Enhanced DeepLabv3+ for Urban Land Use Classification Based on UAV-Borne Images

The research on urban land use classification is of great significance to the efficient management of urban land resources and the sustainable development of cities. The rapid development of UAV industry provides rich data support for urban land use classification. In order to extract higher semantic features of UAV-borne images and improve classification accuracy, this paper proposes a classification method based on enhanced DeepLabv3+ network. In view of the complexity and large amount of information, the initial backbone network Xception is replaced by the lightweight network MobileNetv2, and the classification results are optimized by using the fully connected conditional random field (CRF). The experimental results show that using the method proposed in this paper, the overall classification accuracy can reach 84.75% and the kappa coefficient can reach 0.737, which is higher than U-Net state-of-the-art algorithm, indicating that this method is reliable for urban land use classification and the classification results are highly consistent with the actual categories.