Research On Identification And Detection Method Of Grassland Poisonous Grass Based On UAV Image

Grassland ecological security plays an important role in regional economic development and social stability.Grassland ecological protection is also a key content of my country’s land resource management.In recent years,due to the contradiction between people’s increasing material needs and productivity,people have carried out predatory development of grasslands,and the phenomenon of overloaded grazing and blind reclamation has become serious.This led to excessive consumption of the forage grasses of the original dominant species in the grassland,and weeds and poisonous grasses obtained more growth space and caused flooding.Among them,Aconitum leucostomum was the most severe and widespread.The proliferation of grassland poisonous grass Diphtheria aconitum leucostomum further threatens the growth of fine pastures,and occasional livestock ingestion occurs.Its proliferation seriously affects the development of grassland animal husbandry as well as ecological and food safety.Xinjiang,as a major animal husbandry province in my country,has suffered heavy economic losses in the face of the spread of diphtheria aconitum leucostomum.Therefore,the control of grassland poisonous grass diphtheria aconitum leucostomum has a long way to go.Diphtheria aconitum leucostomum and common pasture have a strong spectral similarity in the visible light range,which will cause a lot of confusion when the image classification method based on the spectral information of the pixel is used to classify the two.Based on the spectral characteristics,morphological characteristics and texture characteristics of diphtheria aconitum leucostomum,three methods are used to identify and detect diphtheria aconitum leucostomum.First of all,we identify based on its spectral characteristics,and observe that the canopy of diphtheria aconitum leucostomum is whitish,and the reflected visible light radiation intensity is much greater than that of ordinary forage grass.Therefore,the difference between the visible spectrum of the canopy of diphtheria aconitum leucostomum in the high-resolution UAV image and the visible spectrum of the forage is used to supervise and classify the distribution of diphtheria aconitum leucostomum.Furthermore,based on the morphological characteristics of diphtheria aconitum leucostomum,we found that the growth characteristics of diphtheria aconitum leucostomum are sporadicly distributed on the grassland,which is much taller than ordinary forage grass.From a distance,it looks like a bulge distributed on a flat grassland.on.Therefore,we propose the relative elevation threshold method,obtain the digital surface model of the study area through drone tilt photogrammetry,and then obtain the digital elevation model of the study area through calculation,and then The difference image is calculated,and the distribution of diphtheria aconitum leucostomum is obtained by setting the relative elevation threshold.Finally,according to the texture characteristics of diphtheria aconitum leucostomum,we propose an automatic classification method of diphtheria aconitum leucostomum and pasture based on high-resolution remote sensing images of convolutional neural network,in order to realize the accurate recognition of diphtheria aconitum leucostomum.UAV was used to obtain the 1cm spatial resolution UAV digital orthoimage of the diphtheria aconitum leucostomum hazard area,two types of training samples of diphtheria aconitum leucostomum and common forage were selected,and three models of VGG16,VGG19 and Res Net50 were used to slice the characteristics of the image Perform abstraction and learning to obtain the deep features of image slices,and then realize the classification and extraction of the two types of features.In this paper,in the identification and detection of grassland poisonous grass Diphtheria aconitum leucostomum,three methods are proposed,and a whole set of research content is completed from method demonstration,data collection,and final determination of relevant parameters.Using drone data to identify diphtheria aconitum leucostomum in the area provides more choices and more accurate methods.