Research On Correlation Between Multispectral Images Gray Values And Chlorophyll Contents During Maize Growing And The Classification Of Its Growth Steps

Chlorophyll content is an important indicator to measure the growth state of crops.By using multi-spectral technology and SPAD-502 Chlorophyll Meter Model during the growth period of maize,the multi-spectral images gray values and chlorophyll contents of maize leaves are collected to establish fitting models.Then,we can use multi-spectral images gray values detect maize's growth state on large scale.In different growth stages of maize,the methods of fertilizing it are different,if we can establish a classification model of maize's growth steps base on multi-spectral images gray values,the model can provide scientific guidance for rational fertilization to increase maize's yield.In the fusion of multi-spectral bands,bands are mapped from original 8-dimensional space to 4-dimensional space,in this way,to extract its spectral characteristics,aiming to improve R-squared,average relative error rate of linear fitting model,and accuracy of the classification modelIn this paper,when establishing the fitting models of multi-spectral images gray values with chlorophyll contents,stepwise regression used to select multi-spectral bands with higher significance and impact analysis used to drop outliers,which are aimed at improving R-squared.What's more residual error analysis operated to test applicability.Then compare established fitting models,the linear fitting model is more suitable relatively,its R-squared is 0.93,average relative error is 9%.When establishing the classification model of maize's growth steps,we use original multi-spectral images gray values to train LinearSVC,KNeighbors,SVC and Ensemble models,then enhance the classifiers' precision based on gray images of confusion matrices and their transformed matrices,among all classifiers,Bagging classifier belonging to Ensemble model gets the highest precision 0.979.In the band fusion,the original 8 wave bands are fused into the final 4 bands by factor analysis and clustering analysis with the same results.The closest two are fused into one band,especially during factor analysis,a linear fitting model established using factor scores,improves R-squared up to 0.94,the average relative error rate being 8%,the accuracy of the Bagging classifier being up to 0.978.