Multi Kernel Based Support Vector Machine For Classification Of High-resolution Remote Sensing Images

With the rapid development of remote sensing technology,the images have significant improvements in spatial resolution,spectral resolution and time-resolved.These provide a huge number of macroscopic and real-time images for the observation of regional and global changes.Among them,researchers favor high spatial resolution images because of their high spatial resolution and strong ability of distinguish ground objects.The high-resolution can be more intuitive to reflect the reality of the surface,but it also provide more data,which increase the difficulty of accurate classification and object extraction.Support vector machine(SVM)is a popular method in the field of kernel machine learning.It can balance the complexity of the classification machine and the sample learning ability through the principle of structural risk minimization.This algorithm can avoid the phenomenon of “dimensionality disaster” and“over-learning”.The classification of high-resolution images shows excellent advantages.This paper analyzes the difficulties in accurate classification of high-resolution images,introduces and optimizes multi-scale-kernel support vector machines,meanwhile,uses intelligent optimization algorithms for parameter optimization to achieve fast and accurate classification of high-resolution images.The main research results include the following aspects:1.Through the analysis and research on high-resolution image features and common classification methods,it has found that the use of only spectral features for classification can cause confusion between features,while the flexible use of spatial features provided by highresolution images can complement the feature information and improve the accuracy of classification.There,the spectral features and texture features of the objects in the highresolution images both used for the learning and construction of the classification machine.2.Multi-kernel support vector machine is a kind of kernel machine with higher flexibility and efficiency.There are many ways about multi-kernel learning(MKL),after theoretical and experimental comparison,multi-scale kernel support vector machine is more flexible and interpretable compared with multi-kernel learning methods such as synthetic kernel and combined kernel.At the same time,the construction method of the original distance matrix can improved after learning the kernel function.Using Mahalanobis distance instead of the original Euclidean distance,a multi-scale-kernel support vector machine based on mahalanobis distance is constructed for the classification of high-resolution images.The kernel function owns better ability to learn globally.3.The classification performance of support vector machine is closely related to the selection of kernel function parameters.The parameters of multi-kernel support vector machine increase with the increase of kernel function,which increase the difficulty of parameter selection.Intelligent optimization algorithm provides a more efficient way to select the kernel parameters.Then,a dynamic differential evolutionary algorithm is used to determine the parameters.Finally,the high-resolution remote sensing images was classified successfully with high precision and efficiency.