Performance evaluation of pansharpening Sentinel 2A imagery for informal settlement identification by spectral-textural features

Abstract

The diversity of informal settlement morphologies across locales makes their mapping inherently challenging in heterogeneous urban landscapes. The aim of this study was to evaluate the potential of pansharpening techniques on Sentinel 2A data, and textural features, in enhancing informal settlement identification accuracy in a fragmented urban environment. Brovey transform, intensity, hue and saturation transform, Environmental Systems Research Institute (ESRI), simple mean, and Gram–Schmidt techniques were employed to pansharpen multispectral bands of Sentinel 2A, bands 5, 6, and 7 in the first group, and bands 8A, 11 and 12 in another, using an average of bands 4 and 8 as the panchromatic band. The main objective was to investigate the efficacy of pansharpening Sentinel 2A imagery and texture analysis in automated mapping of morphologically varied informal settlements. An evaluation of the quality of fused images was undertaken through computation of the correlation between the spectral values of the original multispectral and pansharpened image. Grey-level-co-occurrence matrix texture features were extracted from the pansharpened images, and subsequently incorporated in the classification process, using a support vector machine classifier. Our results confirm that the Gram–Schmidt fusion technique yielded the highest informal settlement identification accuracy (F-score 95.2%; overall accuracy 91.8%). The experimental results demonstrated the potential of pansharpening Sentinel 2A, and the added value of image texture for a more nuanced characterisation of informal settlements.

Keywords:

• pansharpening
• Sentinel 2A
• informal settlement mapping
• remote sensing
• image texture

DATA AVAILABILITY STATEMENT

No data sets were generated during the current study.

CONFLICTS OF INTEREST

The authors declare no conflict of interest.

Additional information

Funding

This study was supported by the UKZN Big Data for Science and Society project (BDSS). The authors also thank the SARChI Chair in Landuse Planning and Management, Grant no 84157 for their generous financial support.