Abstract: In multispectral radiometric thermometry, the development of data processing as a key step has been limited by the unknown emissivity of the target. The traditional solution is to construct a hypothetical model of emissivity based on the intrinsic connection between emissivity and wavelength or temperature. However, due to the complexity of the actual situation, when the predetermined hypothetical model deviates from the real emissivity characteristics of the target object, such deviation will be amplified in the temperature measurement process, which leads to significant errors. Therefore, the core challenge of multispectral radiometric temperature measurement technology lies in constructing an emissivity self-correction mechanism to achieve higher temperature measurement accuracy without the need of pre-set material emissivity models through multispectral data fusion and adaptive temperature inversion algorithms. In this paper, a hybrid optimization algorithm based on improved NRBO-GA(hybrid Newton-Raphson-based optimizer and genetic algorithm) was proposed, and the key innovation of this algorithm lied in the establishment of a constrained optimization mathematical model for multi-wavelength radiometric temperature measurement. Firstly, the spectral emissivity distribution was parameterized, and the spectral emissivity population that meet the radiative transfer constraints was constructed as the initial solution set; then the adaptive objective function was used to quantitatively evaluate the fitness, and the evolutionary algorithm was implemented to dynamically search for the optimum in the feasible solution space; finally, the global convergence mechanism guided by constraints boundaries was used to obtain the temperature-emissivity joint inversion which conformed to the law of black-body radiation and had an optimal fitness. The method ccould synchronize the target spectral emissivity with the real temperature without the need to preset the emissivity model. Simulation experiments were carried out for 6 typical emissivity models, and the results show that the algorithm is highly adaptable to the typical spectral emissivity distribution patterns, such as monotonically increasing/decreasing and multipolar fluctuations. The average relative error of the inversion temperature of the algorithm is lower than 1.31% in 900 K and 1 000 K standard blackbody tests. Applying it to the processing of rocket engine plume flame temperature, the maximum relative error of the inversion temperature does not exceed 0.63% when the set temperature is 2 490 K. The simulation and experimental results fully prove that the algorithm has excellent performance in the inversion of target true temperature and firing rate.