Urban land cover classification method for high-resolution UAV imagery
CHEN Weijie
Abstract：To enhance the accuracy of urban feature classification and support refined urban management,this study constructs the TMSK dataset based on high-resolution UAV imagery,comprising 480 annotated samples across six feature categories:buildings,impervious surfaces,vegetation,bare land,water,and cars.The Deep-UNet model was employed for training and validation,configured with 300 epochs,and integrated with the Ohem loss function and a Warmup learning rate strategy.Comparative experiments were conducted against benchmark datasets such as Potsdam.Results demonstrate that the TMSK dataset outperforms the benchmarks in both image clarity and annotation quality,achieving higher accuracy,F1-score,and mean Intersection over Union(mIoU).This confirms the effectiveness and practicality of the proposed dataset for urban feature classification tasks,providing reliable data support for high-resolution UAV imagery analysis.
Keywords：UAV imageryurban land typedatasetclassification method
Publication Date：2025-11-25
Pages：4( 63-66 )