Correlation Analysis And Research On Multi-band Images Fusion And Chlorophyll Contents

Modern agriculture introduces new technologies and scientific management methods in order to obtain crop growth parameters,and improves crop yields.This is a trend of agricultural development today.The non-destructive detection of chlorophyll contents and the appearance of multi-spectral image technology have provided conveniences for crop growth monitoring.Chlorophyll content is one of the important indicators to measure the growth status of crops.During the growth of the maize in a trial field,in this research,we use a multispectral camera and a SPAD-502 chlorophyll content measuring instrument to collect the multi-spectral images and chlorophyll content values of the different leaves of maize individuals,and then,extract the gray values of the areas on the maize leaves images and,based on their correlation analyses establish a relation equation between the two above so that large-scale rapid non-destructive detect of chlorophyll contents can be implemented autonomously on maize leaves by image recognition.Since the chlorophyll content values of maize leaves have a decreasing trend in the whole growth period in the experiment,we classify the entire growth period of the maize into nine different stages,based on which we can determine the corresponding growth stages of the maize by detecting their chlorophyll contents,thus providing support for effectively monitoring maize growth status.In order to reduce the multi-spectral band information redundancy,this study uses factor analysis to fuse the original 8-dimensional space(it corresponds to 8 bands)into the 4-dimensional(4 combinations of 8 bands)space to reduce the error rate of the prediction model of chlorophyll contents of maize and improve the accuracy of the classification model of maize growth stages.The cross-validation is used to optimize the Elman neural network.Experimental results show that the multi-band images fusion based on factor rotation combining the cross-validation optimization of Elman neural network can produce good effect on the chlorophyll contents prediction with the smallest error of 0.0024,in comparison with the traditional linear and nonlinear regressions,which is decreased by 0.1676,0.1776,respectively,in the experiments.In order to improve the classification accuracy of maize growth stages,based on multi-band images fusion,the random forest method is employed to classify the growth stages of maize after being optimized by cross-validation optimization.Experimental results show that multi-band images fusion combining cross-validation optimization can improve the classification accuracy of random forest model.Its classification accuracy is 96.67%,with Hamming Loss of 0.0037,and compared with the original multi-band images without cross-validation,the classification accuracy is improved by 0.0279 and Hamming Loss is reduced by 0.0031.