城市道路的多特征多核SVM提取方法

针对高分辨率遥感影像中城市道路提取的复杂性及SVM的分类性能，提出了一种城市道路的多特征多核SVM提取方法。首先利用FCM算法将原始影像粗分为建成区和非建成区两类，剔除非建成区；然后根据分水岭分割算法分割建成区并提取分割对象的光谱特征与空间特征，以全局核函数和局部核函数加权组合的方式构建多核SVM对建成区进行二次分类，去除建成区中的建筑物等非道路信息；最后利用数学形态学处理，获得最终的道路提取结果。试验结果表明：文中所提方法能够较精确地提取城市道路信息，分类精度高于单核SVM提取及其他对比方法。

Multi-feature Multiple Kernels SVM-based Urban Road Extraction

For the complex nature of urban road extraction in high-resolution remote sensing image and the classification performance of SVM,a new urban road extraction method of multi-feature multiple kernels SVM-based is proposed.Firstly,the FCM algorithm is utilized to classify the original image into the built-up areas and non-built-up areas,and the non-built-up areas are removed.Then this paper segments the built-up areas using watershed segmentation algorithm,and extracts the image objects spectral features and spatial features. This paper uses the weighted sum approach to achieve multiple kernels SVM ( MKSVM ) by the way of combining the global kernel function and the local kernel function. The secondary classification of built-up areas are achieved using MKSVM and remove the non-road information.Finally,the final road information is processed by using mathematics morphology.The result of experiment shows that the proposed method can fairly well extract the urban road information,and the accuracy of classification is much higher than SVM and other extraction methods.