Research On Hierarchical Subclassification Based On Shape And Proximity Relation Of Water

The self-developed30m Global Land Cover remotely sensed products have thehighest spatial resolution in the six existing land cover products,which fills the gaps inthe fields of remote sensing mapping procucts.At present,water products of GLC datahave been finished,with the spatial distribution and data analysis of water dataobtaining preliminary outcome.As to our GLC products,there is only one land typecode in the attribute table,which can distinguish different land types,such as water andwetland,but can not distinguish different land feature in one land type,such as YangtzeRiver and Poyang Lake.Therefore,it is essential to subclassify every single land typeof remotely sensed products.This paper focuses on subclassification research on GLCwater data,and further discusses how to attach different attributes to classification datawith spatial information,but without spectral information.This paper put forward three filtering basis to hierarchically filter water dataselectively，considering priori knowledge,proximity relation and water shape.First ofall,select whole river data and part lake data of GLC water product according to riverand lake reference data of global geomatics database,and filter another part lakedata,based on spatial relationship between reference data and water data.Secondly,it isnecessary to convert polygon into point considering that calculating distance andangle inside polygons.Filtering scattered river polygons according to proximityamong points with part river data and lake data from step one left.It is important thatdistance between adjacent river points is shorter than that between river point and lakepoint within searching radius along river trend.This is mathematics basis of proximityfiltering.Finally,we separate river polygons from lake polygons quantificationally inthe remaining water polygons based on shape difference,namely roundness.Theseparation of water polygons depends on the selection of threshold.It calls ArcEngine function database for secondary development in Python underPython Shell development environment,develops water subclassification automatically batch program based on three filtering condition and enhancescalculation efficiency immensely.Where program needs interactive operation isthreshold setting.It needs cycle criterion to decide whether running results arereasonable and separates river data from lake data finally.At last,the overall classification accuracy of river is97.40%,while lake is98.72%through grid overlay method.We can see that lake classification accuracy is higherthan that of river,which indicates method in this paper applies to lake subclassificationresearch.The results show that the grid size has influence on classification accuracy tosome extent.