Mapping Urban Impervious Surface With an Unsupervised Approach Using Interferometric Coherence of SAR Images

• Published in: IEEE journal of selected topics in applied earth observations and remote sensing Vol. 15; pp. 2734 - 2744
• Main Authors: Liang, Xindan, Lin, Yinyi, Zhang, Hongsheng
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Classification; Coherence; Hydrology; Image classification; Interferometric synthetic aperture radar; interferometric synthetic aperture radar (InSAR); Interferometry; Land surface; Mapping; Optical surface waves; Remote sensing; SAR (radar); spatial-temporal baseline; Surface topography; Synthetic aperture radar; unsupervised classification; urban impervious surface (UIS); Vegetation mapping
• ISSN: 1939-1404; 2151-1535
• DOI: 10.1109/JSTARS.2022.3149813

Impervious surface is significant in hydrology, urban management, ecology, and other research areas. Therefore, extracting impervious surface is crucial to understanding the change of environment and ecosystem. However, previous supervised classification methods usually rely on comprehensive training samples and human experiences. The article on automatic and efficient impervious surface extraction is still underexplored. In this study, we investigated the potential of using interferometric synthetic aperture radar technology for unsupervised urban impervious surface (UIS) mapping. A total 136 coherence maps of Hong Kong with different perpendicular and temporal baselines were used to classify UIS and non-UIS through setting different coherence thresholds. We proposed a new method, entitled interferometric coherence thresholding method, which can achieve high classification accuracy using coherence map. The result illustrates: first, using a threshold of 0.4, nearly 90% of images achieve an overall accuracy of over 80%. The highest one reaches 88.25%, which is much higher than using K-means and ISODATA method; second, small temporal baselines (12 and 24 days) are likely to reduce the classification accuracy; third, using optimal classification threshold, the coherence of two SAR images would not bring large impact on classification results.