| Soil is the foundation of agriculture and fertilizers are the important goods in agrocultural production. Fertilization can improve soil fertility and increase crop output per unit. However, over-fertilization can also cause the waste of fertilizers and environmental pollutions, thereby bringing about huge losses to the economy. Precise fertilization occupies an important position in precise agriculture, which can contribute to the increase of fertilizer utilization rate and the reduction in environmental pollutions. Currently, there still exist a series of disadvantages in detecting the nutrients of fertilizers and soils using the traditional chemical analysis methods, such as time-consuming, complex, high costs and secondary pollution. It is extremely urgent for a rapid, green, real-time and, low-cost in-situ detection method in agriculture.Visible and near infrared (VIS-NIR) spectroscopy has been proved to be a convenient, rapid, cost-effective and non-polluting method. In this work, the samples of fertilizers and lime concretion black soil (a typical soil type in Huaibei Plain), were collected, and then VIS-NIR spectral technique in comnination with chemometrics was used for rapid measuring the nutrient contents of fertilizers and soils, including total nitrogen (N),total phosphorus (P), total potassium (K) of fertilizers and the organic matter content (OMC), total nitrogen (N), pH value of soils. Finally, the corresponding regression models were built. This study is significant to the promotion of precision fertilization. The major research contents and fruits of this work are as follows:(1) The VIS/NIR spevtroscopy rapid detection models for N, P and K contents in fertilizers and OMC, N contents and pH value of soils were investigated in detail, so as to achieve the rapid non-destructive measurements for fertilizer nutrients and soil parameters. Firstly, the effects of 17 kinds of pretreatment methods were studied systematically, so the optimal pretreatment method for different parameters was obtained respectively. Secondly, the samples in modeling set were selected by Kennard-Stone (KS) and the Sample Set Partitioning based on Joint x-y Distances (SPXY) algorithms, then the characteristic wavelengths were extracted by successive projections algorithm (SPA) and genetic algorithm (GA), and finally the prediction models were built using principal component regression (PCR), partial least aquares regression (PLS) and least squares-suppot vector machines (LS-SVM) based on the whole wavelengths and the selected characteristic wavelengths. The results using different models were evaluated and compared, then, the most optimal rapid detection measurement model of all is achieved.(2) The constructed smooth-SPXY-SPA-LS-SVM model, can provide favorbable prediction results for the N content in fertilizers, with the coefficient of relation (R) of 0.9346 and residual prediction deviation (RPD) of 4.81 for the prediction set. The constructed SNV-KS-GA-PCR model can provide favorable prediction results for the P content,with R of 0.9619 and RPD of 4.52. The constructed MSC+2nd Derivative-KS-GA-LS-SVM model can provide favorable prediction results for the K. content, with R of 0.9105 and RPD of 2.57. According to the classtion standard of RPD, all the above models could be used for quantitative analysis.(3) The constructed smooth+SNV-SPXY-GA-PCR model can provide favorable results for the OMC conent in lime concretion black soils, with R of 0.9138 and RPD of 2.41. The constructed MSC-SPXY-GA-PCR model can provide favorable result for N content, with R of 0.9604 and RPD of 3.59. The constructed 1st Derivative+SNV-SPXY-GA-PCR model can provide favorable results for pH value, with R of 0.9162 and RPD of 2.51. According to the classtion standard of RPD, all the above models could also be used for quantitative analysis.(4) By SPA and GA, the characteristic wavelengths of VIS/NIR spectroscopy can be effectively extracted, which can simplify the complexity of the constructed models without the reduction of prediction and thus lay the technique foundation for research and development of the rapid detection equipments... |
PDF Full Text Request