Assessing the performance of the multi-morphological profiles in urban land cover mapping using pixel based classifiers and very high resolution satellite imagery

Improved spatial and spectral resolution from recent sensor advancements provides opportunities for detailed and enhanced accuracies in the classification of heterogeneous urban landscapes. However, to date, such classifications have remained a challenge for most pixel based techniques. Whereas object based techniques have proved effective in classifying heterogeneous urban landscapes, by providing an effective framework for analysis of high spatial resolution images, challenges such as under and over-segmentation and non-robust statistical estimation impede their optimum performance in the often complex urban landscapes. Therefore, it is imperative that an effective classification approach is identified for effective utilisation of both spatio-spectral characteristics of image objects. Morphological techniques, especially multi-morphological profiles (MMP), provide an effective framework for the analysis of both spectral and spatial information from very high-resolution satellite imagery by performing image analysis based on features such as geometric, texture and contrast. In this study, using Support Vector Machine (SVM) and Maximum Likelihood (ML) classification algorithms, we compare the classification accuracies based on MMP as feature vector against those without MMP as a feature vector. Results from this study indicate that the use of MMP as a feature vector produced significantly higher classification accuracies of 84.8% and 82.2%, compared to 75.77% and 77.6% without MMP as a feature vector for SVM and ML, respectively. The study concludes that MMP can be used as a feature vector to increase the classification accuracy of a heterogeneous urban land use land cover. (C) 2020 The Authors. Published by Elsevier B.V. on behalf of African Institute of Mathematical Sciences / Next Einstein Initiative.