Study On Estimation Model Of SPAD Value Of Rice Leaves Based On Color Characteristics

[Objective] To explore the relationship between the color characteristics of rice leaves and the relative chlorophyll content (SPAD), so as to lay a theoretical foundation for the rapid acquisition of SPAD values of rice leaves. [Methods] The rice leaves were equally divided into three parts: leaf base, leaf middle and leaf tip, and their color characteristics and SPAD values were obtained respectively. Through image processing technology and correlation analysis method, five features highly related to SPAD value are selected from 18 color features as model input and SPAD value as output to build four models, namely BP neural network regression model, convolutional neural network regression model (CNN), weighted least squares regression model (W+LS) and random forest regression model (Rand+Grid+RF) based on random search and grid search parameter optimization.[Results] Rand+Grid+RF model has the best prediction effect for leaf base, and its model evaluation indicators RMSE, R2, R2_Adjusted and MAE are 1.764, 0.609, 0.555 and 1.435 respectively; Rand+Grid+RF model has the best prediction effect for the middle part of the leaf, and its model evaluation indicators RMSE, R2, R2_Adjusted and MAE are 1.843, 0.718, 0.694 and 1.431 respectively. For the tip part, W+LS model has the best prediction effect, and its model evaluation indicators RMSE and R2, R2_Adjusted and MAE are 1.558, 0.680, 0.635 and 1.177 respectively. [Conclusion] The experimental results show that Rand+Grid+RF model has a good prediction effect on the SPAD value of the blade base and middle part, and W+LS model has a good prediction effect on the tip part.