Estimation of anthocyanin content in maize at different growth stages based on hyperspectral technology
Anthocyanin is an important pigment in maize.Convenient and non-destructive estimation of anthocyanin content is of great significance for monitoring maize growth.Based on the Anth values and hyperspectral data of maize canopy leaves at jointing stage,flare opening stage,tasseling stage and milk-ripe stage in Guanzhong area,multiple single-factor models and multi-factor models were established.The results showed that the original spectral characteristics of maize leaves at different growth stages were generally consistent,but locally different.The correlations between the characteristic bands and Anth values of the transform spectrum were better than those of the primary spectrum,and the characteristic bands of the first derivative spectrum was the best.The successive projections algorithm(SPA)had a good dimension reduction effect,and the selected modeling parameters rangeed from two to 27.The accuracy of the optimal single factor model and multiple linear re-gression model was the best at tasseling stage,followed by jointing stage and flare opening stage,and the worst at milk-ripe stage.Among all the models,the sparrow search algorithm-extreme learning machine regression(SSA-ELMR)model based on the first derivative spectrum at the tasseling stage was the optimal model in this study.The R2 values for modeling and verifica-tion of this model were 0.847 and 0.895,respectively,and the corresponding nRMSE values were 6.44% and 7.21%.The results of this study indicate that the tasseling period was the optimal period for inverting the anthocyanin content in maize leaves,and the extreme learning machine could further improve the accuracy of traditional models.
万方数据
维普
