Research On The Quantization Of Terrain Texture Based On Gray Level Co-occurrence Matrix

Topography formation is the result of the internal and external force synergy in the earth's crust,and the differences of topography and landform are complex and diverse in terrain texture,so the comprehensive analysis and quantitative extraction of terrain texture is very important for the study of crustal movement and geomorphic evolution.The calculation of spatial variation characteristic of various micro terrain factors and exploring indirectly the change rule of terrain are main methods which are good at expression of local feature,but not at macro grasp of large scale region terrain in present researches.Because of lack of breakthrough progress in effective analysis of macroscopic terrain in current researches,the texture analysis technology in computer vision field is taken as the breakthrough point in this paper,combing texture analysis technology based on GLCM with computer image processing technology and geological research ideas,then we introduced it into the research field of digital terrain analysis.We quantitatively extracted the terrain texture with a way based on gray level co-occurrence matrix,it can effectively reveal the macroscopic features of landform in a certain extent.Thus,we believed the researches including quantitative analysis of topographic features and terrain recognition recognition classification can make a breakthrough in this way.In this paper,the main contents and conclusions are as follows:(1)we analyzed the construction factors of the gray level co-occurrence matrix how to effect the texture parameters synthetically.Took dem raster data as an example,we analyzed the influence of peer-to-peer distance and angle direction changes on 11 parameter characteristics.Finally we got the more suitable result when distance equals 4 grid cell,however,most of the parameters characteristics changing with four Angle directions is not obvious.So we took the mean of four directions as value of data in quantitative analysis.(2)The quantification of terrain texture is inevitably affected by the resolution scale of the data,so in this paper we built multiscale series data of different resolutions of sample area,calculating 8 parameter eigenvalues of each scale of DEM data,hillshade data,slope data and curvature data changing by the various resolutions:the parameter eigenvalues of hillshade data are most sensitive to the changing of DEM resolution,while the VAR eigenvalue is the most obvious in all parameters.(3)We retained 8 parameters characteristics(ASM,CON,COR,IDM,ENT,VAR,SumAverg,DifEntrp)through analyzing distinguishing ability on landform types from four data of eight sample data respectively.Owning to redundance and repeated description of parameters characteristics,in this paper the method of PCA is used to classify the parameters reasonably:COR parameter is selected to detect the direction;ENT,ASM and IDM parameter are used to quantify terrain complexity,and CON,SumAverg and DifEntrp parameter are used to quantify terrain periodicity.Finally,we proposed comprehensive quantitative indicators for analysis and concluded the indicators from the gray level co-occurrence matrix have certain ability to distinguish the landform types after comprehensive analysis of terrain texture.