GSA-UBS: A Novel Medical Hyperspectral Band Selection Based on Gravitational Search Algorithm

Medical hyperspectral images (MHSIs) provide the possibility of non-invasive disease diagnosis. However, due to the sparsity of MHSIs data in high-dimensional space, the “curse of dimensionality” arises, which reduces the efficiency and accuracy of data processing. Therefore, spectral dimensionality reduction has become a necessary step for MHSIs data analysis and application. To preserve the inherent properties of spectral bands, an unsupervised band selection algorithm based on Gravitational Search Algorithm, called GSA-UBS, is proposed in this paper to search for the best subset of bands. Considering the amount of information and redundancy of the candidate bands, we define an evaluation criterion consisting of a band distance matrix and an information entropy vector. Additionally, we design a simple discrete search strategy that enables GSA to directly obtain the original serial number of the selected bands instead of weighting the bands 0–1. Extensive experiments are conducted on two publicly available in vivo brain cancer MHSIs datasets, and the results demonstrate that GSA-UBS outperforms several state-of-the-art methods.