Monitoring Potato Growth And Yield Estimation Using Unmanned Aerial Vehicle Remote Sensing Images

In recent years,UAV remote sensing monitoring technology has been widely used in agriculture with its efficient,non-destructive and real-time features,providing a new method for real-time monitoring of crop growth and accurate yield prediction,but few studies have been reported on the use of remote sensing technology to explore the relationship between spectral features and potato growth indicators and yield.As the fourth largest food crop in China,potato plays an important role in ensuring food security.Research on rapid and non-destructive monitoring of field growth and yield prediction can provide support for efficient collection of potato field phenotype information and planting management data.This study was conducted in Harbin City,Heilongjiang Province,with a randomized block design and three replications in a field trial with four potato varieties and three different treatments of nitrogen fertilizer application.UAV RGB and multispectral images were collected at key potato growth stages（bud emergence,early flowering,full flowering and final flowering）,and important agronomic indicators were measured simultaneously,and based on the remote sensing images acquired by UAV,multiple spectral vegetation indices were calculated and inverse modeling was performed with growth parameters（plant height,SPAD,above-ground biomass,whole plant biomass）and yield to obtain the estimation models for different growth stages and The inverse modeling was conducted to obtain the estimation models of growth indices for different growth stages and full growth stages,construct integrated growth indices and conduct multi-angle growth monitoring studies,and analyze the yield prediction models for different growth stages,which can provide a theoretical basis for predicting local potato growth and yield.The main results are as follows:（1）Extraction of each growth index of potato can be achieved by using the UAV platform.Based on the spectral information contained in the RGB and multispectral images of the UAV,the correlation analysis was conducted with each growth index of potato（plant height,SPAD,above-ground biomass and whole-plant biomass）,and a model for estimating potato plant height,SPAD,above-ground biomass and whole-plant biomass was constructed using multiple linear regression and random forest.The results showed that the combination of RGB index and vegetation index could improve the estimation of each growth index,and the inverse effect of the multiple linear regression model was better than that of the random forest model.The best estimation models of plant height R²and RMSE were 0.94 and 5.86 cm,respectively;the best estimation models of SPAD R²and RMSE were 0.67 and 2.06,respectively;the best estimation models of aboveground biomass R²and RMSE were 0.53 and 137.2 g m^-2,respectively;the best estimation models of whole plant biomass R²and RMSE were 0.85 and 185.6 g m^-2,respectively.（2）The plant height,SPAD,above-ground biomass and whole-plant biomass were combined into a new integrated growth index using the equal weight method,and the multiple linear regression and random forest were used to establish the integrated growth index estimation models for the bud stage,first flowering stage,full flowering stage,final flowering stage and full growth stage,respectively.The results showed that the integrated growth indexes constructed based on the equal weight method could better characterize the potato growth.Similar to the single growth index,the multiple linear regression model performed better than the random forest model,where the test set R²and RMSE of the multiple linear regression model based on visible+multispectral vegetation index at first flowering were 0.87 and 0.07,respectively,and the validation set R²and RMSE were 0.56 and 0.11,respectively.This multiple linear regression model was the best model with slightly lower accuracy compared to the single growth parameter model.（3）Based on the UAV RGB and multispectral image features,two methods,multiple linear regression and random forest,were selected to construct yield estimation models for potato using spectral features at each growth stage.The results showed that the multiple linear regression model had the best estimation effect,and the final yield estimation effect was from high to low:full flowering stage>full growth stage>final flowering stage>first flowering stage>current bud stage,and the R²and RMSE of the optimal yield estimation model were 0.77 and 0.64 kg m^-2,respectively;the R²and RMSE of the validation set were 0.68 and 0.56 kg m^-2,respectively.