Exploitation of textural and morphological image features in Sentinel-2A data for slum mapping

Abstract

In this paper we use image texture and morphological profiles for mapping slums in Sentinel-2A imagery. Varying sizes of the respective spatial descriptors (GLCM, differential morphological profiles) are tested for classification using a random forest classifier. Results are interpreted based on pixel-based and patch-based accuracy assessment. Best classification results have been reached at the pixel-based level with a kappa of 81.65 for the combined feature set with both GLCM and DMP. At the patch level, the analyses show that higher accuracies are reached with large kernel sizes and detection is better for large slum areas.