Multi-level Feature Extraction for Automated Land Cover Classification using Deep CNN with Long Short-Term Memory Network

Abstract

In remote sensing and satellite imagery, classification of land cover usage are the vital tasks that help to understand about the physical aspects of the surface of the Earth and its utilization. The advancements in information technology significantly contributed in the development of robust methods for interpreting images and recognizing patterns from them. With the rapid development of deep learning methods and increasing availability of dataset, many researchers apply deep learning for satellite and remote sensing image classification. This paper proposes a novel network using deep CNN with LSTM that extracts the features from satellite images for land cover classification. Convolutional Neural Network (CNN) is used to extract the features from the images and long short-term memory (LSTM) network is used to support the sequence prediction and classification. The experiment is carried out on a novel EuroSAT dataset that contains 27000 land cover images that categorizes into 10 distinct classes. The performance of the proposed model is also compared with the benchmark pre-trained CNN models. The proposed model outperforms and achieved 92.90% precision, 93.15% recall and 93.02% F1-score values. https://www.sciencedirect.com/science/article/pii/S2352914820305621