Abstract: In this study, the high-resolution remote sensing image of Vaihingen City in Germany is used as the data source, proposes a extraction method combining multi-scale guided filtering feature and kernel principal component analysis feature. The multi-scale guide filter was used to extract the characteristics of different scales of green space. The kernel principal component analysis algorithm with nonlinear mapping ability was used to reduce the dimension of multi-scale features. Finally, the dimension-reduced features were input into SVM classifier to get the classification results of urban green space. The experiment was compared with the previous green space extraction methods. The results show that this method can make full use of the spatial neighbourhood information classification, get higher classification accuracy than the baseline methods, and significantly reduce the fragmentation, which is often produced by the traditional pixel-level classification methods.