Polygon Entity Matching For Multi-represented Vector Data Based On PSO Neural Network With Application Of Data Updating

As the stage named "primitive accumulation" of fundamental geospatial data ends and the rapid development and wide application of Geographic Information System (GIS), the up-to-date geospatial data is becoming more and more significant and data updating has been a bottleneck of sustainable development of GIS. Therefore, the study of data updating is one of the hotspots and difficulties in GIS. Multi-represented geospatial datasets are multiple representations of the same geospatial phenomena or entities in the real world to meet the different requirements, which used different geometries, topology structures, data precision, attribute integrity, classifications and semantic representations, and spatial relations. These differences make the conflation and updating of multi-represented geospatial data difficult. The entity matching are used to identify the corresponding objects representing the same phenomena or entities in the real world to conflate and/or update the multi-represented datasets, in order to get more accurate, more detailed, and newer geospatial data.A polygon entity matching method for multi-represented vector Data based on neural network is proposed in this paper on the basis of the analysis of the computational model, indices, and causes of spatial similarity and the multi-represented characteristics of geographic entities. Moreover, a propagating update mode is provided based on the proposed polygon entity matching method to keep updated propagating among the multi-represented datasets.Firstly, unified irregular fundamental grids are built according to the existed spatial indexes, into which hierarchy theory and urban morphology are led. The potential matching set is obtained quickly and accurately according to the grid index accompanied by the spatial relation between the Minimum Bounding Rectangles (MBRs) of the matching objects, which not only avoids the complicated overlapped analysis but also narrows the objects in potential matching sets to the minimum.Secondly, the process of matching includes rough matching and accurate matching other than obtaining potential matching set according to the geometrical and multi-represented characteristics of vector data. The rough matching is to determine the rough matched pairs by comparing the ratio of overlapping area to object area with a specific threshold. In addition, the rough matched pairs will be combined if there are intersections between them. Therefore, the matching types of1:N (N>1), M:1(M>1), and M:N (M>1, N>1) are transformed into the matching type of1:1. Furthermore, the intelligent learning is bring into the accurate matching, that is, particle swarm optimization (PSO) and ant colony optimization are used to training the artificial neural network (ANN) to get the connecting structure and weights, which is substituted to empirical determination of indices weights. In addition, the strategy of bidirectional matching is utilized to obtain the maximum matched pairs on the premise of keeping the precision, i.e., backward matching finds matched pairs to supplement the matched pairs in forward matching.In addition, a correlative multi-represented data model is proposed aiming at promoting existing multi-represented data models. There are correlations among the multi-temporal datasets, multi-scale datasets, and adjacent different temporal and different scale datasets. Furthermore, a propagating update mode is discussed based on entity matching. The variation objects can be located rapidly and the updates can be propagated among the multi-represented datasets according to the correlations among these multi-represented datasets.Finally, experiments are taken to confirm the effectiveness of the proposed methods of entity matching and data propagating update. The results of three group experiments (different types of datasets of the same time, different scales of the same type of datasets, and different temporal datasets with different types) indicate that the same index may have different weight in the different data context. It also reveals that the proposed entity matching method has a higher accuracy in all the three group entity matching when compared experiments with the weighted mean method given by previous literature. In addition, the updates can be propagated from the building data to the1:1000electronic map and the1:10000electronic map in the data update experiments, which verify that the updates can be propagated among multi-represented datasets.