CNN-based nonlinearity correction and background radiation suppression of an infrared spectral radiance measurement system

To support the infrared radiance calibration needs in fields such as remote sensing, climate change monitoring, and aerospace, the National Institute of Metrology of China has built up an infrared spectral radiance measurement system (ISRMS) in the 2.0-14.0 µm wavelength range. In this study, two main improvements were implemented in the ISRMS. The first is the nonlinearity correction of the measurement system. A one-dimensional convolutional neural network (1D-CNN) method was employed to correct nonlinearity by modeling the relationship between the measured spectral response of a standard blackbody and its theoretical radiance derived from Planck's law at each wavelength; validation results demonstrated that, compared to both piecewise linear interpolation and polynomial fitting methods, the 1D-CNN model achieved smaller deviations from the theoretical values, suggesting its effectiveness. The second is the suppression of the background stray radiation. A constant-temperature water-cooled plate was designed and installed at the entrance of the relay optics in the ISRMS, and results showed that the repeatability and reproducibility of the system were effectively improved. Furthermore, the measurement uncertainty of the ISRMS was evaluated. At a temperature of 798 K, the expanded measurement uncertainty (with a coverage factor <i>k</i>=2) was 2.74-1.26% in the 2.0-5.0 µm range, 1.26-0.54% in the 5.0-8.0 µm range, and 0.54-0.42% in the 8.0-14.0 µm range.