Automatic Hyperspectral Image Clustering Using Qutrit Differential Evolution

A hyperspectral image serves as a valuable data source for ground cover analysis. However, determining the optimum number of clusters in hyperspectral images faces challenges due to the “curse of dimensionality” and the unavailability of ground truth images. Therefore, employing unsupervised cluster detection methods proves more advantageous in practical scenarios. This paper introduces a qutrit differential evolution algorithm for automatic clustering of hyperspectral images. The proposed algorithm incorporates qutrit Hadamard gates for population initialization and qutrit NOT gates for mutation. A qutrit-based crossover operation is also implemented following the normalization principle. The results of the proposed qutrit differential evolution are compared with the classical and qubit differential evolution algorithms utilizing different statistical tests and the F score. The Adjusted Rand Index serves as the fitness function and is used to validate the clusters. In most cases, the proposed algorithm outperforms the competing algorithms and the K-means algorithm with predefined cluster numbers.