Study On Jujube Quality Detection By UAV Near-ground Multi-angle And Polarization Detection

In order to solve the problem that the zenith angle of light source,zenith angle detection,the difference of relative azimuth angle and the low accuracy of multi-spectral detection of unmanned aerial vehicle(UAV)generally affect the quality detection effect of jujube,this paper takes jujube in southern Xinjiang as the research object,and carries out the quality detection research of jujube in indoor and outdoor unmanned aerial vehicles under multi-angle and polarization modes.Based on the obtained unpolarized and polarized spectral data and physical and chemical indexes,a prediction model of jujube quality was established.By exploring the characteristic angle of jujube quality detection in southern Xinjiang and the method of jujube reflectivity inversion,a multi-angle fusion jujube quality detection model was established to improve the detection accuracy.The specific research results are as follows:(1)The jujube quality detection model under different azimuth angles and the jujube quality prediction model under multi-angle fusion are constructed.Under outdoor natural conditions,multi-angle jujube data were collected by using multi-spectral unmanned aerial vehicle,and the corresponding jujube reflectivity data were obtained when the relative azimuth angles were 0,45,90,135,180,225,270 and 315.The jujube moisture content model and soluble solid model were constructed based on single angle and multi-angle fusion respectively,and the characteristic angles of jujube detection were explored.Compared with the single-angle model,the multi-angle fusion method can significantly improve the detection accuracy of water content and soluble solids of jujube.Among the water content prediction models,the 90 polarization PLSR model based on angle fusion has the best prediction effect.Compared with the single angle model,the correlation coefficient of prediction set is increased from 0.6331 to 0.8404.Among the soluble solids prediction models,the 45 polarization SVM model based on angle fusion has the best prediction effect.Compared with the single angle model,the correlation coefficient of prediction set is increased from 0.5645 to 0.8867.The above results show that the jujube quality prediction model based on multi-angle fusion method can effectively improve the accuracy of multi-spectral detection of jujube quality by UAV,and can provide a new idea for the subsequent large-scale detection of agricultural products by UAV.(2)In order to reduce the workload of multi-angle data collected by UAV,four bidirectional reflectance distribution function(BRDF)models,namely Index,Rahman,Ross-li and Roujean,were used to invert the multi-angle and multi-polarization data collected indoors and outdoors,and to predict the jujube reflectivity of other azimuth angles.The natural lighting conditions in Alar City,the first division of Xinjiang Uygur Autonomous Region,from September to December were simulated indoors.The incident zenith angle was set at 30 ～ 60,the detection zenith angle was set at 0 ～ 60,and the relative azimuth was set at 0 ～ 360.The unpolarized reflectance,0 polarized reflectance,45 polarized reflectance,90 polarized reflectance and 135 polarized reflectance of the whole jujube region were extracted.After the envelope is removed,four BRDF models are used for inversion.The results show that Index,Rahman,Ross-li and Roujean models are all suitable for inversion of indoor jujube reflectance data.The multi-angle and multi-polarization data collected outdoors by four BRDF models are used for inversion.The results show that the four BRDF models are all suitable for the original reflectance data of jujube without polarization.However,for polarization data,the inversion results of the four BRDF models are quite different,and the inversion effects of each model need to be analyzed in detail under different azimuth angles.None of the four BRDF models can well invert the polarization reflectance of all angles.(3)Contrastively analyze the red jujube reflectance retrieved by four BRDF models from different angles and the real multi-angle reflectance data,fuse the reflectance data with the best retrieval effect from each angle,and build the red jujube water content and soluble solid prediction models based on multi-angle fusion.Compared with the established non-polarized and polarized angle fusion models,it is found that the prediction effect of the non-polarized angle fusion model is better.Among the water content prediction models,the unpolarized PLSR model has the best prediction effect,with the correlation coefficient of the prediction set of 0.7585 and the root mean square error RMCEP of the prediction set of 2.0598.Among the prediction models of soluble solids,the unpolarized PLSR model has the best prediction effect,with the correlation coefficient of the prediction set of 0.7493 and the root mean square error RMCEP of the prediction set of 2.3108.The above results show that it is feasible to establish a jujube quality detection model using unpolarized jujube reflectivity data retrieved by BRDF model.