DesU-NetAM: optimized DenseU-Net with attention mechanism for hyperspectral image classification

The utilization of hyperspectral images (HSI) is expanding rapidly with the advancement of remote sensing technology. Accurately categorizing ground features using HSI is a crucial research topic that has garnered considerable interest. The high dimensional space, numerous spectral bands, and lack of labeled training data make categorizing hyperspectral images difficult. We provide a novel hyperspectral image categorization approach based on DenseU-Net to address these issues. The data is first normalized by separating the maximum value of the entire set of data by the average intensity of each pixel. Then we also propose an attention mechanism network. Because spectral and spatial data are extracted independently, there may be less interference between the two types of features in this network. The retrieved spectral and spatial data are integrated to categorize the data. Finally, DenseU-Net method parameters are adjusted using the Tuna Swarm Optimization (TSO) algorithm, which has a global search capability. This optimized DenseU-Net is then utilized to manage a hyperspectral image categorization approach effectively. Three standard HSI datasets utilized for experiments were acquired by various sensors at different acquisition times and used for classification studies. The comparison findings show that the proposed approach outperforms other deep learning methods models in classification effectiveness. © 2023, The Author(s), under exclusive licence to Bharati Vidyapeeth's Institute of Computer Applications and Management.