Rapid Detection And Instrumentation Of Soil Organic Matter Based On Visible And Near Infrared Spectroscopy

Soil organic matter plays an important role in the composition of soil, it is significant in soil fertility and plant growth. In traditional agricultural production, it is usually intensive cultivation,applying a large mount of organic fertilizer and low labor productivity. These situations have a great impact on the environment. Rapid quantitative analyze of soil organic matter play a fundamental role in promotion scientific management of agricultural production and quantitative fertilization. Traditional determination of soil organic matter content is mainly depend on the laboratory chemical reaction methods. Although they have good accuracy, however the shortcomings include higher requirement for operators, take long time and high costs. Because soil organic matter content is one of the important indicators for farmland nutrient grade, it is necessary to find a quick, easy and high precision instrument for farmland soil organic matter content detection and classification. In this study, a total number of 104 soil samples were collected in ten different regions of Zhejiang province in China. The soil samples were scanned by Ocean Optics USB4000 fiber spectrometer with the spectrum range of 350-1000 nm. The main content of this research was as follows:(1) Established calibration models based on full range of soil visiable and near infrared spectroscopy and applied in quantitative detection of soil organic matter content. The outlier of soil samples were analyzed and picked out. Aiming at the problem of noise in soil spectra,smoothing, multiple scattering correction, baseline correction and wavelet threshold de-noising pretreatment methods were implemented for spectral preprocessing. By comparing, the results show that wavelet threshold de-noising method with sym6 wavelet function and level-7decomposition acquired the best effect. The PLS prediction results with correlation coefficient(R2P) of 0.74 for raw spectra and 0.76 for wavelet de-noised, the ratio of reduction deviation(RPD) of 2.00 for raw spectra and 2.07 for wavelet de-noised, respectively. It is indicate that appropriate pretreatment methods can improve the predict accuracy of soil organic matter.(2) The calculate model simplified methods for instrument application was studied. Due to the full range soil spectra contain large amount of data which include the redundant information,this will lead to the models become highly complexity and excessive calculation. Therefore,interval partial least squares, uninformative variable elimination, successive projections algorithm, competitive adaptive reweighted sampling, genetic algorithm, ant colony optimization and particle swarm optimization were used to extract wavelength to establish PLS calibration models. The 26 wavelength selected by particle swarm optimization achieve the best result with R2 Pof 0.81 and RPD of 2.31. The results show that select the characteristic wavelengths could effectively simplifie calibration model and reduce computation time.(3) Design and development the prototype of portable spectrometer based on the wavelength selection method. Adopts the modular hardware this study designed and developed a portable spectrometer prototype. According to the spectrometer hardware, also developed a software based on ARM-Linux embedded system for rapid detection with Java progromming language which has nice user interface and mostly complete functions. With this instrument 20 samples were detected, the result R2 Pof 0.78, RPD of 1.74 and the soil organic matter grades classification accuracy rate reached 85% according to the SOM partition criterion of the standard farmland fertility survey and grading technical specifications of Zhejiang Province. The prototype could be applied to grading soil organic matter after further improved.