Wheat Yield Estimation Based On UAV Canopy Spectrum And 3D Volume Data

Wheat is the main grain crop in our country.Making an estimate of wheat yield in a timely and accurate way is beneficial to the decision-making and management of Chinese agriculture.Traditional crop yield estimation methods mainly rely on human fieldwork,which is time-consuming and laborious.Remote sensing technology provides real-time crop information at multiple spatial scales and provides a means for rapid and non-destructive prediction of wheat yield.In this study,multispectral images of the wheat seedling stage,greening stage,jointing stage,booting stage,and flowering stage were obtained under three densities and four nitrogen fertilizer treatments.12 common color features were summarized,15 planting cover index and 5 texture features(collectively referred to as image features)were correlated with yield by correlation analysis,and features with high yield correlation were screened out,and features with high collinearity were eliminated.Finally,IKAW and MGRVI were selected at the seedling stage,INT and PSRI at the greening stage,ExG,PSRI,and Contrast at the joining stage,and INT,ExR,RGBVI,NDREI and Contrast at the booting stage.ExG,NGBDI,RGBVI,and CIRE were selected for yield estimation at the flowering stage.In order to further improve the yield estimation accuracy,canopy volume characteristics were constructed on the basis of the above characteristics.The height data obtained by UAV was analyzed and used to measure the plant height.The R2 values of the predicted plant height at the joint stage,booting stage,and flowering stage were 0.7710,0.8381,and 0.8376,respectively.The canopy height was measured by the elevation difference between adjacent periods.The R2 values for predicting canopy height at the booting stage and flowering stage were 0.7183 and 0.7515,respectively,which proved the rationality of using elevation data to obtain canopy volume.Correlation analysis was conducted between canopy volume and yield in each period to study the relationship between canopy volume characteristics and yield.Meanwhile,correlation analysis was conducted between canopy volume characteristics and color index,vegetation index,and texture characteristics.It was found that canopy volume characteristics and these indexes were not strongly collinearity,so canopy volume characteristics could be used to optimize the yield prediction model.Canopy volume characteristics obtained in each period were used to optimize the yield prediction model based on image features.The results showed that the optimized canopy volume characteristics improved the yield prediction accuracy,and the model had the best prediction effect at the flowering stage.The RF regression model optimized by particle swarm optimization had the highest prediction accuracy,with R2 of 0.8868 and RMSE of 271.9 kg/ha.The nRMSE was 10.29%,and the wheat yield was predicted successfully.Therefore,canopy volume characteristics can provide a new way to accurately estimate wheat yield by optimizing the yield prediction model.