MetaMamba: Meta-learning with mamba for few-shot vegetation species classification using UAV-based hyperspectral imagery in desert rangeland

Effective monitoring of desert rangeland ecosystems is of crucial significance to regional ecological security. Unmanned aerial vehicle (UAV) hyperspectral remote sensing provides an effective means for the fine identification of vegetation species. However, in practical applications, vegetation classification using hyperspectral images often faces the problem of limited labeled samples, which makes it difficult for traditional deep learning methods to obtain stable and accurate classification results. To address these issues, this study proposes a meta-learning with Mamba (MetaMamba) method for vegetation classification in desert rangeland. This method constructs a local-global dual-branch structure in the feature extraction stage to achieve effective fusion of local and global spatial context information. Specifically, the local branch uses convolutional neural networks (CNNs) to extract fine-grained spatial features, while the global branch models long-distance spatial dependencies based on the Mamba model. Additionally, a meta-learning strategy is introduced to enhance the feature learning and generalization abilities of the model under few-shot conditions. Experimental results show that the proposed method outperforms existing methods across multiple evaluation metrics. The overall classification accuracy (OA) reaches 90.85%, the average accuracy (AA) reaches 91.67%, and the Kappa coefficient reaches 87.77%. The method shows good stability and adaptability under different sample sizes. The MetaMamba model can achieve high-precision classification of desert rangeland vegetation species under few-shot conditions, providing an effective technical approach for ecological monitoring and rangeland resource management.