Nondestructive Diagnosis Of Nitrogen Concentration Of Rubber Tree Based On Spectroscopy Technology

Nature rubber is an important strategy material and its production is directly related to the development of our country's economy.The concentration of nitrogen is critical for the growth of rubber tree and the production of natural rubber.Reasonable use of nitrogen can promote the production of nature rubber.Currently,the empirio-judgment and biochemical analysis method are the most commonly used strategy to detect the nitrogen concentration in agriculture.However,the existing method is time-consuming,harmful to environment and need professional operators.With the development of precision agriculture,more life information need to be obtained fast and precisely to support the field management and the traditional nitrogen estimation method cannot meet the requirement.Thus,it is important to propose a fast and accurate nitrogen estimation method to improve the production of natural rubber and support the development of modern natural rubber industry.By systemically analyzing the spectra character of rubber leave and fractional calculus,a series of nitrogen concentration qualitative and quantitative models were proposed.The main novelty of this paper can be summarized as follows:1?The investigation on nitrogen concentration level estimation model based on BP-Adaboost algorithm and NIR spectra.According to National standard of China GB/T 29570-2013(Technical regulations for foliar nutrient diagnosis of rubber tree),a BP-Adaboost algorithm based nitrogen concentration level identification model is proposed by the steps of pretreatment,feature extraction and model construction.By analyzing the spectra character of raw spectra,15 efficient wavebands are selected as the feature to construct a fast and accurate nitrogen concentration level estimation model.Three models including Byes discrimination model,BP neural network model and BP-Adaboost model are proposed to get the accurate nitrogen content level of rubber tree.The optimal result is obtained by BP-Adaboost model with the recognition rate of 92.98%.2?A fractional calculus based data processing method is proposed to solve the problem of noise amplification while the data processing process using the traditional derivation method.Based on the high degree of freedom and non-local characteristics of fractional calculus,the balance points of detail extraction and noise suppression are sought in the global scope.Thus,the fractional calculus enhanced spectra can be obtained.This method provides a basis for establishing an efficient and accurate nitrogen estimation method of rubber tree.3?A nitrogen concentration identification method is proposed by fractional order augmented spectra with order 0-2.Fractional calculus is applied to get the fractional order augmented spectra.Two modeling methods,including support vertical machine and extreme learning machine,are applied to get the nitrogen concentration level identification model.The recognition performances of different modeling methods and the effectiveness of fractional order treatment are discussed.The results show that the models established by 0.6-order spectra and 1.6-order spectra obtained the best recognition result.The complexity of the models can signification reduced by successive projection algorithm.The optimal recognition result is obtained by extreme learning machine model and the best recognition rate is 95.45%.4?The establishment of fractional calculus based nitrogen concentration quantitative analysis model for rubber tree.11 fractional order spectra are obtained by the method of fractional calculus.Partial least squire regression algorithm is used to construct the nitrogen concentration estimation model.By comparing the prediction results of models established by fractional order augmented spectra with different orders,the optimal order for nitrogen concentration estimation can be assured.The best order for nitrogen content regression model is 0.6-order.The regression coefficient and root mean square error of the best model is 0.9245 and 0.1186,respectively.