Feature extraction from sentinel 2 imagery using deep learning approach for urban land use land cover classification

Abstract

In satellite remote sensing, extraction of features from the imagery newlinefor the urban environment is challenging due to the heterogeneous mix of newlinevarious features. Researchers across the world have been attempting newlinecontinuously to improve satellite image classification accuracy, with an aim to newlinefacilitate better and sustainable development plans. Land Use / Land Cover newline(LU/LC) classification using satellite remote sensing data is quite a challenging newlineprocedure due to the spectral and geographical complexity of the imagery. To newlineaddress this issue, the emergence of Deep Learning (DL) in the advancement newlineof machine learning can pave the path towards processing, analyzing and newlinemanipulating geospatial sensor data. This research aims to frame a hybrid newlinemethodology that can extract the urban features such as settlement, industry, newlinewaterbody and green cover (tree) more precisely for LU/LC classification. newlineIn order to implement this methodology, the selected study area is newlineChennai city with many urban features between 12°58 00quotN to 13°9 00quotN, and newline80°8 00quotE to 80°18 00quotE. An image of Sentinel-2 Multispectral Instrument newline(MSI) acquired on March 22, 2018, with a spatial resolution of 10 meters of 4 newlinemultispectral bands (blue, green, red and near-infrared) and radiometric newlineresolution of 12 bits was used for spectral classifications using Maximum newlineLikelihood Classifier (MLC) and Deep Learning (DL) methods. newline