Measurement Model Of Nitrogen, Phosphorus And Potassium Content For Citrus Leaves Based On Hyperspectrum

Nitrogen, phosphorus and potassium are three essential nutrients in citrus growth. However, traditional chemical measurementmethods of obtaining these three nutrients content of citrus leaves aretime-consuming procedures with complex operations and heavy workload, which can be harmful to citrus leaves. Therefore, rapid, non-destructive and accurate measurement of nitrogen, phosphorus and potassium content for citrus leaves is significant for citrus cultivation and management, variable spraying fertilizer, agricultural machine development, and may provide a theoretical basis for nutritional diagnosis andgrowth monitoring of fruit trees based on hyperspectral technology.In this paper, the modelingwas based on hyperspectral technology and Luogang citrus as testvariety. Healthy leaves were collected during four important citrus growth periods corresponding to germination period, stability period, bloom period and harvesting period. Hyperspectral reflectance and nitrogen, phosphorus and potassiumcontent of citrus leaves were measured by spectrometer(ASD Field Spec 3) and traditional chemical measurementmethods, respectively. In this way, models based on hyperspectral data were applied to the final regression analysis for predicting nitrogen, phosphorus and potassium contentof citrus leaves. The main contentsare as follows:(1) In the research of the measurement model of nitrogencontent for citrus leaves based on hyperspectrum, feature wavelengths were extracted from spectrum by correlation coefficient, successive projections algorithm(SPA) and principle component analysis(PCA), and the number of feature wavelengths is 137,12 and 25 respectively. Prediction models of nitrogen content were establishedbased on back-propagation neural network(BPNN) and support vector regress(SVR). SVR achieved the best performance, and the coefficient of determination(R2) of the validation set was 0.9520, the root-mean-square error(RMSE) was 0.3162.(2) In the research of the measurement model of phosphoruscontent for citrus leaves based on hyperspectrum, orthogonal experiment analysis was used to determine the optimal denoising parameters of wavelet denoising. When “haar” was used as the wavelet basis function, the decomposition layer was 3, “heursure” as the threshold selection and “sln” as the threshold rescaling project,wavelet denoising achieved the bestdenoising performance. WD-SMLR(wavelet denoising- stepwise multiple linear regress) and WD-PLSR(wavelet denoising- partial least square regression) for predicting phosphorus content of citrus leaves were established, and the coefficient of determination(R2) of the validation set were 0.8247 and 0.8313, the root-mean-square error(RMSE) were 0.0675 and 0.0623, respectively.(3) In the research of the measurement model of potassium content for citrus leaves based on hyperspectrum, the dimension of intrinsic manifold was determined based on multidimensional scaling(MDS). Then, multidimensional scaling(MDS), isometric mapping(Isomap), laplacian eigenmaps(LE), locally linear embedding(LLE), maximum variance unfolding(MVU) five manifold learningalgorithms were applied to establish prediction model of potassium content of citrus leaves, while genetic algorithm(GA), particle swarm optimization(PSO), cross validation(CV)were used to optimize the parameters of support vector machine regression model. Experiment results revealed that MDS-GA-SVR achieved the best performance, and the coefficient of determination(R2) of calibration setand validation set were0.9950 and 0.9798, the root-mean-square error(RMSE) were0.3238 and 0.6296, respectively.