Cotton Yield Prediction Model Based on Hyperspectral Feature Fusion and Machine Learning

Cotton, as a globally significant economic crop, necessitates accurate yield prediction for field management and market regulation. This study aims to enhance the precision of cotton yield prediction by integrating hyperspectral feature fusion techniques with machine learning modelling approaches. A method widely used in finance and stock markets (MF‐DFA) is introduced into the agricultural sector to extract multifractal features from cotton canopy spectra. These features are fused with other spectral characteristics, and machine learning algorithms are employed to construct yield prediction models. The results indicate that (1) building prediction models based on fused features effectively improves the accuracy and stability of cotton yield predictions, outperforming models based on single features; (2) the fusion of Vegetation Index (VI) and Fractal Parameters (FP) features yields the best prediction results during the flower and boll period, with R 2 values of 0.6681 and 0.7009 for two consecutive years, respectively; (3) the Random Forest modelling approach exhibits higher prediction accuracy, and feature fusion parameters such as VI&FP and VI&FP&TR also perform well when applied to other model methods, demonstrating excellent generalisation capabilities. In conclusion, the prediction model based on hyperspectral feature fusion and machine learning provides a new and effective approach for cotton yield prediction.