Land Cover Satellite Image Classification Using NDVI and SimpleCNN

Abstract

Image classification and prediction is a task which is embedded with quite a lot of challenges. Introduction of deep learning gave a rapid rise in this area of research. The efficient and the simplest deep learning algorithm that has helped researchers to make immense contributions in the field of image classification is Convolutional Neural Network (CNN). One of the important applications of image classification is in remote sensing, where it is used for land cover classification. In this paper we developed a SimpleCNN architecture for the classification of multi-spectral images from SAT-4 and SAT-6 airborne datasets. Two sets of experiments are conducted using the model by feeding it with different features. First level of experiment is done by providing the model with Near-Infrared (NIR) band information as it can sense vegetation health. The domain knowledge of Normalized Difference Vegetation Index (NDVI) motivated us to utilize Red and NIR spectral bands together in the second level of experimentation for the classification. It is observed from the experiment that the two band information gave better results for land cover classification.