Research on a Hyperspectral Rice Yield Estimation Model Based on Random Forest

This article analyzes and selects 18 characteristic parameters in different growing period and vegetation Index, and establishes 6 random forest model algorithms with different types of parameters for rice yield estimation. Main to analyze the influence of the two key parameters of nTree and Depth in the random forest algorithm on the model accuracy, and analyzing the rice yield estimation accuracy of the six random forest models. The experimental results indicate that: First, the nTree parameter has little effect of the model accuracy. Second, a small Depth value has a significant impact on model accuracy. When the Depth value is greater than 6, the model accuracy tends to stabilize. Third, The training effect of the RF-der model is the best among the six models, while the accuracy of the RF-R and RF-ref models is relatively low. Forth, the first derivative of rice canopy spectrum and vegetation index are more suitable for estimation. Fifth, the spectral indices of heading and filling stages are more suitable for estimating rice yield.