Study On Recognition Methods Of Ground Vegetation Based On Multi-scales Remote Sensing Data

The information of the planting structure,growth and distribution of vegetation has great significance for guiding the country to formulate relevant agricultural policies.Remote sensing technology can achieve rapid acquisition of agricultural information.,which has obvious advantages to be applied in the extraction of vegetation information.Three main types of remote sensing data are widely used:aerospace remote sensing data,aerial remote sensing data,and unmanned aerial vehicle(UAV)remote sensing data.Based on the image features of GF-1 satellite and UAV,this thesis analyzes the relevant characteristics of vegetation classification and fully utilizes the characteristics of the two data.Classification methods of ground vegetation that meet the requirements of multi-scales agricultural applications has been designed,and the ideal identification accuracy is achieved.The main research contents are as follows:(1)Using the image of UAV as the main data source,a set of ground vegetation identification methods under small-scale regions have been designed.Although the spatial resolution of the UAV image can reach centimeter-level,it only contains three bands of red(R),green(G)and blue(B).The UAV image lacks the near-infrared(NIR)band,which has obvious spectral response characteristics of green vegetation.After analyzing the characteristics of GF-1 satellite data and UAV data,this thesis proposes an image fusion method to form a new image,which maintains centimeter-level spatial resolution of UAV data.and the spectral information is consistent with satellite data.Meanwhile,the original UAV image and the fused image are classified by support vector machine(SVM),K-nearest neighbor(KNN)and principal component analysis(PCA)respectively.The experimental results show that the classification accuracy of vegetation based on the fused image integrated into the NIR band is significantly improved.(2)Using the image of GF-1 satellite as the main data source,a set of ground vegetation identification methods under large-scale regions have been designed.In China,the planting methods of intercropping are used.However,the satellite data commonly used is difficult to sample directly from the image due to low spatial resolution,so that the sample-based supervised classification methods cannot be used for identification of scattered vegetation.This thesis proposes a method of data-assisted sampling of UAV to train classifiers,which solves the problem that satellite data cannot be classified by supervised classification when vegetation planting area is small and scattered.Two new verification methods of classification accuracy with UAV image have also been proposed.In this thesis,the main vegetation types in the farming area have been identified by object-oriented classification method.The experimental results show that compared with the traditional sampling methods,the proposed method not only is efficient but also can achieve higher classification accuracy.(3)Different scales are combined with each other to form a multi-scales identification method of ground vegetation.In the study of small-scale areas,the satellite data at large-scale is used to enhance the spectral information of the UAV data.In the study of large-scale areas,the spatial information of the UAV data at small-scale is used to do auxiliary sampling.Multi-scales identification methods of ground vegetation can not only meet the application requirements of high-precision agricultural,such as internal area calculation at small-scale,the information of vegetation distribution at large-scale can also be extracted.