Namib Beetle Firefly Optimization enabled Densenet architecture for hyperspectral image segmentation and classification

Abstract

ABSTRACT Many organisations have concentrated on how hyperspectral images allow for the automatic pixel-level classification and segmentation since each pixel’s underlying spectrum is abundantly documented. Due to the unpredictable nature of the spectrum and the noise in the hyperspectral data, this task is very challenging and calls for specific solutions. The hyperspectral picture segmentation procedure in this instance makes use of the newly developed Namib Beetle Firefly Optimization (NBFO) method, which was created by combining the Namib Beetle Optimization method (NBOA) and Firefly Algorithm (FAO) for tackling optimisation issues. Therefore, in order to segment the images, the U-Net++ model is used. The DenseNet model, which is likewise trained using the NBFO approach, is then used to classify the segmented images. Utilising hyperspectral image segmentation techniques, the NBFO-driven DenseNet model surpassed the competition, resulting in a True Positive Rate (TPR) of 0.906786, a Positive Rate (FPR) of 0.889466, and a False Pixel Accuracy (FPA) of 0.931562.