defined;At last,for multi-polygon,based on the similarity measure of simple polygon,the matching control subgraph is defined by complexity,neighborhood support degree and similarity of simple subgraph.The matching position graph is constructed based on matching control subgraph,and the polygon vectorization description is realized by combining local moment variable based on convex hull.2.Comprehensive Evaluation Method of OSM Building Data Based on Deep Autoencoding NetworkIf there are official vector reference data in the study area,aiming at the problems of incomplete evaluation factors and subjectivity of weights during assessing OSM data quality,this paper designs a deep auto-encoder network.A variety of quality evaluation indicators including shape accuracy,data integrity,position accuracy,direction consistency and semantic accuracy are used as the input of the network.The unsupervised learning method is used to evaluate the data quality.The encoding-decoding reconstruction error of the model is taken as the quality evaluation result.The influence of artificial weights is weakened.By using artificial intelligence method,the data quality can be evaluated objectively,synthetically and quantitatively.The model has strong generalization ability and can be applied to other data quality assessment or abnormal data analysis.3.Evaluation Method of OSM Building Data Integrity and Location Accuracy Based on Multi-Task Convolutional Neural NetworkWhen the official standard vector reference data is limited in some regions,this work builds a quality evaluation model of OSM building data based on the extraction results of high-resolution remote sensing images.First,because building scale is diversity in high resolution remote sensing images,a multi-task Res-U-Net depth convolution neural network is designed to solve the problem.Then,guided filtering method is introduced to optimize the classification results and remove salt and pepper noise.At last,the building footprints are extracted by masking method,which are treated as the reference data for evaluating OSM buildings.The data integrity and location accuracy are calculated to realize the data quality evaluation of OSM buildings.This paper can provide factual basis for the development of volunteer geographic information,so that related projects can be better developed,and guide data collection methods.To some extent,the extracted buildings from remote sensing can supplement the lack of OSM data.This paper creatively applies the related technology and methods of artificial intelligence to evaluate the quality of OSM building data,and combines remote sensing data with geographic information data,which provides new ideas for the follow-up research.