Spectral Indexes Evaluation for Satellite Images Classification using CNN

Deep learning approaches are applied for a wide variety of problems, they being used in the remote sensing field study and showed high performance. Recent studies have demonstrated efficiency using spectral indexes classification because accuracy F1 score increasing comparison with usage only RGB channels. The paper problem satellite images on EuroSAT dataset proposed convolutional neural network. In research set most been selected calculated dataset. Then, novel comparative analysis was carried out. It has established that significant (NDVI, NDWI, GNDVI) increased from 64.72% to 84.19% 63.89% 84.05%. biggest improvement obtained River, Highway PermanentCrop classes.