Monitoring Winter Wheat Growth And Yield Estimation With Multi-scale UAV Images

Wheat is the main food crop in China,so it is very important to monitor its growth and forecast its yield in real time and without damage.With the rapid development of UAV,it can carry more loads and obtain more bands and higher resolution images,making it more suitable for small area,fine growth monitoring and yield prediction.In this paper,the field image data of winter wheat at jointing stage and filling stage in Fu Village,Fuzhen town,Cangzhou City,Hebei Province were obtained by a fixed-wing UAV equipped with a multispectral camera,and the leaf area index(LAI)and chlorophyll content(SPAD)monitoring indexes at the same period and the yield data at maturity stage were also obtained.On this basis,LAI,SPAD and yield inversion models of winter wheat at jointing and filling stages were established based on vegetation indices of images at different resolutions,taking spatial differences of image data and growth differences of winter wheat into consideration.The main research contents of this paper are as follows:(1)Firstly,the image data obtained by the multi-spectral UAV at jointing stage and filling stage were spliced to generate orthographic images.The generated orthographic images were resampled at 0.1m intervals to obtain image data with a total of 10 resolutions from 0.1m to 1.0m.Then,20 planting indices under each resolution were calculated respectively.According to the correlation analysis between calculated vegetation index and measured LAI,SPAD and yield,the top five planting covers with the best correlation between vegetation index and LAI,SPAD and yield at jointing stage and filling stage were selected for follow-up analysis at each resolution.(2)By comparing LAI models constructed at jointing stage and filling stage,it can be found that,the unary linear regression model,unary quadratic regression model and multiple linear regression model constructed at jointing stage are superior to the three models constructed at grouting stage,Therefore,by comparing the inversion models at jointing stage and filling stage,LAI inversion effect at jointing stage is better,the best resolution is 0.1m,and the best model is a multiple linear regression model.(3)By comparing the SPAD model constructed at jointing stage and filling stage,it can be found that,the unary linear regression model,unary quadratic regression model and multiple linear regression model constructed at grouting stage are better to the three models constructed at jointing stage.Therefore,the SPAD inversion at jointing stage and filling stage is better in filling stage,and the optimal resolution is 0.4m.The optimal model is multiple linear regression model.(4)Compared to the output of jointing stage and milking stage build model can b-e found that,the unary linear regression model,unary quadratic regression model an d multiple linear regression model constructed at grouting stage are superior to the three models constructed at jointing stage.Therefore,the yield inversion which was carri-ed out in filling stage is better,and the optimal resolution is 0.5m.The best model is mul-tiple linear regression model.