Identification Of Desert Shrub Species Based On UAV Remote Sensing

In view of the traditional remote sensing classification method,the spectral characteristics of shrub-bare land and shrub-shrub are inaccurate in desert areas which are not suitable for the accuracy of desert shrub classification and extraction.This paper takes the transition zone between desert and desert grassland in the western of Ordos Plateau as the research area.The visible and near-infrared data collected by fixed wing eBee UAV and the visible light data of DJI UAV at flowering stage are processed to obtain high-resolution DOM images.Meanwhile,combining with the measured investigation of vegetation characteristics of field surface shrubs,the object-oriented artificial definition rules,Bayes,KNN,Cart and RF methods were used to identify and extract shrub species in desert areas in high resolution DOM images.Simultaneously,the applicability and classification accuracy of desert shrubs in different methods were evaluated.The main results are as follows:(1)In the desert,vegetation spots can be created by using the mean brightness ofall bands.(2)In the process of recognition,it is not only pay attention to the role of high resolution on classification.In particular,the spectrum of the ground object is complex and the vegetation difference is insignificant.It should be based on the morphological characteristics of shrubs(plant height,crown width and shape),reproductive characteristics(flowering,deciduous period),habitat characteristics(covering-sandy,desert),The morphological and spectral characteristics of the corresponding shrubs are extracted,combining the field survey data and the quantitative characteristics of plant species are counted.The vegetation categories are distinguished by threshold classification.Object-oriented UAV remote sensing technology has good applicability and application potential shrub recognition in desert.(3)The five methods have different degrees of misclassification in the identification,extraction and classification of desert shrubs.The method of artificial custom rules in grazing area and forbidden grazing area has the best effect.The classification results have a high degree of matching with the real ground objects.Moreover,the classification spots are consistent with the actual ground objects,but it dependent on professional knowledge and field experience to a certain extent.In the other four machines learning classification methods,the misclassification is weakened to a certain extent,the effect and speed are not very different,and the classification accuracy is highly dependent on the selection of training samples.The Kappa coefficient of custom rule in grazing area is 0.74,which is 0.38,0.34,0.16 and 0.09 higher than Bayes,Cart,RF,and KNN classification,respectively.The Kappa coefficient of custom rule in forbidden pastoral area is 0.79,which is 0.51,0.3,0.33 and 0.25 higher than Bayes,KNN,Cart and RF classification,respectively.In desert,random forest classification method is superior to other classification methods and the overall accuracy is 95%,which is 45%,9%,9%and 4%higher than KNN,Bayes,custom rules and Cart classification,respectively.