Maize Drought Identification Based On Phenotypic Characteristics

Drought is an important disaster affecting maize production.It is of great significance to innovate the monitoring and early warning of drought in maize.This study was based on maize phenotypic features and BP neural network to identify maize drought,in order to explore a new method for drought identification in maize,and lay a foundation for improving the level of maize drought monitoring and identification.In this study,visible light imaging was used to collect maize images under different drought stress,and phenotypic feature variables were extracted from the maize images through programming.Multiple BP neural networks were used by ensemble learning to construct drought identification models for different growth and development stages of maize.Maize plants under conditions of different drought stress were identified by drought identification models.The study results show that the morphological features of maize are the most effective features to identify drought,and the features of color and texture are important features for identifying drought in the middle and later stages of maize growth.Under different conditions of drought stress,the phenotypic features of maize were significantly different and the phenotypic features of maize were the most unique under no-drought and drought-specific stress.Therefore,the drought identification models had a high degree of identification of maize under no-drought and drought-specific stress.The drought identification models for maize emergence-jointing,jointing-tasseling and tasseling-maturing three stages were all more than 90% accurate in recognizing the drought of maize.The errors of the drought identification models of emergence-jointing and jointing-tasseling two stages identifying different drought degrees of maize were less than 0.15,while the model of tasseling-maturing stage had relatively high errors in identifying different drought degrees of maize,between 0.2 and 0.5.The precision of the model for emergence-jointing stage identifying the different drought degrees of maize was more than 97%.The precision of the model for jointing-tasseling stage identifying the different drought degrees of maize was between 92% and 95%.The model of maize tasseling-maturing stage identified different maize drought degrees with the precision between 90% and 95% at training,and with the precision between 88% and 95% at testing.Using ensemble learning method to build maize drought identification model has obvious advantages in reducing training error,improving accuracy and precision compared with traditional single classifier model,and can effectively reduce the difference in identifying different drought degrees of maize.Comprehensive study results show that using maize phenotypes to identify maize drought has good results,but different phenotypic features have different contributions to drought identification of maize.There are some differences in the results of identifying different maize drought degrees used by the models constructed with ensemble learning of BP neural network,but the overall identification effect is good.