Inversion Of Chlorophyll Content Based On Hyperspectral Remote Sensing Image In Soybean

Chlorophyll content is an important indicator of plant growth and development,which can indirectly reflect the growth status of plants.In soybean production,it is of great significance to establish a rapid and efficient method for chlorophyll content determination,which can be used to monitor the growth and development status of soybean,and to adjust cultivation practice accordingly to achieve high yield.Traditional methods for chlorophyll content determination are time-consuming,labor-intensive,scale-limited and destructive,which cannot meet the requirements of modern agricultural production.With the development of UAV and hyperspectral technology,it is possible to obtain soybean chlorophyll content rapidly and with high-throughput.At present,there are few reports on the reversion of soybean chlorophyll content at different growth stages by using hyperspectral sensor mounted on UAV.In this study,the hyperspectral data at flowering,podding,bulging and mature stages were obtained by hyperspectral sensor mounted on UAV,and the chlorophyll content data of soybean at the four stages were also collected.Using linear regression,exponential regression,logarithmic regression and decision tree algorithm,as well as multiple linear regression,ridge regression,LASSO regression,random forest and neural network algorithm,the univariate and multivariate inversion models of soybean chlorophyll content were established,respectively.The main results were as follows:1.The spectral curve of soybean changed with growth stages and chlorophyll content levels.Based on the results of trilateral parameters,it was found that there was a certain relationship between trilateral parameters and the growth and development of soybean,and trilateral parameters could be used as indicators for judging the growth and development status of soybean.Meanwhile,the chlorophyll content of soybean was closely related to the trilateral parameters,and trilateral parameters might change accordingly with the rise and fall of chlorophyll contents of soybean.2.The first derivative of the original spectrum could improve the correlation between the hyperspectral data and the chlorophyll content of soybean.The first order differential optimal band was the best hyperspectral parameter for predicting soybean chlorophyll content,and the best bands for flowering,podding,bulging and mature stages was 655 nm,508 nm,518 nm and 683 nm,respectively.3.Comparison of models showed that the model based on decision tree algorithm showed the best performance and could be used as a reference algorithm to establish the inversion model based on univariate regression.4.Compared with univariate regression models,multivariate regression models showed better performance in inversion effect and model stability.Among all multivariate regression models,the models constructed by random forest algorithm showed the best inversion effect on soybean chlorophyll content in most cases.Among all inversion models constructed in this study,the inversion ability of the model established with random forest algorithm at whole stage was the best.The random forest algorithm might be used as a reference algorithm for soybean chlorophyll content inversion.5.The inversion ability of the models established in the four growth periods is relatively close,and there is no obvious stage difference;The inversion effect of the model established by the whole period data is better than that in a single period.In summary,this study discussed the changes of trilateral parameters under different chlorophyll levels and different periods,univariate and multivariate inversion models of soybean chlorophyll content at different growth stages were established with different algorithms,and the inversion ability and stability of different models were analyzed in this study.The results should have certain significance for the retrieve of soybean chlorophyll using hyperspectral remote sensing technology.