FAPAR Estimation Of Maize Based On Spectral And Texture Features Of UAV Remote Sensing Image

Fraction of Absorbed Photosynthetically Active Radiation(FAPAR)is one of the most important parameters used to characterize plant canopy,which can be used to predict crop yield.At present,the empirical method of crop FAPAR estimation based on Unmanned Aerial Vehicle(UAV)images is mostly based on spectral features,which has the problem of FAPAR saturation.Texture is an important structural parameter of plants,which can represent the structure of plant canopy.However,few people use crop textures in crop FAPAR estimation.In this paper,summer maize was used as the research object to explore the role of vegetation indices and textures in maize canopy FAPAR estimation.Based on the RGB and multispectral sensors carried by the UAV,vegetation indices and textures of maize canopy were extracted,and the FAPAR estimation models of the maize canopy were constructed.The FAPAR of maize canopy was estimated by vegetation indices,textures and the combination of vegetation indices and textures.The main study contents and results are as follows:(1)The dynamic changes of FAPAR growth in different sowing dates,different growth periods and different growth periods of each maize variety were analyzed,which served as the basis for subsequent analysis of FAPAR estimation data.(2)To explore the effect of textures on FAPAR estimation of maize canopy.Based on the textures,the FAPAR of maize was estimated by the traditional empirical method and partial least squares(PLSR)method using UAV RGB images and multispectral images,and the results were compared with that of maize FAPAR estimated by vegetation indices.(3)The PLSR method was used to construct maize canopy FAPAR estimation models based on RGB images,multispectral images and RGB+multispectral images,respectively,using vegetation indices and textures.The results were compared with those of the traditional empirical method,and indicate that,under the same conditions,the accuracy of PLSR method is higher than that of traditional empirical method in maize FAPAR estimation.For the vegetation indices,the models based on multispectral images are more accurate.For textures,the models based on RGB images are more accurate.Under the same conditions,the maize FAPAR estimation models based on RGB images+multispectral images are more accurate than those based on RGB images or multispectral images alone.(4)The vegetation indices and textures were combined to construct a new estimation model for FAPAR of maize canopy.PLSR method was used to construct FAPAR estimation models of maize canopy based on RGB images,multispectral images and RGB images+multispectral images,respectively,and the accuracy of the three models was compared.The results indicate that,the combination of vegetation indices and textures is better than that of vegetation indices or textures alone for FAPAR estimation of maize canopy.The combination of vegetation indices and textures based on RGB images+multispectral images has the highest accuracy in FAPAR estimation of maize,with determination coefficient(R~2)and relative root mean square error(r RMSE)reaching 0.95 and 3.64%,respectively.