Research On Extraction Of Gully Based On UAV Images Using Object-Oriented Method

The erosion gully has an extremely wide distribution in the Loess Plateau of China,and it is the most important type of erosional landform.The treatment of eroded gully is an important part of soil erosion control.The extraction methods of erosion gully mainly include:visual interpretation,based on pixel analysis method and object-oriented method.The first two methods are time-consuming and labor-intensive,limited by image accuracy,and have high requirements on the interpreter's work quality and work experience.The object-oriented method often only uses the spectral information of the image,lacks topographic information,and affects the classification accuracy.This paper combines high-resolution DEM and DOM image data produced by UAV images,combined with object-oriented image analysis method,using multi-level feature extraction to extract the erosion ditch of Zhifanggou in Ansai District,partially overcoming the traditional erosion ditch.The problem of time-consuming and laborious manual interpretation and topographical mapping of erosion grooves on topographic maps,and high classification accuracy.The main research contents and conclusions of the thesis are as follows:(1)The optimal segmentation scale parameters for the extraction of erosion ditch are determined.The optimal segmentation scale parameters,including shape parameters and compactness parameters,are obtained by the local variance method when the study area is used to extract the gully trench area and shallow trench area.And the most subdivided scale parameters.The optimal parameter combination of the topographic data of the gully area is(215-0.4-0.7),and the optimal parameter combination of the terrain and image data of the gully gully area is(187-0.4-0.6),and the topographic data of the shallow ditch area The optimal parameter combination is(152-0.5-0.7),and the optimal parameter combination of terrain and image data is(133-0.3-0.5).(2)Constructing the feature space of object-oriented classification:using the texture features,spectral features and geometric features of the R language segmentation results to of the feature space,and then for several major The variables are retained and then the performance is calculated.Through such repeated calculations until the most suitable variables are found,the feature space of object-oriented classification is constructed.Among them,there are 17 features of gully terrain data,18 features of terrain plus image data,and 10 features of shallow trench topography and image data.(3)Based on the object-oriented two-level method,the erosion ditch extraction is completed:based on the object-oriented classification feature space,the object-oriented classification idea is used to extract the gully gully area.The research results show that the terrain plus image data classification The accuracy is higher than the classification accuracy of terrain data.The random forest classification method has the highest accuracy in terrain and image data classification.The overall classification accuracy is 92.82%and the Kappa coefficient is 0.85.The shallow trench area was extracted by rule classification,and the user accuracy of the classification was 86.96%.(4)The slope and aspect characteristics of each eroded trench are analyzed:the slope and slope of each eroded trench are calculated,and the slope,slope and frequency curves are calculated.The results show that the slope statistical distribution is close to positive.State distribution,the statistical distribution of the aspect of the slope,the proportion of the sun slope is greater than the shady slope.