Estimation Of Canopy Nutrition, Flower Quantity And Yield Of Citrus Plant By Low-altitude Remote Sensing

The canopy multispectral imagery of Tarocco blood orange plants in flower bud physiological differentiation period, dormancy period and flowering period were collected using eight-rotor unmanned aerial vehicle(UAV) equipped with multispectral camera array Canopy leaves for nutrient content determination were collected, and flower quantity and yield of per plant were investigated in the test area. Then spectral remote sensing image processing and statistical analysis technology were employed to study on the estimation of canopy nutrition, flower quantity and yield of citrus plant based on low-altitude remote sensing technology. This work would provide a theoretical basis and technical support for citrus nutrition, flowering and fruit setting amount detection technology research and development based on low-altitude UAV remote sensing information. The main contents and results of research were as follows: 1. Estimation of canopy nutrition contents of Tarocco blood orange plant by low-altitude remote sensing(1) By the analysis of low-altitude remote sensing spectral information of Tarocco blood orange plant canopy in flower bud physiological differentiation period, dormancy period and flowering period, it was found that low-altitude remote sensing spectral reflectance of Tarocco blood orange canopy was very similar to that of typical plants, at low reflectance in visible spectral region, high reflectance in the near infrared region. The trend of canopy spectral reflectance curve in three phenophases was similar, but the values of spectral reflectance were significantly different, the variation trend was flower bud physiological differentiation period > dormancy period > flowering period.(2) Based on the acquisition of low-altitude remote sensing spectral information of Tarocco blood orange plant canopy in three phenophases, four kinds of methods(Normalization, Multiplicative scatter correction(MSC), De-Noise, and Standard normal variable transformation(SNV)) were adopted to preprocess the original spectrum, and the estimation models for plant canopy nutrient content prediction based on the pretreated spectra and the original spectrum were established by partial least squares(PLS), multiple linear regression(MLR), principal component regression(PCR) and least squares support vector machine(LS-SVM). By means of comparing the accuracy of different models for plant canopy N, P, K, Ca, Mg, S contents prediction, the results showed that the optimal model for plant canopy N content estimation by low-altitude remote sensing was the PCR model based on original spectrum in dormancy period. Prediction model established by MLR for canopy P content based on original spectrum in dormancy period was the best. LS-SVM model for canopy K content based on original spectrum in flowering period had the best prediction accuracy. By SNV pretreatment canopy original spectrum of dormancy period, using LS-SVM to establish model for canopy Ca content could reach the optimal estimation accuracy. In order to estimate the content of Mg in canopy, the PLS model based on SNV pretreatment spectrum in flower bud physiological differentiation period was the optimal. Take advantage of the original spectral in dormancy period to establish LS-SVM model can be used to estimate canopy S content better. 2. Estimation of flower quantity of Tarocco blood orange plant by low-altitude remote sensing(1) Based on the acquisition of low-altitude remote sensing spectral information of tarocco blood orange plant canopy in three phenophases, four spectral pretreatment methods(Normalization, MSC, De-Noise and SNV) were adopted and four kinds modeling methods(PLS, MLR, PCR, and LS-SVM) were employed to estimate flower quantity of plant canopy. The results showed that the best period for flower quantity of Tarocco blood orange plant estimation by low-altitude remote sensing of canopy is dormancy period, the best model was LS-SVM prediction model based on De-Noise pretreatment spectrum, RC and RP of the model were 0.6571 and 0.6519, NRMSEP and NRMSEC were 0.3150 and 0.2852 respectively.(2) For further optimization of the model for plant flower quantity estimation, three vegetation indexes(NDVI, DVI, RVI) were adopted to transform spectral data. Then, vegetation indexes(NDVI+DVI+RVI), original spectrum combined with vegetation indexes were treated as the independent variables respectively. PLS, PCR, LS-SVM models for canopy flower quantity estimation were established, by comparing prediction accuracy of all the models. It was found that the dormancy period was the best period for flower quantity of Tarocco blood orange estimation. The best estimation model was PCR model based on the original spectrum combined with vegetation indexes as independent variables(RC = 0.6766, NRMSEC = 0.2912, RP = 0.6963, and NRMSEP = 0.3928).(3) Through the comprehensive analysis of all the estimation models for plant flower quantity, the results showed that the estimation accuracy of the PCR model based on original spectrum combined with vegetation indexes was the best. 3. Yield estimation of Tarocco blood orange plant by low-altitude remote sensing(1) Based on the acquisition of low-altitude remote sensing spectral information of Tarocco blood orange plant canopy in three phenophases, four spectral pretreatment methods(Normalization, MSC, De-Noise, SNV) were adopted and four kinds modeling methods(PLS, MLR, PCR, LS-SVM) were employed to estimate yield of the plant. The results showed that the best period for yield of Tarocco blood orange plant estimation by low-altitude remote sensing of canopy was dormancy period, and the best model was LS-SVM prediction model based on MSC pretreatment spectrum(RC and RP of the model were 0.7290 and 0.7237, NRMSEP and NRMSEC were 0.2347 and 0.2549 respectively).(2) Three vegetation indexes(NDVI, DVI, RVI) were adopted to transform spectral data, Then, vegetation indexes(NDVI+DVI+RVI), original spectrum combined with vegetation indexes were treated as independent variables respectively. PLS, PCR, LS-SVM models for plant yield estimation were established. By comparing prediction accuracy of all the models, it was found that the dormancy period was the best period for yield estimation of Tarocco blood orange. The best estimation model was LS-SVM model based on the vegetation indexes as independent variables(RC = 0.8727, NRMSEC = 0.1729, RP = 0.8081, and NRMSEP = 0.2585).(3) Through the comprehensive analysis of all the estimation models for plant yield, the results showed that the estimation accuracy of the LS-SVM model based on the vegetation indexes was the best.