<p>Hyperspectral imaging is an innovative technology that uses a high-resolution camera to capture images of the Earth’s surface in hundreds of spectral bands. Each band corresponds to a specific wavelength of light, allowing for the detection of subtle differences in surface properties. This technology is beneficial for Land Use Land Cover (LULC) classification as it can accurately identify and map different land cover types and their boundaries. Despite its benefits, working with hyperspectral images to classify LULC can pose certain challenges due to the limited availability of training samples. Manual pixel-wise labelling of hyperspectral images is a time-consuming and resource-intensive process, presenting a significant challenge in land use and land cover (LULC) classification tasks. To mitigate this issue, we propose a novel Multiple Encoder Attention-on-Attention (MEAoA) framework that enhances classification performance, even when limited labelled data are available. Our approach first applies Principal Component Analysis (PCA) to reduce data dimensionality and extract the most informative spectral features. We then employ Hybrid Rice Optimization (HRO) to select the optimal bands for classification. The AoA mechanism, integrated into a multi-encoder architecture, enables the model to focus on salient spectral-spatial patterns, thereby improving its discriminative ability with fewer annotations. We evaluate our approach on three benchmark hyperspectral datasets—Indian Pines (IP), Salinas, and Pavia University (PU)—achieving an overall accuracy of 98.12% on the Salinas dataset, which demonstrates the method’s effectiveness in addressing the labelling challenge.</p>