Development And Application Of Food Detection Software Based On Near-infrared Spectroscopy Technology

In recent years, food security has become a major issue causing serious publicconcerns. The accidents of adulteration occur one after another causing serious harm topeople’s life and health. Strengthening food security’s supervision is becoming moreand more important. Near-Infrared (NIR) spectroscopy is one of the most commonlyused nondestructive testing technologies in recent years. Comparing to the traditionalchemical analysis method, it is fast, reliable, and it is also operating simply.Near-Infrared (NIR) spectroscopy develops so fast in recent years that it is widely usedin on-site and on-line testing of samples. The main contributions of the thesis aresummarized below:First of all, this thesis presents a study on developing food detection software(NIR1.0) based on NIR spectroscopy. The basic principles about NIR are reviewed andthe Near-Infrared (NIR) spectroscopy detection technology has some advantagescompared to traditional chemical detection. The latest research and developments infood detection using NIR are also expounded in this thesis.Secondly, the thesis reviews some chemometric algorithms. And chemometric isindispensable part of NIR.Thirdly, food detection software based on MATLAB has been developed. Thesoftware system can achieve spectroscopy preprocessing, unusual sample analyzing andmodeling, model invoking, and it also includes many chemometric algorithms. Thissoftware optimizes model building and the help system is added, so the software is moreconvenient.Fourthly, the thesis has research on the application of NIR spectroscopy indetecting adulterate milk powder. The experiment mixing maltose in milk powder istaken, and the thesis uses PLS pattern recognition and SIMCA algorithms to build twomodels, comparing several spectroscopy preprocessing methods and the different rangeof NIR wave. The result indicates that the optimal discriminant accuracy of the bothmodel are100%. That is, Near-Infrared (NIR) spectroscopy is feasible in the qualitativediscrimination. Finally, the thesis makes a study in detecting moisture in wheat flour using NIRspectroscopy. It collects101wheat flour samples, and selects the range of moistureabsorption peak (1460nm and1960nm). It builds two calibration models by the meansof linear and nonlinear method that is partial least squares (PLS) and back propagationartificial neural network (BP-ANN). Both of them get good result. Compared toBP-ANN, the PLS model is more accurate. And the subsequent experiments are taken toconfirm the dependability and universality of the PLS model in wheat flour andmedium-strength wheat flour. The thesis puts up a new design of portable and NIRmulti-wavelength moisture meter based on the PLS model. Comparing to the traditionalmoisture meters using2or3wavelength, the new moisture meter adopts16wavelengths, in which12wavelengths are used in measuring and the other4are used inreference. So the meter obtains12absorbance. The new design overcomes the influenceof temperature variation to accuracy and expands the range of the meter’s applications.