<p>Band-pass multispectral infrared radiometry enables pixel-level retrieval of spectral emissivity and temperature at low cost and over a wide field of view. However, band-integrated measurements strongly couple radiance, temperature, and emissivity within each channel, leading to an underdetermined and highly non-convex inversion problem. In this work, we formulate the problem as a derivative-free, bounded multi-peak optimization constrained by Planck’s law and radiometric calibration. An adaptive escaping grey wolf optimizer (AEGWO) is proposed by incorporating chaotic initialization and stagnation-aware escaping to maintain population diversity. The objective function integrates band-pass energy consistency and channel-wise gain or bias compensation to suppress calibration and transmittance uncertainties. Experimental results show that under an equal computational budget, AEGWO achieves lower error, faster and more stable convergence, and higher stability than PSO, IGWO, LGWO, HHO, and SAGA. The proposed method ensures physical interpretability and robustness by enforcing energy consistency and applying channel correction. It is gradient-free, low-complexity, and deployable with few filters over a large field of view, making it suitable for rapid coating evaluation and in-situ monitoring.</p>