| Chlorophyll is an important pigment used in crop light energy,which directly affects the energy conversion and transmission process of crops.The change of its content directly indicates the ability of crop photosynthesis and reflects the nutritional status of crops.Using remote sensing method to obtain chlorophyll content information of potato is a non-destructive rapid high-flux chlorophyll content monitoring method,which can provide timely and effective reference for farmland fertilization management.In this paper,the potatoes of Xiaotangshan National Precision Agricultural Research Demonstration Base in Beijing were taken as the research object,and the UAV multi-spectral image,UHD185 imaging spectrometer and potato chlorophyll content in the test area were acquired from May to September 2018.The main research contents and conclusions are as follows:(1)Based on the multi-spectral image of UAV,the variables such as vegetation index and texture feature of multi-spectral image were extracted firstly,then the correlation with chlorophyll content was analyzed,and the superior feature variables were selected.Based on the full subset of Adj.R~2 and K-fold cross validation.Analytical methods to estimate potato chlorophyll content.Finally,the vegetation index and texture features were constructed by principal component fusion to construct a new comprehensive index to estimate the chlorophyll content.The results show that in the bud stage,the R~2 ratio of the comprehensive index model is 0.104 and 0.136,respectively,while the nRMSE is reduced by 1.3%and 1.6%.The tuber formation period,the R~2 ratio of the comprehensive index model,the vegetation index model and the texture feature model.Increased by 0.04,0.101,nRMSE decreased by 0.5%and1.2%;tuber growth period,comprehensive index model R~2 increased by 0.075 and0.111 compared with vegetation index model and texture feature model,nRMSE decreased by 0.9%and 1.3%;starch accumulation period,comprehensive index The model R~2 is increased by 0.017 and 0.046,respectively,compared with the vegetation index model and the texture feature model,and the nRMSE is reduced by 0.2%and0.6%.(2)Based on the hyperspectral data of UAV,the sensitive characteristic index is extracted by fractional derivative,spectral position and area feature,and the salient characteristic variables are screened by continuous projection transform(SPA)and continuous wavelet decomposition.Finally,stepwise regression analysis(Step),support vector machine(SVM)and random forest(RF)were used to estimate and validate chlorophyll content.The results showed that:1)Potato chlorophyll content model based on fractional differential transformation,budding stage and tuber formation stage,SVM-based estimation model of potato chlorophyll content is better than MLR and RF model,tuber growth stage and starch accumulation stage,and MLR-based estimation model of potato chlorophyll content is better than SVM and RF model.2)Based on the characteristic index model of potato chlorophyll content,stepwise regression analysis was used to estimate the optimum chlorophyll content of potato at budding stage,tuber formation stage and starch accumulation stage,which was 3%and17%,4%and 12%,6%and 16%higher than that of SVM and RF respectively.RMSE decreased by 0.34 and 1.19 g.cm-2,0.31 and 0.57 G.cm-2,0.34 and 1.15 g.cm-2,respectively.At tuber growth stage,SVM estimated chlorophyll content.The content of R~2,RMSE and nRMSE were 0.71,3.59 g.cm-2 and 9.65%respectively.The SPA-PLS model was the best in budding and tuber growth stages,followed by Wavelet-Step model,tuber formation stage and starch accumulation stage,Wavelet-Step model was the best and SPA-PLS model was the worst.The results showed that SPA+PLS model was the best model for estimating the chlorophyll content of potatoes in budding stage and tuber growth stage,and characteristic+stepwise regression model was the best model for estimating the chlorophyll content of potatoes in tuber formation stage and starch accumulation stage. |
PDF Full Text Request