Research And Application Of Water Body Extraction Method Based On Remote Sensing Image

The extraction of water body information plays a very important role in the investigation,monitoring,utilization,and protection of water resources,environmental monitoring,and the early warning and prevention of flood and waterlogging disasters.With the continuous progress of remote sensing technology,it has become more convenient and efficient to obtain water body information using remote sensing image.This thesis proposes water body extraction methods based on water body indices,support vector machines,and U-Net,and develops water body extraction application software based on these methods.The software is used to extract water body information from Landsat-8 remote sensing image at different periods.The main research content of this thesis is as follows:Firstly,the research area and data sources are introduced,and the basic parameters of Landsat-8 satellite,sensor,and remote sensing image are described.The remote sensing image are preprocessed using ENVI platform,including radiation calibration,elevation calculation,and atmospheric correction,which lays the foundation for the subsequent water body extraction.Secondly,water body information is extracted from remote sensing image based on the normalized water index method.Two methods,empirical threshold method and Otsu algorithm,are used to determine the classification threshold of water body,and the accuracy of the water body extraction model based on water body indices is evaluated based on the confusion matrix.The experiment shows that the water body extraction method based on water body indices can effectively extract water body information.Thirdly,support vector machine(SVM)is used as the water body classification model to extract water body information from remote sensing image.SVM is used to train water body classification on remote sensing image with different band combinations.The model classification accuracy is calculated based on validation samples,and compared with the water body index method.The results show that SVM has good performance in water body extraction from remote sensing image and has better applicability than the water body index method because it does not require threshold calculation and is less restricted by the features of objects in the image.Then,the U-Net deep learning network is used to construct a water body extraction model.After introducing artificial intelligence,deep learning,convolutional neural network,and U-Net network structure,the process and idea of using U-Net for water body extraction are given.A dataset for U-Net training is constructed based on transfer learning.The training hyperparameters of U-Net are set,and the training,testing,and accuracy evaluation of U-Net for water body extraction are carried out.It is found that the accuracy of U-Net water body extraction model is better than the other two water body extraction methods.Finally,the U-Net water body extraction model with the best performance was selected for application,and water body extraction application software was constructed with the completed U-Net water body extraction model as the core.This software is used to identify water information in Landsat-8 remote sensing image at different times,analyze and evaluate the generalization ability of the U-Net water body extraction model,and analyze the changes of water body in different periods based on the extracted water body.