Deep learning

To conduct a thorough examination of building infrastructure and develop a deeper understanding of human-urban dynamics, this study introduced a multi-task network, MT-BR, to extract unified building information. This study proposed a method for selecting representative samples for robust network training that considers both their location and the features of the urban environment they represent, so that they can be used effectively across large urban landscapes.

Methods to integrate Earth system modelling (ESM) with deep learning offer promise for advancing understanding of Earth processes. This Perspective explores the development and applications of hybrid Earth system modelling, a framework that integrates neural networks into ESM throughout the modelling lifecycle.

Roadside noise barriers (RNBs) are important urban infrastructures to ensure a city is liveable. This study provides the first reliable and nationwide vectorized RNB dataset with street view imagery in China. The generated RNB dataset is evaluated in terms of two aspects, i.e., the detection accuracy and the completeness and positional accuracy. The method is based on a developed geospatial artificial intelligence framework.

The first rooftop area dataset for 90 cities in China is of high quality and can be used for data support and decision-making that can facilitate sustainable urban development effectively.

To overcome these two challenges of delineating roof structure lines from satellite imagery, the Deep Roof Refiner—a detail-oriented deep learning network and a synthetic strategy were proposed.