Automated Building Extraction from Remote Sensing Imagery

Automated building extraction from remote sensing imagery has emerged as a critical component in geospatial analysis, enabling rapid and precise mapping of urban and peri-urban environments across the globe. By leveraging high-resolution satellite, aerial and unmanned aerial vehicle (UAV) imagery, these techniques can systematically delineate building footprints, heights and structural characteristics without manual intervention. Core methodological advances have been driven by deep learning architectures, especially convolutional neural networks that perform pixel-level semantic segmentation and instance segmentation, often enhanced by super-resolution algorithms to counteract varying image quality. Integration of spectral, spatial and elevation data—such as digital surface models derived from photogrammetry or LiDAR—has further improved accuracy and the capacity to infer three-dimensional attributes. The resulting datasets support a multitude of applications, from urban planning and infrastructure monitoring to disaster response and environmental assessment. As computational efficiency and access to open-source imagery continue to grow, fully automated pipelines are increasingly robust, offering consistent outputs at continental scales while addressing challenges of generalisability across diverse typologies and lighting conditions.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Automated Building Extraction from Remote Sensing Imagery publication trend

The graph below shows the total number of articles in automated building extraction from remote sensing imagery across all publications each year (not limited to Nature Index journals).

Technical terms

Semantic segmentation: the process of assigning a label to every pixel in an image, classifying each as building, vegetation, road or other categories.

Instance segmentation: an extension of semantic segmentation that distinguishes and delineates each individual object instance within the same class.

Convolutional neural network (CNN): a deep learning model that employs convolutional filters to automatically learn spatial hierarchies of features from image data.

Super-resolution: techniques used to enhance the spatial resolution of imagery by reconstructing high-resolution details from lower-resolution inputs.

Monocular remote sensing imagery: single-view images captured by aerial or satellite sensors, as opposed to stereo or multi-angle acquisitions, often requiring specialised methods to infer depth.

Digital surface model (DSM): a three-dimensional representation of the Earth’s surface that includes natural and built features, used to augment two-dimensional imagery with elevation information.

References

1. Large-scale individual building extraction from open-source satellite imagery via super-resolution-based instance segmentation approach. ISPRS Journal of Photogrammetry and Remote Sensing (2023).
2. CNNs for remote extraction of urban features: A survey-driven benchmarking. Expert Systems with Applications (2024).
3. 3DCentripetalNet: Building height retrieval from monocular remote sensing imagery. International Journal of Applied Earth Observation and Geoinformation (2023).