Research On Drip Irrigation Winter Wheat Estimation Model Based On Hyperspectral Imaging Technology

[Objective]Water shortage is the main bottleneck for sustainable agricultural development in arid and semi-arid areas in northwest China.Water is the main component of crops and an important limiting factor for crop growth and development.When the water is scarce,the external morphology and internal physiological and biochemical effects of the crops will be affected,which will directly affect crop growth,yield formation and quality.Therefore,timely and accurate monitoring of crop water conditions is of great significance to the reasonable irrigation,quality improvement and yield increase of crops.The water content of crops directly reflects the water status of crops.Traditional methods of obtaining moisture content for crops are usually drying,which is time-consuming,labor-intensive,and extremely destructive.The number of samples is limited and it is difficult to meet the needs of modern agricultural development.Hyperspectral imaging technology combines spectral information with image information to obtain a large amount of spectral and spatial dimensional information.It has the characteristics of multi-scale,multi-band and high resolution.This paper accurately estimates moisture content based on hyperspectral data of winter wheat leaves and canopy in a typical drip irrigation in Xinjiang,and establishes a water content estimation model to achieve non-destructive estimation of water content.This paper also provide technical support for timely monitoring of crop water surplus-deficit and shortage,and guidance for the formulation of a scientific and rational irrigation system.[Method]This study takes Xinjiang typical drip irrigation winter wheat as the research object,and establishes the hyperspectral water content estimation model of leaves and plants as the main research goal.A total of 5 water treatments W1(150 mm),W2(300 mm),W3(450 mm),W4(600 mm),and W5(750 mm)were set.Field sampling and indoor analysis and testing methods were used to obtain wheat agronomic traits,and portable hyperspectral images were used to obtain wheat top 1 leaf(L1),top 2 leaf(L2),top 3 leaf(L3)and canopy hyperspectral data.Smooth,derivative,reciprocal,root,logarithmic and other data processing methods were used for transforming the raw spectral reflectance with the canopy of wheat leaves,to analyze and compare three methods:Simple linear regression(SLR),Principal components regression(PCR),and Partial least squares regression(PLSR).The effect of model estimation is calculated by using typical correlation analysis(CA)and continuous projection algorithm(SPA)to select feature bands,and to construct an optimal estimation model of wheat water content and carry out accuracy tests.The results can provide technical support for real-time monitoring of regional crop water conditions and scientific and rational irrigation system formulation.[Results]The main conclusions of this paper were as follows:(1)The agronomic traits of winter wheat change with the advancement of the growth process.From jointing stage to mid-grouting stage,plant water content(PWC)and leaf water content(LWC)continue to decrease;plant dry matter accumulation(PDMA)shows a continuous increase.Leaf area index(PLAI and LLAI)showed a trend of increasing first and then decreasing,the lowest in jointing stage and the highest in flowering stage.The SPAD value increased first and then decreased,the highest during the flowering stage and the lowest during the mid-grouting stage.From the perspective of different water treatments,with the increase of irrigation amount,PWC and LWC showed a continuous increasing trend;SPAD values showed an increasing trend in the range of W1-W4,and may slightly decrease during W5 treatment;while PDMA,PLAI and LLAI showed a trend of increasing first and then decreasing,the highest in W3 and the lowest in Wl,the overall performance is W3>W4>W5>W2>W1.From different leaf positions,LWC and LLAI showed L3>L2>L1 as a whole,while SPAD value showed L2>L3>L1 at jointing stage,and L1>L2>L3 in other stages.According to the correlation analysis of agronomic traits of plants and different leaf positions,the water content of L1,L2,and L3 were significantly correlated with PWC at each growth stage.The accuracy of the model using the water content of the top 1 leaf and the plant water content reached 0.8489,can better estimate the water status of the entire plant.(2)The correlation between the original spectral reflectance of winter wheat leaves and LWC is poor.SG',SG",(?)?(?),(1/SG)',(1/SG)",(lgSG)',(lgSG)" transformation can be significantly improved Correlation between spectral reflectance and LWC.The model achieved the best effect when SLR used(1/SG)" transformation,PCR used(lgSG)" transformation and PLSR used(lgSG)"transformation,with model Rp2 equal to 0.7965,0.9158 and 0.9207,respectively,and RMSEp equal to 4.5800%,2.8666%and 2.7815%respectively,and the PLSR method as a whole is better than the SLR and PCR methods.Optimize the modeling band and compare the residual prediction deviation(RPD)of the estimated model.The SPA method is better than the CA method.The SPA method is superior to the CA method.The(1/SG)"-SPA-PLSR model constructed using the 11 feature bands screened by SPA has the highest estimation accuracy,Rp2 is 0.9449,RMSEp is 2.3185%,and the RPD value is 4.3175,and the model verification R2 is 0.5668.This model simplifies the number of modeling bands,improves the estimation efficiency,and can more accurately estimate the water content of winter wheat leaves compared with full-band and significance test band modeling.(3)Data transformation of the raw spectral reflectance of the canopy can improve the correlation between the spectral reflectance and PWC.Among them,the lgSG-PLSR model constructed using the lgSG transform has the highest estimation accuracy,Rp2 is 0.8808,and RMSEp is 3.2512%.RPD value is 2.9343.Optimize the modeling band and compare the estimated RPD value of the model.The SPA method is better than the CA method.The SPA method is superior to the CA method.The lgSG-SPA-PLSR model constructed using the 9 feature bands screened by SPA has the highest estimation accuracy,Rzp2 is 0.8925,RMSEp is 3.0880%,and the RPD value is 3.0894.From different growth stages,the estimation models of jointing and heading stages have lower accuracy,and the model RPD values are less than 2.0,it cannot estimate the plant water content.The estimation model of the flowering stage,the pre-grouting stage and the mid-grouting stage is relatively high,and the model RPD value is greater than 2.5,it has a good ability to estimate the water content of the plant.Among them,the SPA method is used to screen the lgSG-SPA-PLSR constructed in 13 feature bands has the highest accuracy,Rp2 is 0.9048,RMSEp is 1.3811%,and RPD value is 3.4547.This model simplifies the number of modeling bands,improves the estimation efficiency,and has a better ability to estimate the water content of winter wheat mid-grouting stage.