Inversion of Potato Chlorophyll Content Based on Radiation Transfer Model and Machine Learning Algorithm

Abstract

Chlorophyll content greatly correlates with crop growth status, nitrogen content, yield, etc. It provides valuable information about plant physiology and can reflect the aging process of plants. In this study, we obtained the leaf hyperspectral data and leaf chlorophyll content data at the potato tuber formation stage, tuber growth stage, and starch accumulation stage based on the 2022 potato N and K fertilizer gradient experiment at the Kershan Farm of Qiqihar Branch of Heilongjiang State Administration of Land Reclamation. Machine learning regression algorithms and RTM combing hybrid methods:GPR_PROSPECT, KRR_PROSPECT, and PLSR_PROSPECT, were used to construct the inversion model of potato leaf chlorophyll content after the active learning algorithm selected the optimal modeling training data set from the RTM simulation dataset. The results demonstrated that the AL techniques could enhance inversion precision and modeling efficiency and mitigate some discomfort associated with hybrid inversion methods. Among the three hybrid methods, the GPR_PROSPECT method performed the best throughout the entire growth stage (R2=0.978, RMSE=2.869ug/cm2, and NRMSE=4.202%) and was best suited for the inversion of potato chlorophyll content. The GPR_PROSPECT model also showed promising results when applied to different growth stages. The results showed that the GPR_PROSPECT inversion model could provide a reference for potato growth monitoring and health status evaluation.

Keywords: Potato, Chlorophyll content, Hyperspectral remote sensing, Hybrid method, Active learning