Paper Title
An Ensemble CNN Model for Land Cover Classification

One of the interesting uses of Earth observation satellite data is ‘Land Cover Classification’. In this paper, data containing geospatial images from the Sentinel-2 satellite have been utilized to categorize land cover as ‘Forest’, ‘Annual Crop’, ‘Highway’, ‘Herbaceous Vegetation’, ‘Industrial’, ‘Pasture’, ‘Residential’, ‘River’, ‘Permanent Crop’ and ‘Sea Lake’. After pre-processing the dataset of satellite images, it was passed through 3 pre-trained Convolutional Neural Network models, namely ResNet18, VGG16 and DenseNet121 in order to classify these images into the above-mentioned 10 classes. In order to further improve accuracy and make more accurate classifications, an ensemble model was created using the 3 CNN models and trained on the same data. It was observed that the ensemble model actually generated accuracy (96.852%) higher than the highest performing CNN model which in this case was the Resnet18 with 95.208% accuracy. Keywords - ResNet, DenseNet, VGG, Convolutional Neural Network, Land Cover Classification, Satellite Images, Ensemble