关键词: 玉米, 机器学习, 高光谱, 生理指标, 叶面积指数, 产量

Abstract: As one of the main crops in Henan Province, maize yield prediction holds significant importance for regional trade and food security. In order to establish a simple, timely, and accurate model for predicting crop LAI and yield, this study employs multiple linear regression (MLR), partial least squares regression (PLSR), and decision tree (DT) machine learning techniques. These techniques are combined with multi‑factor data, including maize physiological parameters (P1), spectral characteristic bands (P2), soil property parameters (P3), and meteorological parameters (P4), to construct estimation models for maize LAI and yield. The study results indicate that among the three machine learning methods, the LAI estimation accuracy during the grain filling stage is significantly higher than in other growth stages, while the yield estimation accuracy during the maturity stage is significantly higher than in other stages. Among the five multi‑factor combinations, the PLSR algorithm combined with the P1+P2+P3+P4 multi‑factor combination has achieved the highest accuracy, with the highest LAI estimation at Rv2=0.84 and RMSEv = 0.38, and the highest yield estimation at Rv2=0.79 and RMSEv = 982 kg/hm2. These findings provide technical support and theoretical basis for regional maize growth and yield prediction in the maize‑growing areas of northern China, enhancing prediction accuracy and efficiency, and are of great significance for agricultural production management and decision‑making.

Key words: maize, machine learning, hyperspectral, physiological indicators, LAI, yield