Evaluation of Land Use/Land Cover Classification based on Different Bands of Sentinel-2 Satellite Imagery using Neural Networks

Abstract

Spatial data analytics is an emerging technology. Artificial neural network techniques play a major role in analysing any critical dataset. Integrating remote sensing data with deep neural networks has led a way to several research problems. This paper aims at producing land use land cover map of Bangalore region, Karnataka, India with various band combinations of sentinel satellite imagery obtained from google earth engine. LULC map classes include water, urban, forest, vegetation and openland. Band combinations of satellite images represent different characteristics of spatial data. Hence, several band combinations are used to build LULC maps. Also, classified maps are generated using different neural networks with pixel-based classification approach. Appropriate performance metrics were identified to evaluate the classification results such as Accuracy, Precision, Recall, F1-score and Confusion Matrix. Among neural networks, Convolutional Neural Network technique outperformed with 98.1 % of accuracy and less error rates in confusion matrix considering RGBNIR (4328) band combination of satellite imagery.