Joint Optimization of Band Fusion and Multi-Level Thresholding Via a Hybrid Metaheuristic Framework: Genetic-Cauchy and Runge-Kutta-Driven Adaptive Intensive Mutation (GC-RUNIM)

Abstract

The analysis of multispectral satellite images requires a complex optimization process due to the high-dimensional data structure and spectral overlap among bands. In this study, a novel hybrid adaptive metaheuristic method, called Genetic-Cauchy and Runge-Kutta-driven Adaptive Intensive Mutation (GC-RUNIM), with a unified objective function (joint objective function) is proposed to simultaneously optimize band fusion and multi-level thresholding processes in multispectral satellite imagery. The proposed algorithm integrates a GA-based global exploration phase with a Runge-Kutta (RUN)-based local exploitation phase within an adaptive framework. The proposed framework incorporates an Adaptive Intensive Mutation (AIM) operator that adaptively regulates local perturbation intensity according to solution quality and search progress, improving exploitation efficiency during joint band fusion and multi-level threshold optimization. Experimental studies are conducted on sentinel-2 satellite imagery with a spatial resolution of 2500 & times; 1902 pixels and 32-bit depth. The obtained results demonstrate that GCRUNIM achieves a value of 0.3901 +/- 0.0064 in the inter-class contrast metric, providing clear superiority over competing algorithms (SCSO: 0.3577, WOA: 0.3639, DGWO: 0.3005). The SSIM (0.9543) and PSNR (35.9537) values indicate that the algorithm preserves spectral fidelity while maximizing class separability. Furthermore, in comparison with traditional principal component analysis-based approaches, the proposed joint-objective structure significantly improves contrast performance from 0.1247 to 0.3901. The full source code of the proposed GC-RUNIM algorithm and implementation details are available as open source at https://github.com/mfaruk-sahin/GCRUNIM.git to support scientific reproducibility and future research.