Sentinel-2 Data for Land Use/Land Cover Mapping: A Meta-analysis and Review

Machine learning and deep learning algorithms are extensively used in the fields of remote sensing image analysis. In this study, the major concepts pertinent to Sentinel-2 satellites are introduced, in which a total of 177 articles from conference proceedings, journals and book chapters were taken into study. Initially, a meta-analysis and review was conducted to analyze the usage of Sentinel-2 images in terms of various applications i.e., in the fields of Agriculture, Land Use/Land Cover, Forest Cover and Urbanization. Various methods for study selection and data extraction was discussed including the refinement process. This review comprises a detailed description of the Sentinel-2 Satellite Mission Programme which comprises of brief introduction, characteristics and properties and data products of Sentinel-2 satellite images. Pre-processing phases like Geometric Correction, Atmospheric Correction and cloud cover masking are discussed elaborately. Further, Land Use Land Cover Classification methods i.e., Unsupervised & Supervised methods, Object-based Image Analysis(OBIA) and Pixel-based image analysis have been discussed. Any classification result is incomplete without Accuracy assessment. Therefore, the accuracy assessment of the classification methods was evaluated and compared in graphically considering 50 case studies from the literature. This review paper also discusses a few Deep Learning Algorithms like CNN networks, Recurrent Neural Networks, Restricted Boltzmann machines and deep belief networks. A further analysis of various applications of Sentinel-2 images are also discussed. This review covers nearly every technology and application using the Sentinel-2 images ranging from pre-processing to accuracy assessment. Finally, a brief conclusion is presented concerning the state-of-art methods and directions for future research.