Research On Remote Sensing Recognition And Classification Of Anthropogenic Soil And Water Loss Plots

Production and construction projects are one of the important causes of humaninduced soil erosion in the region.Soil and water conservation departments urgently need to strengthen the supervision of man-made soil erosion plots.At present,there are few researches on automatic identification and extraction of anthropogenic soil and water loss plots based on remote sensing images.Traditional monitoring methods of visual interpretation and field investigation have long cycle,low efficiency and high cost of manpower and material resources,which are difficult to meet the needs of production practice.In this paper,Neixiang County,Henan Province,which is rich in disturbance types and widely distributed in production and construction projects,is selected as the research area.Based on GF-6 and Sentinel-2 remote sensing images,a remote sensing classification system is constructed,including vegetation,water,cultivated land,impervious layer and man-made soil and water loss plots.Select the sample points of each land type in the study area,conduct statistical analysis on their spectral,shape and texture features,and obtain appropriate feature combinations.Multiple analysis methods were used to realize the remote sensing recognition,extraction and classification of man-made soil erosion land parcels.The main research contents and conclusions of the paper are as follows:(1)In terms of spectral characteristics,in the near infrared band range,the spectral differences of all kinds of ground objects are great,so the traditional characteristic indexes such as NDVI,NDWI and NDBI are selected to analyze their sensitivity to the manmade soil erosion plots.In terms of shape characteristics,the features of length-width ratio of different ground objects are obviously different.In terms of texture features,all kinds of ground objects have obvious differences in contrast.(2)Extraction of artificially soil and water loss plots by pixel method is based on Google Earth Engine cloud platform,which completes the rapid acquisition of Sentinel-2images and the selection of sample points in the platform,and selects features such as spectrum,texture and terrain factors on the basis of feature analysis,and the plots of anthropogenic soil erosion were extracted by random forest classifier.The extraction accuracy of soil erosion plots based on pixel classification method is more than 90%,which can well meet the extraction requirement of large-scale disturbance areas.(3)The object-oriented classification method based on multi-scale segmentation and the object-oriented classification method combined with vector segmentation are proposed to extract the artificial soil erosion plots..The result of nearest neighbor classification combined with vector segmentation is better than that of multi-scale segmentation.The user accuracy is 90.42%,and the production accuracy is 71.13%,which can better obtain the boundary information of man-made soil erosion plots.(4)A classification method of multi-source and time series remote sensing image fusion is proposed,which reduces the misclassification and omission in the extraction process of man-made soil erosion plots.Through the analysis of classification results,the user’s low accuracy is mainly reflected in the leakage of cultivated land and impervious layer.Through the NDVI time series data of cultivated land,the difference value of NDVI of cultivated land was calculated,and the pseudo disturbance area of cultivated land was eliminated by setting the threshold value.In the impervious layer,the RII index and BAI index were set to distinguish impervious layer from artificial soil and water loss plot,which effectively reduced the leakage of artificial soil and water loss plot in impervious layer.The experimental results show that the production accuracy is 80.2% and the user accuracy is 92.9%,which effectively solves the problems of misclassification and leakage between cultivated land and impervious layer and anthropogenic soil erosion plots.(5)For this study area,a classification system of anthropogenic soil and water loss plots under the background of soil and water loss was proposed.Anthropogenic soil and water loss plots were divided into four categories: excavation engineering,urban construction engineering,railway engineering,highway engineering and agricultural and forestry development and construction engineering.The characteristic index system of disturbance classification was constructed based on multi-source data and topographic features.Based on the classification rules of each production and construction project type,the threshold value was set for the experiment,and the overall accuracy of the classification results was higher than 82%.Taking the anthropogenic soil and water loss plot as the research object,the disturbance area is identified based on pixel and a variety of object-oriented analysis methods by making full use of the existing vector results.Combined with multi-source data and topographic factors,the types of the man-made soil and water loss plot are subdivided and applied to the supervision of soil and water conservation in production and construction projects.It provides solutions for relevant soil and water conservation departments and supervision departments.