Considering the complexity and fragmentation of urban forest structure, using high-resolution remote sensing data and intelligent machine learning algorithms such as deep learning to accurately extract urban forest information is crucial for intelligent monitoring and urban forest resources management. This paper used WorldView-3 high-resolution satellite imagery of urban areas in Yuhang District, Hangzhou and an improved U-net deep learning neural network combined with an object-oriented multi-scale segmentation algorithm developed to extract urban forests classification information. Firstly, through a large amount of training data, it obtained optimized parameters. And then, the U-net network is used to get the classification map. Finally, it carried out an object-oriented multi-scale segmentation algorithm to modify the urban forest extraction results to obtain accurate urban forest information. The research showed that: (1) the overall classification accuracy of urban forest based on the improved U-net deep learning neural network is 90.50%, and the Kappa coefficient is 0.886; (2) analyzing with an object-oriented multi-scale segmentation algorithm, it has corrected the salt and pepper noise and errors of boundary features in the U-net deep learning neural network’s results have. The total classification accuracy increased to 93.83%, and the Kappa coefficient rose to 0.9295. Therefore, the U-net network model combined with object-oriented multi-scale segmentation algorithms can effectively improve remote sensing vegetation classification precision.