Research On Detection Of Spatiotemporal Object Inconsistency For Land Cover Update

Land cover is a complex of natural characteristics.Land cover spatial distribution and change over time play an important role in climate simulation,ecological assessment and geographical conditions monitoring.With the increasing development of the Earth observation technology,remotely sensed images have become the main data source for land cover information extraction.However,due to the spectral,texture,and temporal complexity and diversity,in addition,not all regions can be covered by temporal high-quality remotely sensed data,which result in multi-temporal land cover data inconsistency after updating.To ensure land cover update quality,in practice,land cover data quality checking gives priory to manual inspection,and partially automated detection,which is time-consuming and labor-intensive.Hence,an automated inconsistency detection method for land cover update is urgently needed.In geospatial data inconsistency detection,topographic vector data inconsistency detection has been paid much attention,while for raster classification data it is not.So,it is an important task to study raster data inconsistency detection for land cover update in academic and industrial societies.Taking the spatial inconsistency detection towards topographic feature for reference,a spatiotemporal object inconsistency detection method for land cover update based on topological relationships in raster space is proposed in this paper,which includes topological relationships calculation in raster land cover data,spatial initial judgment rules construction based on confidence interval and spatiotemporal object inconsistency detection based on sequential topological chain matching.The main research and outcome of this essay are described briefly as follows.(1)A modern logic quantifiers-based method for topological relationships calculation in raster is proposed in this paper.The composition-based and whole-based spatial relationships models are summarized.And the usability of current models 4IM(4-intersect model)and 9IM(9-intersect model)for topological relationships calculation in raster land cover data is studied.The geometry complex of an object with a hole and the neighbor paradox for Jordan curve in boundary expression of land cover object are discussed.Based on universal quantifier and existential quantifier,a 3TQM(3-tuple quantifiers model)for describing topological relationships in raster space is proposed.In combination with morphology dilation and map algebra,a neighbor search algorithm for topological relationships calculation is designed,achieving the calculation of spatial relationships between a simplicial land cover object and a land cover object with a hole.(2)A confidence interval-based spatial initial judgment rules construction is proposed.As the complexity of land cover distribution and change over time,as well as the different classification system,checking rules for land cover misclassification is hardly obtained.Current methods for rules discovery are summarized,and statistical histograms for spatial relationships quantifying between pairs of land cover objects are constructed.A set algebra method for spatial initial judgment rules discovery is proposed.The rules are represented by spatial predicates and logical connectors,resolving the issue of checking rules absent for land cover misclassification.(3)A sequential topological chain matching method for spatiotemporal object inconsistency judgment is proposed.In comparison to topographic vector data inconsistency judgment,the result of land cover update inconsistency judgment by spatial initial rules is uncertain,and it is found that the sequential continuity of topological relationships for judgment of land cover object misclassified is indispensable.A matching technical framework for detecting inconsistent object is proposed based on topological relationship consistency constraint,geometry consistency constraint and semantic consistency constraint,as well as a sequential topological chain matching method is designed,achieving the posteriori judgment of object misclassified in update,and meanwhile eliminating the uncertainty in initial judgment by spatial initial judgment rules.(4)A system prototype for detection of spatiotemporal object inconsistency in land cover update is developed.The functional design and interface design are realized using Microsoft Visual Studio 2010 and Esir ArcEngine 10.2.According to the strategy of topological relationships calculation in raster,spatial initial judgment rules construction and sequential topological chain matching,taking six study areas(about 120000 square kilometers per area)in a 2015 update of GlobeLand30 for example,4 kinds of rules and more than 20 items are produced towards cultivated land,and land cover objects misclassified in update are labelled by points.By reference to remotely sensed images,the efficiency for inconsistency detection is checked manually,reaches 88.2%totally.After manual verification,through uniform point sampling(40x40)and statistical calculation,the overall data quality is improved by 10.3%,and the feasibility of the method is confirmed,which provide a maneuverability method for land cover update in large area to hold data quality.