Identification Of Plant Species In Desert Steppe Based On UAV RGB Image

Accurately obtaining information on the spatial distribution of grassland plant species is an important basis for biodiversity monitoring,community reconstruction and ecological function maintenance in grassland ecosystems.However,there is currently a lack of effective technical means to obtain timely and accurate spatial distribution of plant species.Unmanned aerial vehicle remote sensing continues to evolve,enabling plant species monitoring.In order to build a remote sensing identification technology system for grassland plant species,this thesis takes Xilamuren desert grassland in Baotou City,Inner Mongolia as the research area,and sets up two kinds of sample plots: grazing area and enclosed area,using UAV RGB remote sensing data,combined with machine learning algorithms and orientation With the object image analysis technology,the remote sensing identification of desert steppe plant species was carried out at the pixel scale and the object scale respectively.The main findings are as follows:1.High-resolution UAV RGB remote sensing images can provide an effective data source for remote sensing identification of desert steppe plant species,and give full play to its advantages of high spatial resolution.2.The pixel-scale minimum distance,support vector machine,and random forest methods can all get the best classification results at the end of August in the grazing area,and their overall accuracies are 37.9%,76.7%,and 80.4%,respectively.The Kappa coefficient 0.3,0.6,and 0.7,respectively.The minimum distance in the enclosed area got the best classification result in mid-July.The overall accuracy was 31.1%,and the Kappa coefficient was 0.1.The support vector machine got the best classification accuracy at the end of August.58.7%,Kappa The coefficient is 0.3,while the random forest obtained the highest classification accuracy of 64.4% and Kappa coefficient of 0.4 at the end of July.3.Image segmentation is the most important step in the object scale.When the segmentation scale is 10,the image can well interpret the characteristic information of different species.Among the various characteristics,the index characteristic is the best.The plant species in the grazing area in mid-July The overall classification accuracy was the highest at 81.3%,with a Kappa coefficient of 0.7,while the enclosed area achieved the best classification accuracy of 50.4% at the end of August,with a Kappa coefficient of 0.3.4.In desert steppe plant species identification,random forest is superior to support vector machine and minimum distance.Among them,the identification accuracy of plant species based on random forest is higher than that of support vector machine,and the identification accuracy of support vector machine is higher than that of minimum distance.All in all,the identification results of plant species at the object scale are higher than those at the pixel scale.5.The multi-temporal image data of desert steppe can effectively identify plant species under the blue light feature,and the overall classification accuracy of plant species in the grazing area is higher than that in the enclosed area,53.3% and 50.4%,respectively.Finally,it is shown that the classification results of plant species of single-phase data are better than those of multi-temporal data.