Study On Inversion Of Soybean Canopy Chlorophyll And Nitrogen Element Content Based On UVA Multispectral Imagery

As a grain and oil crops,soybean plays a key role in ensuring food security.During the growth and development of soybean,the chlorophyll and nitrogen content are important agronomic parameters that reflect the growth status of soybean.Through the use of unmanned aerial vehicle(UAV)remote sensing platforms,the chlorophyll and nitrogen content of soybean can be rapidly and non-destructively monitored,providing timely guidance for soybean field management based on these agronomic parameters.This study was conducted in Huojia County,Henan Province from July to September 2022.The DJI Phantom 4 RTK drone was used to obtain multispectral images of different soybean varieties during their main growth stages,and the chlorophyll and nitrogen content in the soybean canopy was measured simultaneously.The multispectral images were preprocessed,and 17 multispectral vegetation indices were extracted.The relationship between vegetation indices and chlorophyll and nitrogen content was analyzed,and quantitative estimation models for chlorophyll and nitrogen content based on vegetation indices were established,including the univariate regression(UR)model,stepwise regression(SR)model,partial least squares regression(PLSR)model,and BP neural network model.The main results of the study are as follows:(1)The statistical analysis of the chlorophyll and nitrogen content of four soybean varieties during three growth stages shows that the chlorophyll and nitrogen content of the four soybean varieties increase rapidly from the V2 stage to the R2 stage,and show differences from the R2 to R5 stage.Among them,the chlorophyll and nitrogen content of Pudou 857 and Zhengdou 365 increase slowly,while the changes in chlorophyll and nitrogen content of Wansu 1208 and Zhonghuang 301 are not obvious.The chlorophyll and nitrogen content are significantly positively correlated in all three growth stages and the full growth period.The study shows that both soybean variety and growth stage significantly affect the chlorophyll and nitrogen content of soybean.(2)Regression models for chlorophyll content were constructed for four soybean varieties during the three growth periods using UR regression model,SR regression model,PLSR regression model,and BP neural network model.During the V2 period,the BP neural network model had the best inversion performance among the four models of four soybean varieties,with the determination coefficient R2 of both the modeling and validation sets above 0.95 and the RMSE below 0.2;During the R2 period,the BP neural network models of Pudou857,Wansu 1208,and Zhengdou365 had the best inversion performance.The determination coefficients R2 of the modeling and validation sets were above 0.94,and the RMSE was below 0.2.However,the superiority of the PLSR model and BP neural network model of Zhonghuang 301 was not significantly different.The determination coefficients R2 of the modeling and validation sets were both above 0.96,and the RMSE was below 0.3;During the R5 period,the inversion model was influenced by variety differences.The BP neural network models of Wansu 1208 and Zhonghuang 301 had the best inversion performance,with a determination coefficient R2 of over 0.93 for the modeling and validation sets and RMSE of below 0.2.The PLSR models of Pudou857 and Zhengdou365 were the best,with a determination coefficient R2 of over 0.90 and a RMSE of below 0.2 for the modeling and validation sets.Comprehensive analysis shows that using BP neural networks during the V2 and R2 periods can quickly and accurately invert soybean chlorophyll content,providing guidance for field nutrition diagnosis.In the R5 period,due to the differences between different varieties,it is necessary to study the chlorophyll content of a single variety.(3)UR regression models,SR regression models,PLSR regression models,and BP neural network models were constructed for four varieties in three growth stages.Among the four of nitrogen element content inversion models of soybeans in the three periods,the inversion effect of the BP neural network model was the best,and the determination coefficients R2 of the modeling set and the verification set were both above 0.90,and the root mean square error RMSE was below 0.2.In the R5 period,there were differences in the inversion models of nitrogen element content in different soybean varieties,but the BP neural network model also showed good results,further indicating that compared with linear regression,the BP neural network is more conducive to revealing the changes of physiological and biochemical complex internal parameters such as vegetation nitrogen elements.Through comprehensive analysis,the use of BP neural network can quickly and accurately invert the nitrogen elements of soybeans in V2,R2,and R5 periods,providing guidance for soybean field formula fertilization.