Fuzzy Machine Learning Algorithm for Different Roof Type Based Transitioned Building Footprint Mapping using Temporal Data

This research study is aimed at employing fuzzy machine learning approaches to map different roof type based transitioned buildings footprints. A comparative study has been carried out to bring out the differences between two methods of training data: ‘mean’ approach and ‘individual samples as mean’ approach. One pair of remotely-sensed temporal data used for the study was acquired from Operational Land Imager onboard Landsat-8 and Multi-spectral Sensor Instrument onboard Sentinel-2A and the other pair of temporal datasets, six years apart, were taken from Google Earth. Class-Based Sensor Independent-Normalized Difference Vegetation Index (CBSI-NDVI) approach was adopted to generate a temporal indices database, which reduces spectral dimensionality while retaining the temporal dimensionality. The temporal indices database was subsequently used as an input in the Modified Possibilistic c-Means (MPCM), fuzzy machine learning algorithm mapping of different roof type based transitioned building footprints. Based on the composition of the roof: concrete, red – and yellowcoloured rooftops were distinguished and the footprints were extracted accordingly. On evaluation of accuracy using F-measure, better performance was observed while using the ‘individual samples as mean approach’ for all the three types of rooftops.