Unsupervised Land Use Land Cover Mapping of Sentinel-2 Data using Convolutional Autoencoders

The increasing importance of Multispectral image analysis in recent years is particularly evident in the domain for monitoring land use land cover. This is crucial for addressing complex issues such as food production loss, heightened vulnerability to disasters, and the broader impact on climate. One significant challenge in this realm is the proper management of land use, which necessitates accurate and up-to-date land use land cover maps. Our work employs an unsupervised approach to generate land use land cover maps using Sentinel-2 Multispectral datasets of 2018. A Convolutional Autoencoder extracts high-level features from season-wise stacked data, followed by the Hierarchical Clustering Algorithm to systematically organize and classify land use patterns. This unsupervised technique demonstrates the effectiveness of hierarchical clustering in creating detailed and informative land use land cover maps over time. This highlights the application of unsupervised learning in generating meaningful insights from land use land cover maps, showcasing its potential for enhanced analysis and interpretation.