Research On Multi-scale Classification Method Of Landforms

The production and life of human beings and the development of ecological civilization are closely related to landform.As the most direct expression of landform type,geomorphology is one of the important contents of geomorphology research.The classification of landform can make humans understand its spatial distribution more clearly and provide scientific basis for rational use of land resources and optimization of human activity pattern.Throughout the existing research results,there are two main shortcomings in the classification of landform.Firstly,the consideration of the feature factors involved in the classification is not comprehensive enough.The feature factors of landforms classification are mainly composed of DEM and its related topographic factors,texture features and structural information factors that can express geomorphological features,but most of the studies do not consider the structural information comprehensively.Secondly,the existing methods for automatic classification of landform are mostly based on a single scale,but the objects of landform are often of different sizes and span a specific spatial scale.However,landform objects of human interest often have multi-scale features,which restricts the accuracy of single scale classification.To address the above issues,the work conducted and the main conclusions obtained in the paper are as follows.(1)Using DEM with 30 m resolution as the base data,nine basic topographic variables involving elevation,slope of slope,hill shading,surface roughness,surface curvature,elevation of variation,slope,surface cutting depth and terrain relief are extracted from the typical sample areas of Loess Plateau,and screened quantitatively.Meanwhile,for the 3 macro topographic factors involved,elevation of variation,surface cutting depth and terrain relief,the average terrain relief is extracted from the window size of 3×3,4×4 and 33×33 as an example,and finally the optimal window size of 14×14 is determined by using the mean variation point method.(2)Used correlation as well as Sheffield entropy method,the terrain factors are quantitatively screened and combined,and the final terrain variables are elevation,slope,hill shading,surface curvature,terrain relief,slope of slope and elevation of variation.(3)Based on the topographic factors as the main feature factors and the texture features as the auxiliary factors(mainly elevation homogeneity,slope mean,slope entropy,mountain shading mean,mountain shading variance,mountain shading entropy,mountain shading correlation,mountain shading angular second order moment,mountain shading phase dissimilarity,mountain shading homogeneity,and mountain shading contrast),new ground element structure reflecting structural information is added The results show that the structural information factor of landform elements can improve the classification accuracy in landforms classification.(4)This paper proposes a comprehensive multi-scale classification method(IMC)for landform,which is composed of three steps: multi-scale segmentation,sequential screening by scale and multi-scale merging,taking into account scale spanning.Among them,sequential screening by scale is an iterative confirmation process based on multi-scale feature extraction and supervised classification,with small-scale(small size)priority and probability maximization as the criteria for the classified objects.he experimental results with a typical sample area of Loess Plateau show that IMC has good performance(overall accuracy reaches75.16%,Kappa coefficient is 0.71),is simple and reliable,and can be used for fine classification of landforms.In summary,this paper tries to realize the automatic classification of landform by using the proposed IMC method with good results,which provides more references for the classification method of landforms.This method is not only applicable to the classification of landforms,but also expected to be applied in the classification of traditional remote sensing images to solve similar problems.