Classification Study Of Desert Steppe Species Based On UAV Hyperspectral Remote Sensing

In recent years,under the influence of climate change and human activities,grassland degradation is very serious,threatening ecological balance,causing grassland production reduction,sandstorms and other disasters.The change of dominant species in grassland ecosystems is a significant feature of grassland degradation,and some species have important implications in the process of grassland degradation.Therefore,rapid,non-destructive and large-area monitoring of grassland species is conducive to the correct judgment of grassland degradation degree,and is of great significance for grassland ecological environment management and grassland animal husbandry production.At present,due to the limitation of spatial resolution,grassland monitoring based on satellite remote sensing cannot reflect the structure of grassland species.Grassland monitoring based on field surveys is time-consuming and laborious,and it is difficult to meet regional monitoring needs.With the rapid development of unmanned aerial vehicle(UAV)and hyperspectral imaging technology,it provides a new method and technical basis for solving regional grassland species classification.In this study,the Inner Mongolia desert steppe was taken as the research object.The multi-rotor UAV was used as the remote sensing platform.The hyperspectral imager was installed on the UAV to collect the grassland hyperspectral remote sensing image during the grassland vegetation growth period.In view of the weak spectral difference of grassland species,it is difficult to directly use the spectrum to distinguish species.In this study,the spectral transformation method was used to increase the spectral gap of grassland species.On this basis,the vegetation index was established,and the classification threshold of the spectral transformation vegetation index was determined by the maximum inter-class variance method.The classification results show that the classification method based on spectral transformation vegetation index effectively extracts the spectral characteristics of grassland species,and the classification method is simple and feasible.In view of the data redundancy of spectral dimension inherent in hyperspectral imagery,considering the correlation of the hyperspectral bands,the effective information content of the bands and the separability of objects,a stepwise method was proposed to select the characteristic bands,which realizes the feature spectrum extraction and reduces spectral dimension.The hyperspectral images represented by the feature bands were used as the input data of the deep convolutional neural network(CNN).The spatial features of the hyperspectral image were further extracted by CNN.In view of the nonlinear problem of hyperspectral data,the ReLU activation function was used to increase the nonlinear expression ability of CNN.In view of the over-fitting phenomenon,the L2 regularization was introduced on the loss function,and Dropout was used in the fully connected layer.These measures effectively prevent over-fitting.Compared with the 4 classification methods,the classification accuracy based on the feature band and CNN classification method is about 96.67%,and the Kappa coefficient is 0.95,which is higher than other classification algorithms.Combining the classification method based on characteristic band and CNN with grassland vegetation phenology,and making full use of the different spectral and spatial characteristics of grassland species phenological period,the best time for hyperspectral remote sensing classification of different species in Inner Mongolia desert steppe is obtained.In this study,the hyperspectral image of the desert steppe in Inner Mongolia was obtained by the UAV hyperspectral remote sensing platform,and the classification model of desert steppe species was established based on the characteristic band and CNN.The research results provide some data support for species coverage estimation,degeneration succession judgment and pasture management of desert steppe in Inner Mongolia,and provide research basis for ecological restoration of desert steppe.