Research On Growth Monitoring And Yield Prediction Of Processing Tomato In Xinjiang Based On UAV Images

In order to effectively and nondestructive monitor the growth of processing tomatoes,and to timely and accurately predict their yield,we conducted the study in the processing tomato experimental base of COFCO Tunhe Tomato Co.,Ltd in Xinjiang.Firstly,Phantom 4 Multispectral UAV was used to obtain visible light images and multi-spectral images of processing tomato canopy during main growth periods,and to simultaneously measure the chlorophyll content and yield of processing tomato.Then,we preprocessed the UAV images,extracted 10 visible light vegetation indices and 10 multi-spectral vegetation indices,and evaluated the most sensitive variable of chlorophyll,yield,and vegetation indices according to the correlation analysis.Finally,based on linear regression,multiple stepwise regression,and partial least squares regression,we established chlorophyll inversion model and yield prediction model of processing tomato during main growth periods to obtain the best chlorophyll inversion model and the best yield estimation model.The results of this study are as follows.(1)The chlorophyll content of processing tomato decreased from the initial flowering stage to the full flowering stage,and increased from the full flowering stage to the fruit bearing stage,reached the peak in the fruit bearing stage,and decreased from the fruit bearing stage to the fruit maturing stage.In the correlation analysis between chlorophyll content and vegetation index,the soil adjusted vegetation index(SAVI)and canopy chlorophyll content index(GCCI)had the highest correlation with the chlorophyll content(r=0.931).(2)In the chlorophyll inversion model of processing tomato,the model based on partial least squares regression with visible light vegetation indices as the independent variable had the best effect,with a determination coefficient(R2)of 0.734,root mean square error(RMSE)of 3.16,standard root mean square error(nRMSE)of 8.79%,mean relative error(MRE)of 5.25%,and a model of y=-73.134xEXG+30.962 xNGRDI-33.132xRGBVI+7.337xMGRVI+35.151.(3)In the correlation analysis between yield and vegetation indices,the ratio vegetation index in fruit expanding stage had the highest correlation with yield(r=0.793).The precision of the yield estimation model in each growth period was characterized by the fruit expanding stage>fruit maturing stage>initial flowering stage>full flowering stage>fruit bearing stage.(4)In the yield estimation model of processing tomato,the prediction model based on partial least squares regression with the multi-spectral vegetation indices in the fruit expanding stage as the independent variable had the best effect,with determination coefficient(R2)of 0.642,root mean square error(RMSE)of 0.64,standard root mean square error(nRMSE)of 4.22%,mean relative error(MRE)of 0.70%,and a model of y=-135.836xNDVI+3.241xGNDVI+85.803xLGI+4.576xOSAVI-104.026xNDRE-0.355xRVI+0.111xEVI+146.945xGCCI+1.830.This study showed that the UAV remote sensing technology can be used to retrieve the chlorophyll content and predict the yield of processing tomato,providing strong support for the nutritional diagnosis of processing tomato in the field and the formulation of local agricultural policies.The growth monitoring and yield prediction of processing tomato provide a new method for the large-scale crop phenotype research of precision agriculture,offering effective technical support for the cultivation of high-yield varieties of processing tomato and fine management of field crops.