Research On Geomorphology Recognition And Classification Based On Multi-Modal Data Fusion

The land surface consists of a variety of landforms,each of which is a natural spatial entity with different visual and physical characteristics.Landforms are formed during the evolution process of different types and intensities which has been continuously functioning on the earth surface.After several decades' qualitative and quantitative research,Chinese researchers have built their own landform classification system.In recent years,with the in-depth development of geomorphological quantitative analysis research,quantitative landform interpretation,with researches based on DEM model,directed by geomorphology theory,and with the help of technologies in different fields,like geographic information system(GIS),remote sensing(RS),DAT,has become an important branch of geomorphology quantitative research.The development of cutting-edge technologies,such as cloud computing,artificial intelligence(AI)and big data,has brought great opportunities and huge challenges to the geographic field.In particular,the AI technology represented by deep learning,has greatly promoted the intelligent and automatic development of landform interpretation.Through introducing the spatial cognition theory and AI technology into the geography field,this paper carries on theory research and model methods research on distinguishing the micro-fluctuation patterns of earth surface,complex landform patterns and similar landforms with the same cause attribution.Focusing on the landform recognition and classification in landform interpretation,this paper puts forward a fusion model integrating the physical cognitive features of landforms for interpreting,builds a multi-model landform recognition network model and designs a pyramid-based landform classification network model by using deep learning.The main research findings in this paper are:(1)The proposal of a multi-modal geomorphological variable classification system.This paper systematically analyzes the function of existing terrain factors in the process of landform interpretation.According to the essential features of the terrain factors,this paper proposes a multi-modal terrain factors classification system for the study of landform recognition and classification.In addition,by quantitative calculation and analysis on the relativity of abundant terrain factors,this paper verifies the rationality of the proposed multi-modal terrain factor classification system.(2)The establishment of a landform recognition model based on multi-modal landform data.Based on the study of the applicability of convolutional neural network in landform feature construction,this paper proposes a new landform recognition framework based on deep learning.The consistent deep learning framework is employed to effectively integrate the landform feature extraction network and landform recognition network in order to build an end-to-end landform recognition network.The rationality of the multi-modal feature extraction network and fusion strategy has been validated through a great number of experiments.(3)The establishment of a landform classification network based on pyramid model.This paper put forwards a pyramidal feature extraction and fusion network model to extract landform features suitable for landform classification and employ the spatial context information in an effective way.This model extracts multi-modal landform features effectively and greatly extends the spatial field of the deep convolutional network by fusing landform features of different scales based on pyramid model features.A large number of experiments have been conducted whose results show that the proposed method significantly improves the accuracy of landform classification,verifies the feasibility of deep learning in landform classification research,and realizes the depth integration of landform research and deep learning technology.(4)Research on the validation system.Based on the research of theory and related technology,this paper introduces the design idea of the prototype system,integrates all the research results into the prototype system,and verifies the correctness and practicability of the related algorithm through the application of the system.