Study On Rice Growth Monitoring Based On Multi-scale Hyper-spectral Data

Crop growth monitoring plays an important role in precision agriculture,which lays a solid foundation for crop yield estimation and crop grain quality monitoring.In this study,we select late rice as the research object.Parameters reflecting crop growth such as SPAD,leaf area index(LAI),leaf nitrogen content(LNC),above ground biomass(AGB),and plant nitrogen accumulation(PNA),have been obtained through the field experiments of variable nitrogen fertilization in the fall of 2019-2020.ASD of leaf and canopy of rice,meanwhile,and images of unmanned aerial vehicle(UAV)were collected.Furthermore,according to the rice growth parameters,sensitive bands of spectral at leaf,canopy and UAV levels were analyzed and selected,and the estimation model of different parameters of the key growth period of rice was constructed.This study provides data support and a new way of thinking for understanding the growth of rice and obtaining the growth information of rice accurately.The specific content and main conclusions of the study are as follows:(1)The study of monitoring rice growth and nutrient parameters had been carried out based on hyper-spectral of leaf and canopy scale.The research results show: at leaf scale,parameters,such as SPAD and LNC,have a higher correlation with spectral at the heading stage.In the range of 520 nm-600 nm,the correlation coefficient between the spectral and SPAD is above-0.9.At canopy scale,there is a better correlation with parameters(SPAD,LNC,LAI,AGB,PNA)at differentiation stage in the range of visible light(454 nm-680 nm),while in the nearinfrared part correlation of each parameter is slightly different.Then the sensitive bands of the normalized differential spectral index are screened for all parameters at different scales.The sensitive bands of each parameter were concentrated in the green light near 550 nm,the "red edge" near 680-760 nm and the red edge region near 780nm-950 nm,which was related to the composition and the cell structure of green vegetation leaves.(2)Research on rice growth and nutrient parameter monitoring based on UAV imaging hyperspectral data has been carried out,too.In this paper,linear spectral unmixing(LSU)method was used to extract spectral data for plots of different nitrogen fertilizer treatment.And the relationship between the spectral data before and after unmixing and the parameters of rice has been analyzed.The research results show that LSU can effectively improve the monitoring accuracy parameters of rice under different coverage.For SPAD,when the rice is under low coverage(coverage <70%),the model accuracy can be improved to 0.889;for high coverage(>85%),the accuracy of the model is improved to 0.722.For LNC,the accuracy of the medium coverage(coverage >70% and coverage <85%)model is increased to 0.473,and the accuracy of the high coverage model is increased to 0.638.For LAI,the accuracy of the low coverage model is increased to 0.94,and the accuracy of the high coverage model is increased to 0.711.For PNA,the accuracy of the low coverage model is increased Increase to 0.74.(3)This research uses multiple linear regression,Gaussian processes regression(GPR),support vector regression(SVR)and partial least squares regression(PLSR)to construct estimation models of rice parameters.The results showed that at the leaf scale,the optimal monitoring models for rice leaf SPAD and LNC were both GPR,with accuracy of 0.896 and0.505,respectively.At the canopy scale,the optimal monitoring model for SPAD is PLSR with an accuracy of 0.636;the optimal monitoring model for LNC is GPR with an accuracy of 0.394;the optimal monitoring model for LAI is SVR with an accuracy of 0.607;The optimal monitoring model is GPR,with an accuracy of 0.82;the optimal monitoring model for PNA is GPR,with an accuracy of 0.723.At the UAV scale,the optimal model for parameter monitoring is GPR,and the accuracy of SPD,LNC,LAI,AGB,and PNA are 0.83,0.615,0.734,0.892,0.824,respectively.Comparing and analyzing the rice parameter spectrum monitoring models at different scales,it can be seen that in SPAD monitoring,the accuracy of the leaf model is better than that of the canopy and UAV hyper-spectral models;for the leaf area index,biomass and nitrogen accumulation,UAV spectroscopy performed best.