Inverse model for the photosynthetic pigment content of peanut leaves using coupling algorithm

Peanut is one of the most important economic and oilseed crops in world. It is vital to ensure the stability of the production for national oil security, due to rich in oil, protein, dietary fiber, and micronutrients with multiple functional components and very high nutritional value. Furthermore, pigments (such as the chlorophyll and carotenoids) are greatly contributed to the photosynthesis in plants. Among them, the content of chlorophyll is very closely related to the photosynthetic capacity, growth and development, as well as nutritional status of vegetation, representing the stress, growth and senescence symptoms. Carotenoids can absorb and transfer the solar radiant energy, and also dissipate the excess energy to protect the photosynthetic system, particularly for the exposure to strong sunlight. It is a high demand for the accurate acquisition and prediction of photosynthetic pigment content for fine planting management. Hyperspectral technology has been widely used for the rapid detection of crop physiological indicators and growth in recent years, due to its high efficiency and non-destructive means. In this study, an inversion model was established for the photosynthetic pigment content of peanut leaf using coupling algorithm. The research object was taken as the peanut canopy leaves at the flowering hypocotyl stage. The data source was collected by ASD Field Spec4 field portable hyperspectrometer. A systematic investigation was also conducted to determine the photosynthetic pigment uptake of peanut canopy leaves. The raw spectral data was preprocessed using Savitzky-Golay smoothing (SG) combined with the standard normal variables (SNV). Seven single screening algorithms of feature wavelength selection were compared, including the correlation coefficient analysis (CC), successive projections algorithm (SPA), random frog (RF), iteratively retains informative variables (IRIV), uninformative variable elimination (UVE), competitive adaptive reweighted sampling (CARS), and genetic algorithm (GA). Four models were combined with the partial least squares regression (PLSR), support vector regression (SVR), gradient boosting decision tree (GBDT), and eXtreme gradient boosting (XGBoost). Full wavelength models were used for the comparative analysis, where the evaluation indexes were used as the determination of coefficients (R²) and root mean square error (RMSE). Three optimal algorithms were selected for pairwise coupling. The results showed that (1) In the single algorithm experiments, the superior performance was achieved in the UVE, IRIV and GA. Specifically, the R² reached 0.591, and 0.565 in the prediction model of chlorophyll and carotenoid content, respectively. There was no significant improvement in the accuracy in the CC, RF, SPA, and CARS. (2) In the coupled algorithm experiments, the UVE-IRIV, GA-IRIV, and GA-UVE were greatly improved the prediction accuracy, while effectively simplified the modeling complexity. Among them, the GA-IRIV-XGBoost inversion model of chlorophyll content reached the highest accuracy with R²=0.622 and RMSE=0.235 mg/g, whereas, the UVE-IRIV-XGBoost inversion model of carotenoid content reached the highest accuracy with R²=0.575 and RMSE=0.056 mg/g. (3) The prediction accuracy of chlorophyll was better than that of carotenoids. In summary, the coupling algorithm can effectively compress the variables for the simplified model with the improved robustness. A rapid, nondestructive and accurate prediction was achieved in the photosynthetic pigment content in peanut leaves. The better effect was coupled for the high accuracy and stability of the improved model using the variable screening. The structure of improved model was also simplified to extract a small number of effective information variables. The finding can provide the promising data base to detect the photosynthetic pigment content of peanut leaves for the yield estimation of peanuts.