In order to adequately correct the response nonlinearity of infrared detectors, the multi-point method is often used to replace the traditional two-point method for the radiometric correction of satellite-borne infrared spectral imagers, and the number of required calibration temperature points increases with the increase of the dynamic range. Aiming at the redundancy of temperature points in multi-point calibration, a calibration temperature point selection method based on relative standard deviation threshold division is proposed. Firstly, the experimentally obtained instrumental response curve is divided into multiple segments using the standard deviation method, and each segment is linearly fitted using the least-squares method to ensure the fitting accuracy; secondly, the calibration coefficients derived from the fitting are inversely performed to determine the blackbody temperature points for two-point calibration in each segment. The same method is used to divide the nonlinear response curves within all spectral segments, and the corresponding calibrated temperature points for all sub-segments of each segment after division are found. Finally, all the temperature points are combined and compressed to determine the final calibration temperature points. In order to verify the feasibility of the method, simulation experiments and real experiments were carried out respectively. The nonlinear response of a single spectral band was simulated, and the standard deviation method and the bisection method were used to divide multiple bands under the same division conditions (relative standard deviation of 1%), and the least-squares method was used for linear fitting after division, resulting in the need for 14 calibrated temperature points after the division of the standard deviation method and the need for 22 calibrated temperature points after the bisection method, and the degree of nonlinearity of the two methods after the division of the fit was less than 1%. It can be proved that the standard deviation method can reduce the number of calibration temperature points under the same division requirements. When the number of calibration points is the same (20), the standard deviation method and the uniform division of the two methods are used to fit the nonlinear response curve of a single spectral band. The maximum value of nonlinearity was 0.404 6% after fitting by standard deviation method and 1.059 9% after uniform division and linear fitting. It can be proved that when the number of calibration temperature points is the same, the standard deviation method is better than the uniform division. Combined with the designed infrared spectral imager, 17 spectral channel response curves at mid-wave were simulated, and the nonlinear response curves of each spectral channel were divided into multiple segments using the standard deviation method, and the 17 spectral segments were divided into 72 segments under the condition of 1% relative standard deviation. The blackbody temperature points used for two-point calibration were derived for each sub-segment. Since the temperature points derived from the inversion of each segment are not necessarily unique, and the same blackbody temperature point can be adapted to radiometric calibration in different bands. Therefore, the temperature points can be combined according to their frequency of occurrence. The combined calibrated temperature points (90) are compressed. Using the compressed 10 calibrated temperature points to calibrate 17 spectral bands, the nonlinearity of the calibrated multiband linear response curve is less than 1%. Infrared radiation acquisition experiments were conducted, and the nonlinear response curve of the mid-wave infrared camera in the temperature range of 308.15~396.15 K in the blackbody grew linearly to saturation was measured. Using the standard deviation method for multi-segment division, 14 calibration points are required after division, and the maximum value of the nonlinearity after linear fitting is 0.61%, and the mean value is 0.48%. The results of both simulation and measurement experiments show that this method can be effectively used for the selection of temperature points for the multi-segment two-point calibration method, which is of certain significance for reducing the burden of the infrared spectral imager in orbit, lowering the number of temperature points for blackbody calibration, and realizing efficient on-planet blackbody nonlinear calibration.