Abstract: In order to improve the extraction of urban vegetation from high-resolution remote sensing image, an novel full convolutional neural network model was proposed. The best model parameters were obtained through a large amount of training data, and the vegetation information was extracted. The vegetation information extracted by support vector machine, object-oriented method and classical FCN model method was compared and analyzed. The results show that the proposed network model can not only effectively alleviate the "salt and pepper phenomenon", but also ensure the accuracy of small-area vegetation extraction and vegetation area boundaries. The method can automatically integrate multiple features, so it can effectively reduce the misclassification and leakage of vegetation pixels and improve the accuracy of vegetation extraction.