Informal settlement extraction from high resolution satellite images using mathematical morphology and deep learning based approaches wavelet frame transform

Abstract

As the worldwide urban population increases due to the influx of rural migrants, most cities in emerging nations are confronted with the formation and expansion of informal settlements. Approximately one-fourth of the global urban population lives in informal settlements at present. Despite the fact that the expansion of informal settlements affects residential land-use planning, newlineidentification of informal settlements is important for urban analysis, disaster management, land cover change assessment and updating geographical databases. But researchers are confronted with challenges in identifying accurate remote sensing-based informal settlement extraction approaches with reliable and detailed information due to their microstructure, instability and wide variety of textures. During the past decade, informal settlements have been mapped using census, survey and participatory methods, which require a time-consuming field survey. Recent advancements in sensing technologies have resulted in abundance of very high resolution (VHR) remotely sensed imagery with excellent spatial and spectral information. Lacunarity-based multiscale analysis, statistically-based grey level co-occurrence matrix (GLCM), spectral-based wavelet frame transform (WFT) and Deep Convolutional Neural Network (DCNN) have all been shown to be effective techniques for detecting informal settlements in VHR images. However, these approaches consider newlinescale, spatial and spectral features, but exclude the effects of contextual and structural features in distinguishing slums. newline newline