Unsupervised method for estimating the number of endmembers in hyperspectral images

Accurately determining the number of pure elements, or endmembers, in a mixture is crucial for unmixing applications in hyperspectral image processing. This work introduces a new unsupervised method, called ’Number of Endmembers by Energy Criteria’ (NEEC), for estimating the number of endmembers in homogeneous solutions of organic compounds in the liquid state, such as esters, hydrocarbons, and alcohols. The NEEC method utilizes eigenvalue analysis and incorporates an energy concept based on the eigenvalues of the sample correlation matrix. Experiments were conducted on both real and synthetic samples to assess the effectiveness of the proposed algorithm. Synthetic mixtures were created using a non-linear method. The results demonstrate that the NEEC method is highly effective, achieving 86.6% accuracy in estimating the number of endmembers. This highlights its potential for analyzing non-linear samples. This research contributes to the advancement of hyperspectral image processing techniques for unmixing applications. •Linear and non-linear hyperspectral sample processing capabilities.•Auto-adaptation to the input hyperspectral image without the need for external parameters.•Estimation of the number of endmembers incorporating the functional concept based on the eigenvectors of the hyperspectral sample correlation matrix.