We propose an efficient hyperspectral classification framework for the Krishna River Basin using LSTM-based auto-labelling and hybrid ensembles, achieving 99% accuracy and enabling robust land cover analysis in data-scarce remote sensing scenarios.