Classification of Remotely Sensed Data Using Fisher’s Linear Discriminant

In place of time-consuming and expensive data collecting on the ground, remote sensing allows for rapid, repeated coverage of enormous areas, with widespread practical applications in fields as diverse as meteorology, disaster reporting, and climate science. Because there are many different LULC classes that can be analyzed, it is important to investigate the classifiers’ classification performance to learn about their strengths and weaknesses. In this study, we use Fisher’s linear discriminant analysis approach to classify two sets of multispectral medium-resolution remote sensor (RS) data and evaluate its performance in recognizing LULC classes, extracting LULC classes, and distinguishing between spectrally overlapping classes. Ten randomly selected pixels from the data are used to illustrate Fisher’s LDA’s pixel assignment approach. The classification analysis shows that Fisher’s LDA is very good at extracting classes that are spectrally dominating, but it is not very effective at extracting classes that are spectrally subservient.