Landscape mapping and pattern analysis of fine-grained urban functional zones (UFZ) are of great research value in the field of urban planning and urban development assessment. Previous studies employing multispectral images have largely enabled only coarse-grained land-use classification of urban areas. By contrast, hyperspectral images can provide very detailed spectral features and have shown some potential to facilitate fine-grained classification of urban areas. In this article, we evaluate and analyze the classification of fine-grained UFZ in the central city of Wuhan, Hubei, China, using GaoFen-5 (GF-5) hyperspectral satellite imagery. We first compare the performance of hyperspectral data (GF-5) and classical multispectral data (Landsat 8) for the classification of fine-grained functional zones of cities by employing two classical classification algorithms and two deep learning methods. We also propose the deep learning-based spectral-spatial unified networks combined with a fully connected conditional random field (SSUN-CRF) algorithm for fine-grained UFZ mapping to enable better landscape pattern analysis. We then analyze the landscape pattern of the main urban areas of Wuhan by combining ten landscape indicators based on the precise UFZ classification results. Experimental results illustrate the followings. First, compared to multispectral images, hyperspectral images can allow for a more accurate UFZ classification. Second, deep learning classification algorithms can better exploit hyperspectral image data, with the SSUN-CRF algorithm, in particular, being able to achieve an overall accuracy of 93.86% and a Kappa coefficient of 92.08%. Finally, the landscape pattern analysis demonstrates that hyperspectral remote sensing imagery shows significant potential in mapping fine-grained UFZ. It is beneficial for further urban study and planning.

Convolutional neural networks (CNNs) have made significant advances in remote sensing scene classification (RSSC) in recent years. Nevertheless, the limitations of the receptive field cause CNNs to suffer from a disadvantage in capturing contextual information. To address this issue, vision transformer (ViT), a novel model that has piqued the interest of academics, is used to extract latent contextual information in remote sensing scene classification. However, when confronted with the challenges of large-scale variations and high interclass similarity in scene classification images, the original ViT has the drawback of ignoring important local features, thereby causing the model’s performance to degrade. Consequently, we propose the hierarchical contextual feature-preserved network (HCFPN) by combining the advantages of CNNs and ViT. First, a hierarchical feature extraction module based on ResNet-34 is utilized to acquire the multilevel convolutional features and high-level semantic features. Second, a contextual feature-preserved module takes advantage of the first two multilevel features to capture abundant long-term contextual features. Then, the captured long-term contextual features are utilized for multiheaded cross-level attention computing to aggregate and explore the correlation of multilevel features. Finally, the multiheaded cross-level attention score and high-level semantic features are classified. Then, a category score average module is proposed to fuse the classification results, whereas a label smoothing approach is utilized prior to calculating the loss to produce discriminative scene representation. In addition, we conduct extensive experiments on two publicly available RSSC datasets. Our proposed HCPFN outperforms most state-of-the-art approaches.