Research On The Detection Methods And Instrumentation Of Soil Properties Using Spectral Analysis Technology

The development of modern agriculture has benefited from the use of fertilizers and pesticides. Fertilization improves soil fertility, and increases crop output per unit. However, in order to pursue high-yield, the crops are often fertilized excessively. The increase in application amount and decrease in utilization rate of the fertilizers not only bring huge losses to the national economy, but also cause serious environmental pollution. For example, the eutrophication in surface water was aggravated, and the content of nutrient elements in groundwater and vegetables exceeded the allowed figure. Therefore, quick access to the information of soil nutrient content, and fertilize crops according to their demands, which have a great significance for the sustainable development of China’s agriculture. For measuring soil nutrient content, traditional chemical analysis methods are complex, time-consuming, high costs and poor real-time. A rapid, in-situ, continuous and pollution-free method for the detection of soil nutrient content is an urgent need in agricultural production.Visible and near infrared (Vis-NIR) spectroscopy has been shown to be a rapid, cost-effective, convenient, and non-polluting method. In this work, Vis-NIR spectral technique combined with chemometrics was used for the measurement of soil nutrient content. The spectral characterization and fast determination method of soil total nitrogen (N), organic carbon (OC), available phosphorus (P) and available potassium (K) have been investigated. A portable instrument which has a good man-machine interface was developed for the fast analysis of soil nutrient. These results were meaningful for the precision management and operation on farmland. The main conclusions were as follows:(1) The spectral detection models were established for the fast and nondestructive determination of soil N, OC, P and K. A complete comparison was performed among different spectral preprocessing methods, and the optimal preprocessing method was standardized normal variable (SNV). Two variable selection methods, namely, Monte Carlo-uninformative variable elimination (MC-UVE) and successive projections algorithm (SPA) were applied to select effective variables. Linear multiple linear regression (MLR) and partial least squares (PLS), and nonlinear least squares-support vector machine (LS-SVM) models were developed based on full-spectrum variables and effective variables. The proposed combination of MC-UVE-LSSVM achieved the optimal prediction performance for soil N, OC, P and K based on NIR spectral data. The values of coefficient of determination (R2) and residual prediction deviation (RPD) were respectively0.88,0.89,0.59and0.75, and2.9,3.1,1.5and2.0.(2) The effective variables obtained by chemometrics methods were analyzed with the chemical absorption groups which were related to N, OC, P and K. It was found that:some of the effective variables of N and OC were directly associated with their functional groups, whereas some impacted on the prediction accuracy by measuring soil moisture content. Soil P and K do not possess direct spectral response in the Vis and NIR region. They were found to be measurable with different levels of agreement, which was partly attributed to the covariation of moisture content and illite in soil. This finding explained how the effective variables affected the prediction results, and provided an important basis for the prediction of soil N, OC, P and K based on effective variables.(3) The establishment of dynamic models was proposed for the prediction of soil N, OC, P and K. Dynamic models were able to learn information from the latest samples by updating effective variables and regression coefficients during the prediction process, so that the robustness of the established models was enhanced and the application range was expanded. The recursive partial least squares regression (RPLS) models and three variable-updating models, namely variable importance in the projection combined with RPLS (VIP-RPLS), VIP-PLS, and uninformative variable elimination combined with PLS (UVE-PLS), all achieved better prediction results than the PLS models for the prediction of soil N, OC, P and K. The proposed combination of VIP-RPLS achieved the optimal prediction performance based on NIR spectral data. Compared with the PLS models, the prediction accuracy was improved by9%-17%.(4) A new design of measurement instrument for the rapid detection of soil nutrient content was proposed. It adopted Windows CE as the embedded operating system, and applied miniature spectrometer module USB4000as the sensing element. The main hardware and data processing software were implemented. The instrument was able to real-time display the analysis results. When using this instrument to measure soil N and OC content, the values of R2and RPD were respectively0.53,0.57, and1.5,1.5. The results indicated that this instrument has certain predictive ability for the prediction of soil N and OC, and laid a solid foundation for further developing high-precision and multifunctional instrument for the fast measurement of soil properties.The above work provided an important foundation for the rapid and non-destructive detection of soil nutrient content, and has a promising application prospect.