| Chemical defoliation and ripening make cotton defoliated and mature in batches,which is a necessary link and an important prerequisite for mechanized cotton harvesting.The monitoring of traditional cotton defoliation effect is mainly through manual field investigation,which has problems such as low efficiency and strong subjective factors.In this paper,four types of agronomic data,including the relative content of chlorophyll in cotton leaves,the actual photosynthetic efficiency of cotton leaves(phi2 parameter),hyperspectral characteristics and texture characteristics of cotton leaves,were obtained by designing a cotton defoliant stress test.The learning model is trained to build a monitoring model for the degree of defoliant stress in cotton,which provides a new method for efficiently and accurately judging the degree of defoliant stress in cotton,and provides scientific guidance for the secondary variable spraying of defoliants and reducing the amount of chemical used.,The main contents are as follows.(1)Collect cotton leaf-related characteristic data by designing a chemical stress degree test measured by defoliant stress days.The experimental site was set up in the spraying area with defoliant treatment and the control area with clear water,and collected the relative chlorophyll content(Soil and plant analyzer development,SPAD)and cotton Leaf phi2 parameter,collected hyperspectral images of cotton leaves at 1,3,5,7,and 9days after spraying in the experimental site at the same time,extracted the spectral reflectance of cotton leaves in the range of 450nm-950 nm,and used SG algorithm to smooth the curve.The change of reflectance curve with the days of chemical treatment was analyzed,and the preprocessed cotton leaf hyperspectral image was subjected to principal component analysis,and the first principal component image was selected for the extraction of cotton leaf texture features.Correlation analysis was carried out between cotton leaf SPAD value and phi2 parameter and chemical treatment days,respectively.The cotton leaf SPAD value and phi2 parameter in the spraying area had strong correlation with spraying days,and the correlation coefficients were all above 0.96.26 spectral feature parameters,8 texture features of cotton leaves,SPAD value and phi2 parameter of cotton leaves measured in the spraying area,a total of four types of feature parameters were extracted to form a multi-feature data set for training and testing of machine learning models.(2)Using multi-feature datasets to train and test three machine learning models,Support Vector Machine(SVM),Naive Bayesian Model(NBM)and Random Forest(RF),to build a monitoring model for the degree of defoliant stress on cotton.The results show that the classification accuracy of the SVM model for the degree of cotton defoliant stress is 97.168%,the overall classification accuracy of the NBM model is 93.642%,and the overall classification accuracy of the RF model is accurate.The rate is 95.722%.(3)The Relief F algorithm is used to reduce the dimensionality of the multi-feature data set,and the first 27 feature parameters with higher weight values are selected to form the feature data set for the training and testing of the three machine learning models.The results show that the SVM model among the three models is used.The ability to discriminate the degree of cotton defoliant stress was good,the overall classification accuracy was 93.422%,the NBM model classification accuracy was 90.761%,and the RF model classification accuracy was 91.533%.In conclusion,the SVM model has the best classification accuracy,which can efficiently and non-destructively monitor the stress degree of defoliants on cotton,formulate the application plan of defoliants,and provide guidance for the realization of precision agriculture by scientific application. |
PDF Full Text Request