Enhanced Feature-Based Clustering for Urban Land Use Pattern Detection

Urban land use patterns are essential for urban planning and environmental management. Traditional analysis methods, such as manual surveys and satellite imagery interpretation, are often labor-intensive and struggle with large-scale data. In contrast, data-driven approaches offer promising alternatives. This paper addresses the challenge of detecting urban land use patterns by proposing an efficient framework that leverages the Vision Transformer (ViT) and Variational Autoencoder (VAE) models to generate robust feature representations from images. To enhance the clustering process, UMAP (Uniform Manifold Approximation and Projection) is used for dimensionality reduction, followed by k -means clustering. Our approach overcomes the limitations of existing image clustering tools by generating embedding features that capture both high-level semantic information and underlying generative structure. Comparative experiments were conducted on the real-life Urban Atlas dataset to evaluate the performance of the proposed framework against various feature extraction options in the proposed methods. Experimental results show that the proposed method outperforms others by capturing the complexities of urban land use patterns, offering enhanced flexibility and robustness.