A Principal Components Analysis based Hotelling’s T2 Control Chart in Combination with Random Forest for Urban Structures Extraction

With the growing impact of populations, land cover changes associated with urbanization have gained a significant interest. In recent years, the land use mapping of urban structures primarily employed very high resolution (VHR) images. Such data are usually characterized by large sizes in addition to being often subjected to noises. Indeed, it is frequent for VHR images, especially open source data, to contain outliers and cast shadows that may hinder the extraction of the needed information while generating a high computational time.In this article, we investigate a combination of a principal components analysis based Hotelling’s T2 control chart with the random forest (RF) algorithm to classify a VHR image given its high dimensionality and noisiness. In addition, a data integration-based technique that combines urban spectral indices (USI), color space transformations (CST) and multispectral satellite images is used to enhance the extraction of urban elements.The obtained results are validated through established criteria showing a significant improvement in terms of accuracy. Computational times are also reported to further demonstrate the efficiency of the proposed methodology in comparison with conventional use of the RF.