Sugarcane crop classification using time series analysis of optical and SAR sentinel images: a deep learning approach

Remote Sensing-based multi-time point imagery helps incorporate the potentiality of time series analysis in crop classification studies. Single timepoint imagery has limitations in this aspect due to a comparatively similar spectral signature of crops. Even in multi-time point imagery, the viability of optical datasets is thwarted by clouds in crop growing seasons. Hence, researchers have used the all-weather imaging capabilities of Synthetic Aperture Radar (SAR) for crop classification. This all-weather capability of SAR and the multispectral capability of Optical datasets combined with their time series data hold great potential in remote sensing-based crop classification. Nevertheless, the underlying patterns of the combined SAR and optical time series haven't been meticulously explored, especially for socio-economically important crops like sugarcane. Therefore, this study proposes using the Sentinel-1 and Sentinel-2 time-series datasets and crop phenology information to develop a Neural Network capable of identifying sugarcane crops in the sugar belts of Western-Uttar Pradesh. Studies are done for the year 2020 using a Long short-term memory (LSTM) neural network obtaining high accuracies in classifying sugarcane and non-sugarcane classes using optical and SAR time-series analysis datasets. The synthetic dataset has also been developed to enhance the convergence of the deep learning model, producing higher accuracy than utilizing the original dataset. Consequently, the study contributes to remote sensing-based crop classification and sets a foundation for further advancements in this domain.