Study On The Calibration Transfer Of Near Infrared Spectroscopy Detection Model For Soil Organic Matter Content

Near infrared spectroscopy is a fast and nondestructive testing technology. Its unique advantages have been shown in the field of agriculture information detection. Soil organic matter is an important index of soil fertility. It is very necessary to detect soil organic matter with fast and timely detection methods.Calibration transfer of near infrared spectroscopy detection model is a key problem in the field of near infrared spectroscopy research. In this study, calibration transfer of near infrared spectroscopy detection model for soil organic matter prediction was analyzed. The soil samples were collected and the spectra were detected in different times (50soil samples were detected in the2012by using ASD Fieldpro3. namely master detection), and the other9soil samples were detected in2013by using ASD Fieldpro3, namely slave detection). Soil organic matter prediction model was built by using partial least square (PLS) analysis method in master detection samples, and the probability of using this PLS model to predict slave detection sample were analyzed. The specific research work and conclusions were as follows:1. The emergence, development of near infrared spectroscopy, near infrared spectroscopy detection technology, and detection method were reviewed.2. The principles of calibration transfer of near infrared spectroscopy detection model, and the domestic and foreign research were summarized.3. Master detection was carried out and partial least squares regression analysis method was used to establish the quantitative soil organic matter prediction model (the calibration samples were41). The correlation coefficient of the model is0.962, the Root Mean Standard Error of Calibration (RMSEC) is0.383, and Standard Error of Calibration (SEC) is0.358. Self testing on the master detection sample was carried out by using another9samples. The correlation coefficient of prediction samples was0.961, the Root Mean Standard Error of prediction (RMSEP) is0.600, and Standard Error of prediction (SEP) is0.597.4. The content of soil organic matter on the slave samples were quantitatively predicted by using PLS model built by using master detection samples. The prediction correlation coefficient is negative. It shows that the prediction model of salve sample is not feasible.5. The new PLS model was built by using new calibration samples, that is,5salve detection samples were went into calibration, the result shows that, when the number of mater detection sample and salve detection sample is similar.the good correlation coefficient will be obtained. The new PLS model was also be used to predict the soil organic matter content of salve samples.the prediction result is not very satisfactory.6. The db4function of the5layer wavelet transform was used in this research, the result shows that, both the master and salve diction samples have the similar high frequency information, but the low frequency information of the two difference is very big. The low frequency coefficient a5was used as salve detection samples spectra, and the soil organic matter content was predicted, the result shows that, wavelet transform is not feasible for model transfer.7. Orthogonal signal correction methods were used to correct salve detection samples. The result shows that, when the regression coefficients of PLS by10, the best calibration transfer model was obtained. The prediction correlation coefficient is1, SEP was0.3181, and RMSEP was0.3052.