The Research On Extraction Of Maize Phenotypic Information Based On Unmanned Aerial Vehicle

Phenotypic information is an important manifestation of crop varieties and growth status,it's also an important factor affecting crop yield.As the larger global population base,food demand is increasing,leading to serious food supply problems.Rapid and accurate extraction of large-scale crop phenotypic information to monitor the growth of crops,and timely and effective management measures for the breeding of high yield and quality of crop varieties,safeguarding China's food security has far-reaching significance.However,the use of artificial field measurement method to obtain phenotypic information has the high accuracy,but the regional coverage is low,it's not suitable for large-scale breeding field.With the rapid development of remote sensing technology,it is possible to obtain real-time,fast and non-destructive surface information of large-scale.The aim of this study is to provide a theoretical basis for obtaining the phenotypic information of crop based on the micro-UAV high-throughput remote sensing platform,and to provide auxiliary support for the study between genotype and phenotypic information of maize varieties.A high-throughput phenotypic information acquisition test for micro-UAVs was carried out in the "National Precision Agriculture Demonstration Research Base" in June-September 2015 including image feature extraction(maize height,tasseling time,vegetation cover,leaf color change)and LAI inversion.The main research work and research results are as follows:(1)Using high-definition digital photo data obtained from UAV high-throughput remote sensing platform with three methods of ISODATA method,SVM method,decision tree classification based on HSV color space transformation to extract canopy coverage,overall accuracy and Kappa coefficient were 59.06%,0.26;92.70%,0.96;98.32%,0.96 respectively.It could be seen that the decision tree classification based on HSV color space transform was the highest,and could be used to extract the coverage of multiple growth period images.(2)The maize tassel extraction was carried out by using two methods of decision tree classification based on HSV color space transformation and object-oriented classification(combining texture information,HSV color space transform,NDI vegetation index and geometric information).The overall accuracy of the classification was 83.79% and 85.91% respectively.Compared with the decision tree classification based on HSV color space transformation,the object-oriented classification method had higher accuracy.Therefore,using object-oriented classification method extracted maize tassel,and then extracted maize tasseling time.The extraction accuracy was 65.62%.It was found that the extraction of maize tasseling time was feasible by using this method.(3)Using the decision tree classification based on HSV color space transformation to extract maize leaf color during the growth period.The use of image hue values could significantly distinguish the color of the leaves,so as to achieve the need of maize leaf color extraction.(4)During maize LAI inversion of growth period,for univariate models,the results showed that NDVI was better than that of other vegetation indices.The R2 and RMSE of linear model and power model were 0.525,0.711 and 0.530,0.717 respectively.NDVI could be used to monitor the LAI changes of growth period;for multivariate inversion,firstly the principal component analysis was carried out on the eight vegetation indices,and then analyzed by multiple linear regression and BP neural network.The results showed that BP neural network had good ability for maize LAI inversion,R2 was 0.605 and RMSE was 0.745,which could predict the change of LAI in maize growth period.(5)In the process of extracting the plant height,the height of the 6884 breeding materials extracted from DSM image had a good linear relationship with the value of the measured,R2 was0.527 and RMSE was 0.223.Therefore,this method could replace the traditional artificial plant height measurement method;through the plant height distribution map,the plant height distribution,changes and other information could be seen more intuitively.