Study on the Calibration Transfer of Soil Nutrient Concentration from the Hyperspectral Camera to the Normal Spectrometer

Calibration transfer between hyperspectral cameras and conventional spectrometers remains a key challenge for quantitative soil nutrient analysis. This study collected 164 soil samples from three regions in Qingdao, China, acquiring spectral data from both a normal spectrometer and hyperspectral camera, along with total carbon (TC) and total nitrogen (TN) concentrations. A stable PLSR prediction model was established using the spectrometer data, and multiple calibration transfer strategies were tested, including single conventional algorithms, combined methods, and algorithms combined with spectral pretreatment. The prediction performance was evaluated by the absolute coefficient \(R_{t}^{2}\) and root mean square error of prediction (RMSEP). Results showed that the Repfile-PDS and Repfile-SNV methods achieved the best transfer performance. For Repfile-PDS, the optimal \(R_{t}^{2}\) and RMSEP were 0.627 and 2.351 for TC, and 0.666 and 0.297 for TN. For Repfile-SNV, the maximum \(R_{t}^{2}\) reached 0.701 (TC) and 0.722 (TN) with 120 standard samples. This study provides a feasible solution for calibration transfer between hyperspectral cameras and spectrometers, supporting rapid quantitative prediction of large-scale hyperspectral image data.

Journal of Spectroscopy