This paper’s primary goal is to classify sugarcane crops using deep learning models using Sentinel imaging data. The sugarcane crop is a cash crop used in the production of ethanol and sugar. The classification of sugarcane is crucial for agricultural oversight and management. The conventional crop classification techniques that rely on limited ground-based data collection or manual examination take a lot of time and are usually unreliable. Consequently, an automated and effective approach is proposed, that requires the utilisation of imagery from satellites data and the deep learning techniques. The Convolutional neural network, ResNet and Inception-ResNet are the models applied for sugarcane classification using multispectral satellite data. The categorisation helps the farmers in timely decisions and management of the crop for good yield. The remote sensing through Sentinel data warns the farmers for pests and early detection of disease as categorisation of sugarcane is done prior. Also helps in monitoring the crop for better yield.The models used for sugarcane categorisation utilising multispectral satellite data are the convolutional neural network, ResNet, and Inception-ResNet. The classification aids farmers in making prompt judgements and managing their crops for maximum production. In order to help farmers, scientists, and other stakeholders make better decisions about crop management, yield forecasts, and resource efficiency in sugarcane farming, the initiative intends to improve sugarcane categorisation techniques. Deep learning models have identified the sugarcane despite the little dataset, and performance measures like as accuracy, recall, and F1 score are shown. The farmers' human interference may be resolved if more remote sensing imaging data is used and databases with Sentinel imagery are made available. In the future, integrating deep learning algorithms with a larger sugarcane or hyperspectral dataset might improve accuracy. Additionally, the classified sugarcane crop may be used for crop monitoring and crop quality evaluation. Techniques like data augmentation and transfer learning may be used to improve a model's resilience and generalisability by adjusting it for a comparable job and artificially growing a dataset. Data augmentation is the process of creating new samples by altering old ones using various techniques, like as rotation, scaling, and flipping, in order to improve a model's robustness and generalisability. There hasn't been much focus on using technology for self-supervised learning on tiny datasets, and more research is required in the field of agriculture. While few-shot and zero-shot learning techniques are useful for optimising performance from sparse data, data augmentation, suitable assessment criteria, and various model ensembles are essential for tiny samples and ought to be investigated in the agricultural remote sensing field.

https://www.taylorfrancis.com/chapters/edit/10.1201/9781003596707-4/sugarcane-categorisation-using-deep-learning-models-sentinel-dataset-mansi-kambli-bhakti-palkar is the link