| The rapeseed cultivation index is an important dynamic monitoring parameter reflecting the growth status of rapeseed.In recent years,unmanned aerial vehicles(UAVs)have received extensive attention in agricultural information collection because of their flexible data acquisition methods,low atmospheric impact,low cost of use,and high efficiency.The different fertilization level and planting density test were designed in the cultivation center of Hunan Agricultural University.The tested varieties were 708 and 710 for Brassica Napus.The canopy images of rapeseed seedling stage and pod fruit stage were collected by the DJI Inspire 2 UAV equipped with high-definition camera.The chlorophyll content of leaves and pods was collected by SPAD measuring instrument and CM measuring instrument respectively.The leaf area index of the seedling canopy was collected by the leaf area index measuring instrument.The color index of the canopy image was extracted by the crop digital image analysis system CMS developed by MATLAB.The correlation analysis was carried out to select the color index with significant correlation to establish the inversion model.The followings are the results of research.(1)The color index of canopy image of rapeseed by UAV had a significant correlation with SPAD value,CM value and leaf area index.It can be used to construct the inversion model of rapeseed cultivation index.(2)The color characteristics of the most significant rapeseed UAV canopy images correlation with the leaf area index of rapeseed seedlings were a*,G-R,Hv*Diff,Hi*Diff,R-Y,G-Y.The Linear,Robust,Binomial,Random Forest and Support Vector Regression inversion model were established.The cross-validation results showed that the linear,model was the most stable,and the relative error rate was 9.17% to 11.37%.(3)The color index of the most significant rapeseed UAV canopy images correlation with the CM value(chlorophyll content)were G,Y,V,L*,G-R,R-B.The Linear,Robust,Binomial,Random Forest and Support Vector Regression model were established.The cross-validation results showed that the linear model was the most stable,and the relative error rate was 3.79% to 4.11%. |
