Abstract: Support vector machines use a single kernel function for impervious surface extraction that produces high time complexity and low extraction accuracy.In order to solve the above problems, this paper introduces polynomial kernel functions on the basis of radial basis kernel functions and proposes an impervious surface extraction method with mixed kernel functions. Firstly,since the features with different properties have similar spectral information, this paper combines the spectral information with the image entropy texture information in the process of feature extraction. It is possible to distinguish between the categories of things more clearly. Then,on the basis of the radial basis kernel function, polynomial kernels are introduced to obtain the feature information of images from local and global perspectives respectively, and improves the extraction accuracy of impervious surfaces. Finally,based on the results of impervious surface extraction, the spatio-temporal evolution analysis is carried out.In this paper, Landsat images from 2009 to 2021 in the main urban area of Fuxin city are used for experiments.The experimental results show that the combination of spectrum and entropy texture can improve the feature extraction effect and extraction accuracy of impervious surface.Compared with the single kernel function extraction method, the impervious surface extraction accuracy of the proposed method is improved by 2.5%, indicating the effectiveness of the proposed method.

Key words: impervious surface, texture features, support vector machine, mixed kernel functions, spatio-temporal analysis