Research On Semi-automatic Extraction Of Water Distribution Information Based On Hyperspectral Remote Sensing Images

Surface water,as the most important terrestrial resource,profoundly affects the balance of the ecosystem and the development of social economy.The real-time,rapid acquisition and automatic identification of water body information are of great significance to water resources surveys,flood monitoring and river basin remediation.Hyperspectral remote sensing image is rich in information,has high spectral resolution and wide band range,and has the ability of more accurate identification of ground feature information,which provides the possibility for dynamic monitoring and automatic extraction of surface water in real time and accurately.The paper is based on the ”Zhuhai-1” hyperspectral satellite(Orbita Hyper Spectral,OHS)image,and take four different backgrounds and different types of water bodies in Anshan City,Funing County of Qinhuangdao City,Anyi County of Nanchang City,and Xiangzhou District of Zhuhai City as the research objects.The main content and conclusions are as follows:(1)The characteristics of the integration of spatial information and spectral information of OHS hyperspectral image data are introduced,and the OHS hyperspectral data is preprocessed,including data reading,radiation correction,atmospheric correction and orthorectification.The cosine distance method is used to calculate the distance between the spectral vector of the designated water pixel and the spectral vector of the pixel to be classified,and the appropriate threshold value is set to automatically and quickly obtain the classification label needed for supervised classification.(2)Combine the water extraction results of the support vector machine(SVM)that only considers spectral information and the variant full convolutional neural network(VFCN)that only considers spatial information to obtain more accurate surface water information.(3)Quantitative comparisons were made between the research method of this article and the water index method(WI),variant full convolutional neural network and support vector machine,and found that in the four research scenarios,the quantitative evaluation parameter fluctuation of the method proposed in this paper is the smallest,that is,the research method in this paper is less affected by the ground features and the nature of the water body.Moreover,the extraction effect of the proposed method is better than that of the other three methods.The overall accuracy of the proposed method is all above 98.822%,and the Kappa coefficient is all above 0.902.This paper proposes a supervised learning method for semi-automatic surface water extraction,which realizes real-time,fast and accurate monitoring and analysis of inland surface water,and provides a basis for mapping inland surface water in China.