Multi-label land cover classification is the process of classifying the land into different classes based on the type of land. A well-defined land classification is very useful, as we can find out the type of land with the satellite images of that particular area, which helps the users decide whether the land is suitable for their purposes or not. Several research efforts using machine learning techniques have been underway to accurately label the land, but there is still room for improvement. To improve the classification accuracy, in this paper we propose a 2D convolutional neural network (CNN) model with convolution and max-pooling, and that is fully connected, with dense layers. The proposed 2D-CNN model consists of twoConv2D layers, a flattened layer and two-dense layers. The proposed network comprises of 5,329,361 parameters/nodes out of which 5,329,169 and 192 are trainable and non-trainable parameters/nodes respectively. We classify the images into 17 labels such as agricultural, airplane, baseball, diamond, beach, buildings, chaparral, dense residential etc., with 2D-CNN model with 80% accuracy. We classified the land in this research using the 2D-CNN model. We examined 2100 satellite images to evaluate the model's performance. The experimental study shows that multiple labels in remote sensing images is predicted most accurately by the proposed CNN model. It distinguishes trees, pavement, water, and other labels in remote sensing images considerably well. The tabulated results show that a state-of-the-art analysis was done to learn varying land cover classification models. In the future, we want to investigate graph-based multi-label classifiers and design more effective algorithms for remote sensing image annotation.

Classification, Land Cover, Deep Learning, CNN