2026· IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing· Vol 19, pp. 24115-24129· 0 citations· 34 references
Abstract
Identifying buildings and urban/built-up features has become increasingly important as cities grow more complex and traditional pixel-based classification methods struggle with mixed-land-cover signatures. Despite their high spatial resolution and rich multispectral capabilities, openly available datasets, such as Sentinel-2, often remain insufficient for extracting fine-scale urban information. For example, roads, buildings that are smaller than their spatial resolution, or urban areas alongside gardens, are the most challenging features to identify. To detect and map built-up structures that fall below the sensor's nominal pixel size, or have complex mixed land-cover signatures, we propose to use regression-based subpixel mapping. We rely on pretrained Alpha Earth Embeddings, which provide rich, multitemporal feature representations that help disentangle built-up features in mixed pixels. These embeddings are combined with freely available very high spatial resolution land-cover map for training. The proposed method could successfully identify small buildings and narrow roads, having a size smaller than the spatial resolution of the considered satellite data. In particular, it was able to detect buildings with areas as small as <bold><inline-formula><tex-math notation="LaTeX">$20 \,\,\mathrm{m}^{2}\,$</tex-math></inline-formula></bold> and roads represented by buffers as narrow as <bold><inline-formula><tex-math notation="LaTeX">$1.5 \,\mathrm{m}$</tex-math></inline-formula></bold>. Our results indicate that combining subpixel mapping with embedding representations enables improved identification of complex built-up features having a size smaller than <bold><inline-formula><tex-math notation="LaTeX">$10 \,\mathrm{m}$</tex-math></inline-formula></bold> when compared to existing land-cover maps.
Recently, low-resolution remote sensing (RS) images have received significant attention because of their widespread spatial coverage, minimum acquisition cost, quick transmission ability, and large-scale earth observation suitability. However, land-use and land-cover (LULC) classification using low-resolution satellite imagery remains challenging due to restricted spatial information, spectral similarity amongst land-cover classes, noise differences, and complex scene heterogeneity. Though recent deep learning-based models have exhibited effective outcomes, they still suffer from insufficient feature representation, inadequate contextual dependency learning, and minimal classification accuracy when processing low-resolution RS images. To resolve these issues, this study develops a lightweight feature extraction model with Temporal Convolutional Networks for low-resolution remote sensing image classification. The proposed model initially preprocesses the images to improve feature consistency and quality. The feature extraction phase then employs MobileNet-V2 to identify and represent relevant spatial patterns in RS images, followed by a temporal convolutional network for RSI classification, enabling effective modeling of sequential and contextual dependencies in spatial features. Furthermore, adaptive fine-tuning of model parameters is performed using an artificial rabbit optimization algorithm to enhance classification accuracy and convergence behavior. Extensive experimental evaluation of the LFEARO-LULCRSI model on the benchmark EuroSat Dataset from Sentinel-2 imagery demonstrates improved performance over existing methods, achieving an accuracy of 98.57%. An ablation study is also performed to examine the contribution of individual model components. The proposed model thus proves useful for effective geospatial analysis in agriculture, urban planning, disaster assessment, and sustainable environmental management, enhancing feature discrimination and contextual dependency learning in low-resolution satellite imagery.
G. Sravanthi, A. Gnanasekaran, G. Ramesh· VFAST Transactions on Softwa...· 0 citations
Building materials and their spatial distribution play a significant role in determining the outcomes of human-caused and natural disasters in urban and peri-urban areas. However, building-level data on building and roofing materials are scarce. Here, we explore the feasibility and performance of a Convolutional Neural Network (CNN) model using spectrally transformed high-resolution multispectral imagery to map roofprints (i.e., classifying and delineating roofing materials at the building footprint-level) in Washington, District of Columbia (D.C.) and Denver, CO, United States. To generate consistent training data, we integrate geospatial vector data of individual building footprints with real estate industry-derived building-level roofing material data to create labeled image data from Planet SuperDove imagery. We compare the CNN classifier to a pixel-based machine learning (ML) model to demonstrate the capability of our roofprints mapping approach. With F1-scores ranging from 0.56 to 0.95 for the most common roof material classes, the CNN model outperformed the pixel-based ML classifier by 15% and 17% in Washington, D.C., and Denver, respectively. Results demonstrate within-domain robustness for the studied metropolitan areas, which are characterized by differing building densities, roof morphologies, and material patterns. While cross-region transferability was not evaluated, our findings provide a controlled comparison of pixel-based and context-aware approaches for rooftop material mapping and highlight the importance of hierarchical representations that integrate spectral information with roof texture, edge characteristics, spatial arrangement, and neighborhood context for improving classification performance. Accurately mapping building materials has the potential to advance urban planning and environmental policies, including assessments of heat exposure, energy demand, as well as hazard risk and community resilience.
C. Amaral, Maxwell C. Cook, Johannes H. Uhl et al.· Remote Sensing· 0 citations
Building extraction from high-resolution remote sensing imagery is critical for urban planning and smart city development, yet it faces challenges such as blurred boundaries, missing fine details, and severe background interference. To address these issues, this study proposes an improved model named GSU-HRNet, which integrates attention mechanisms and boundary refinement strategies on the basis of UHRNet's high-resolution parallel backbone. An enhanced Pyramid Squeeze-and-Excitation (PSE) module is embedded in the lateral feature transmission paths of each hierarchical stage, capturing multi-scale contextual information via adaptive average pooling of multiple sizes to strengthen semantic responses for buildings and suppress background noise. A Gated Bottleneck Convolution (GBC) module is further introduced in the feature fusion stage, adopting a dual-branch structure with gating mechanisms and residual connections to selectively regulate fused features, alleviate redundant feature accumulation, and improve the stability of feature representation. Experiments were conducted on the aerial imagery subset of the WHU Building Dataset (covering 450 km2 in Christchurch with 8,189 512×512 image tiles), which was split into training, validation and test sets at a ratio of 6:1:3. Ablation experiments verify the effectiveness and complementarity of PSE and GBC modules, with the combined model achieving optimal performance. Quantitative comparisons show that GSU-HRNet outperforms classical models like U-Net and PSPNet, reaching an IoU of 89.93% and an F1-score of 94.50%. Qualitative analysis demonstrates that the proposed model yields clearer building boundaries, more complete structural preservation, and reduced false detections and omissions, even in challenging scenarios with complex building structures and shadow interference. The results confirm that GS-UHRNet effectively enhances feature representation and boundary delineation accuracy, and exhibits strong generalization ability across different building extraction datasets, providing a robust solution for automated building extraction from hig-hresolution remote sensing imagery.
Shi He, Shiye Zhang, Xiujuan Liang et al.· GEOINFORMATICS· 0 citations
Building footprint extraction in high-density urban areas remains difficult because spectral confusion and shadows obscure building–background differences, while dense adjacency and complex boundaries hinder boundary recovery and adjacent-building separation. To address these issues, this study combines Digital Orthophoto Maps (DOM) and normalized Digital Surface Models (nDSM) and develops ACM-PSPNet for high-density built-up areas of Hong Kong. ACM-PSPNet extends the PSPNet baseline by integrating atrous spatial pyramid pooling and convolutional block attention for high-level contextual enhancement and feature recalibration, together with an MS-Fuse decoder for progressive multilevel spatial-detail recovery. The main experiments were independently repeated three times and evaluated on a spatially independent test set. In the input-source comparison, DOM+nDSM yielded a mean IoU of 84.94%, compared with 68.23% for DOM and 83.23% for nDSM alone. Using DOM+nDSM as the common input, the model comparison showed that ACM-PSPNet provided the highest performance among the eight evaluated models, with mean IoU, F1 Score, and Boundary F1 values of 87.81%, 93.13%, and 79.78%, respectively. In selected typical scenes, the under-segmentation proportion decreased from 35.7% for PSPNet to 5.0% for ACM-PSPNet. The results support ACM-PSPNet as an accuracy-oriented framework for dense urban building mapping, particularly where reliable boundary recovery and adjacent-building separation are required, although broader cross-city and cross-sensor validation remains necessary.
Land cover maps are important in spatial planning and urban development as well as for environmental monitoring. It provides an overview on various types of land cover and that information is needed for the current and future land development and planning. While it is recognized that land cover maps are important, not many organizations have access to high resolution imagery due to budget constraints and therefore they will use whatever available satellite imagery to support their current business operations. With the available of Sentinel imagery data that provides 10-meter resolution with no cost or free for public to use, it serves as an initial data source when higher-resolution imageries are not available. However, the 10-meter resolution has a limited number of bands (Blue, Green, Red and Near-Infrared). This limitation reduces the spectral depth and often poses challenges in processing and disseminating the objects that have similar classes such as vegetation and forest. Therefore, this study evaluates the performance of the commonly used pixel-based classification technique under the limited spectral conditions. The findings provide practical guide for planners, researchers or even practitioners who rely solely on 10-meter resolution from Sentinel 2 to support their operational needs.
Busiai Bin Seman, Tarmiji Masron· Arab world geographer· 0 citations
Remote sensing-based land cover classification and change detection are essential for ensuring environment, planning cities, and managing resources. Accurately extracting spatial and temporal information from high-resolution multi-temporal images remains challenging due to feature inconsistency, class imbalance, and limited integration of spatial–temporal representations in existing methods. Traditional deep learning approaches often fail to capture subtle variations, struggle to generalize across complex and heterogeneous landscapes, leading to reduced accuracy in change detection. To overcome these constraints, this research introduces an advanced deep learning framework, GeoFusion-ChangeNet, for high-resolution land cover mapping and automatic change detection. The proposed model integrates hierarchical convolutional feature encoder with a residual deep feature backbone to effectively capture rich multi-scale spatial features. An attention-based mechanism is included to improve feature representation by concentrating on informative regions while suppressing irrelevant information. Furthermore, a multi-temporal feature fusion method that combines feature concatenation and feature differencing is used to effectively capture temporal variations between input images. The proposed system is also deployed through an interactive interface to enable real-time visualization of classification and change detection results. The experimental results on the SECOND benchmark dataset achieved Accuracy of 95.00%, IoU of 86.00%, F1-Score of 93.00%, and Kappa Coefficient of 84.00%. In addition, cross-dataset evaluation on the LEVIR-CD benchmark achieved Accuracy of 93.12%, F1-Score of 90.61%, IoU of 82.83%, and Kappa Coefficient of 81.47%, indicating that GeoFusion-ChangeNet maintains stable performance under an additional remote sensing change-detection setting.
Yi Chen· Scientific Reports· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
