Urban land use classification is of great significance for capturing current situation of cities accurately and optimizing urban spatial structure. An urban land use classification model specifically for China is proposed by using remote sensing images. Firstly, a multi-resolution feature fusion convolution neural network is designed to recognize urban land use types. Besides, according to the distribution characteristics of urban functional areas in China, a new dataset for the urban land use classification is proposed. Experimental results show that the proposed work can reach 88% accuracy on six urban land use types, which validates the effectiveness of the algorithm in the classification of urban land use. Finally, a case study for part of Beijing’s main urban districts demonstrates the value and effectiveness of the proposed model for providing data support in the field of urban planning.