Weighted split-flow network auxiliary with hierarchical multitasking for urban land use classification of high-resolution remote sensing images

Abstract

The progress of research on urban land use (ULU) classification is slow primarily because of the complexity of urban scenes with strong presence of human activities and the phenomenon of one ULU presenting multiple urban scenes. To improve the accuracy of ULU classification based on high-resolution remote sensing images, a weighted split-flow network (WSNet) with K-fold cross validation (K-CV) and hierarchical multitasking is proposed in this study. The split-flow strategy and attention module are applied to optimize the learning ability of the WSNet, K-CV is used to strengthen the robustness of the WSNet, and the hierarchical multitasking can address one ULU presenting different scenes. To verify the effectiveness of the proposed method, two datasets and verification data were used to conduct experiments of model training and evaluation, respectively. The results show that on ImageNet-1K and the Urban dataset, the WSNet is superior to the other models. Furthermore, on the Urban dataset, the results of the WSNet with hierarchical multitasking and K-CV were significantly improved in all metrics compared with the WSNet, and the accuracy indicator reached 92.17%. Additionally, the effectiveness of WSNet with hierarchical multitasking and K-CV was also confirmed in the verification experiment. Therefore, the proposed method can effectively improve the performance of ULU classification of high-resolution remote sensing images and provide technical support for urban management and decision-making.