| Precision fertilization is one of the core research content in the field of precision agriculture, and the critical problem is how to acquire the content of nutrient elements in the soil quickly and accurately for achieving precision fertilization. At present, the major method of soil nutrient detection is sampling in the farmland firstly, and then preprocessing, testing the samples in laboratory, but this method can’t meet the requirements of rapid testing, and so, a kind of technique which can analyze the soil nutrient real-timely is needed. Because of the advantages of the laser-induced breakdown spectroscopy (LIBS) technique, such as no sample pretreatment, shorter testing time, simultaneous measurement of multiple elements and so on, it is widely applied in many fields, and it has been researched in the fields of soil nutrient detection and soil heavy metal detection in our country and abroad. Using this technique to detect the content of soil nutrient elements can provide fundamental data for precision fertilization.Based on the LIBS system in the laboratory and use the soil samples collected in the same field from Long Kang farm as the research objects, this work mainly studies the quantitative analysis methods for the partial soil nutrient elements by using LIBS. Magnesium (Mg) is a kind of nutrient element for crops, with the quantitative analysis of Mg in soil as an example, the calibration curve of Mg was constructed by using double-spectrum line of Mg, and the quantitative analytical model of Mg was constructed by using support vector machine regression (SVR) in this work.Because the quantitative analysis by using LIBS is susceptible to matrix effect, self-absorption effect, the variations of system’s parameters, environmental noise and other interference, the accuracy of quantitative analysis for elements concentration by using single characteristic spectral lines as analytical line will be affected. In order to improve the accuracy of measurement, this paper proposes to select two characteristic spectral lines of Mg as analytical lines, one is atomic line, the other one is ionic line, and then use the sum of the spectral net intensity of the two lines to construct the calibration curve for the concentration of Mg in the soil, the predictive effect of this calibration curve in testing set which include 9 soil samples is better than use only one of the two lines, the squared correlation coefficient between the real value and the predicted value is 0.95, and the mean relative error is 1.84%.Because the constituent of soil is very complex, the interference of different elements is serious. Considering the influence produced by the interference of different elements on LIBS quantitative analysis, this paper proposes to use SVR to construct the quantitative analytical model for the concentration of Mg in the soil. First of all, the spectral feature vector is constructed according to the existence form of Mg in the soil, the spectral feature vector include the characteristic spectral intensity of Mg and many other elements, such as Fe, Mn, Al and Ca, which can cause interference to the analysis of Mg. The 30 soil samples are divided into training set and testing set firstly, and then the SVR quantitative analytical model for the concentration of Mg in the soil is constructed based on the raining set which include 22 soil samples, the predictive squared correlation coefficient of this model in 8 testing samples is 0.946, and the mean relative error is 2.06%. The predictive effect of this model shows that the method of SVR can be used to construct the LIBS quantitative analysis model for soil nutrient, but the relevant parameters need further adjustments.Moreover, corresponding software system was developed on the basis of LIBS system in the laboratory. Experimenter can operate equipment conveniently and visually by using this software system, furthermore, the soil spectral data can be obtained more automatically through this software. |
PDF Full Text Request