Research On Grassland Drought Monitoring Based On Remote Sensing Image

Drought is a common natural disaster that has a great impact on human production and life.With the acceleration of global warming,drought events occur frequently,which seriously affects the healthy development of the global ecosystem.Grassland drought is one of the main causes of grassland degradation and desertification.Establishing a stable and accurate grassland drought monitoring model is helpful to grasp the drought situation in time and take corresponding measures.Satellite remote sensing images provide multi band spectral data,which makes it possible to realize large-scale dynamic drought monitoring in grassland.Therefore,this thesis takes Siziwang Banner in Inner Mongolia as the research area,uses Landsat8 satellite remote sensing data to extract grassland information,combines drought monitoring indicators and meteorological data,and adopts machine learning and depth learning methods to carry out grassland drought monitoring research based on remote sensing images.The main research contents and conclusions are as follows:(1)A grassland classification method for remote sensing images based on hybrid3D-2D convolutional neural network combined with vision Transformer(Hybrid CNN-Vi T)is proposed.Aiming at the problem of insufficient global feature extraction in CNN,Vi T is introduced into 3D-2D CNN,and on the basis of preserving the spatial spectral features of satellite remote sensing images by using 3D and 2D convolution kernels,the global sequence information is extracted by the multi-head attention mechanism of Vi T to improve the classification stability.Experiments compare 3D-2D CNN and RF methods,and the results show that the Hybrid CNN-Vi T method improves the grassland classification accuracy of remote sensing images and is an effective and stable grassland extraction method.(2)A grassland drought monitoring model based on machine learning is constructed.Firstly,NDVI,AVI,VCI,RVI,EVI,NDWI 6 vegetation indexes extracted from Hybrid CNN-Vi T in five grassland areas were calculated as drought monitoring indicators,and their applicability was determined by correlation analysis with Pa,so as to construct a data set of grassland drought monitoring indicators;Then,four machine learning methods of SVM,RF,KNN and XGboost are used to construct a grassland drought monitoring model with six drought monitoring indicators as independent variables and Pa as dependent variables;Finally,the accuracy and applicability of four grassland drought monitoring models are analyzed and compared.The results show that when the training set test set ratio is 7:3,the monitoring effect of the four models is the best,and the accuracy rate is above 85%.Among them,the XGboost grassland drought monitoring model has the highest accuracy and the best generalization performance,and its accuracy rate reaches 90.76%.(3)A grassland drought monitoring model based on GAN and transfer learning is proposed.Due to the limited data set,it is difficult to directly apply deep learning to grassland drought monitoring,so the model first uses GAN to pre-train the pre-training set constructed by drought monitoring indicators in four regions,including cultivated land and desert.Then,the initialized network is trained by transfer learning with the grassland drought monitoring index data set to obtain the grassland drought monitoring model.Finally,the model is compared with machine learning method and its applicability is analyzed.The results show that the grassland drought monitoring model based on GAN and transfer learning excels the machine learning methodwith an accuracy of 94.07%,revealing a higher degree of fitting and precision.(4)A grassland drought monitoring system based on remote sensing images is designed and implemented.The system uses Tkinter module of the Python platform to build a graphical interface,and its functions of remote sensing image preprocessing,grassland information extraction,grassland drought monitoring index extraction and grassland drought monitoring are implemented,which is convenient for the users to quickly and accurately monitor grassland drought based on remote sensing images.