We have downloaded one scene of Sentinel-2 multi-spectral images from Copernicus Open Data Hub that was acquired on September 21st, 2017. It covers most of the regions in Western New York area.The spatial and radiometric resolution of Sentinel-2 imagery varies by spectral band. Sentinel-2 data are acquired in 13 spectral bands in the VNIR and SWIR: four bands at 10 m: Band 2, Band 3, Band 4, and Band 8; six bands at 20 m: Band 5, Band 6, Band 7, Band 8A, Band 11, and Band 12; three bands at 60 m: Band 1, Band 9, and Band 10.
we have used Sen2Cor, a python based processor develop by ESA for Sentinel-2 Level 2A product for formatting and processing (such as atmospheric correction, aerosol optical thickness correction, water vapor retrieval, surface reflectance retrieval from TOA, geometric correction with DEM). All bands were resampled at 10 m resolution in Sentinel Toolboxes.In this tutorial we will use a sub-set of images with bands 2, 3, 4, 5, 6, 7, 8, 8A, 11 and 12. Prior to go for classification of Sentinel-2 satellite image using a Deep Neural Network, we need to process the spatial data. Then we will train our Deep Neural Network model with 5 feature classes extracted from Google Earth imagery: (1) parking/road/pavement, (2) building, (3) trees/bushes, (4) grass, and (5) water bodies. We used on-screen digitization in QGIS to create polygons representing members of these feature classes. A subset of atmospheric corrected surface reflectance of Sentinel-2 bands and feature class shape files are avilable for download as rar, 7z and zip formats.
First, we will convert the polygons to 2.5-meter x 2.5 meter raster grid, and then convert them to spatial points. We will next extract values from Sentinel-2 Band's B2, B3, B4, B5, B6, B7, B8, B8A, B11 and B12 bands and add them to the point data set. We also convert all raster bands to spatial point data frame and then to a CSV file. This grid-point data file will be used for prediction of landuse classes.