Research On Application Of Near-infrared Spectroscopy In Several Aquatic Products

Near-infrared spectroscopy (NIRS) has become increasingly important in many fields as a rapid non-destructive analytical technique, such as agriculture, petrochemical, medicine and foodstuff. This technique requires no professional and does no harm to the environment. Research about NIRS in aquatic products is rare, because the moisture of aquatic products is very high and the ingredients of them are highly complex. As a result, it is very difficult to analyze spectral data. In this article, a portable instrument (FQA-NIR GUN) is used to analyze aquatic products with NIRS. The purpose is to bring the advantages of NIRS into aquatic field. Then it could supply theoretic for the design of on-line detection instruments and new machines.1. The core of NIRS is setting up a steady and accurate calibration model. Based on mathematical theory of statistics, process of founding a multivariate linear model and its acceptance standards are explained while sources of errors are analyzed. Techniques such as coefficient of determination, parameter hypothesis testing of model, root mean square error are introduced in detail.2. NIRS models about Pseudosciaena crocea's fat content and freshness (K value) are studied. The absorbance spectra of Pseudosciaena crocea without any processing are collected with NIR GUN. Then fat content and K value is detected with chemical approach. The spectral data are pretreated with different methods. After optimized by genetic algorithm(GA),NIRS models are set up by partial least square method (PLS). Wavelet transform is proved to be an efficient method in distilling useful information from spectral data. GA could optimize the wavelength of spectra and find the most relevant ones. The NIRS model of Pseudosciaena crocea's fat content has good precision and constancy, and is available to application. The model of K value is not as good as that of fat content, but it is also precise enough for quantitative analysis. 3. Study on the NIRS models about fabricated products, taking Lepidotrigla microptera surmi products as an example. The surmi products are made and stored as the ones on the market. Different contents of starch are blended in the products. The products are frozen for some days and thawed, after then absorbance spectra are collected. Different methods are used to pretreat spectral data and wavelength selection is finished by GA subsequently. NIRS models of starch and Lepidotrigla microptera content are made by PLS. The results show that both of the models have good precision and constancy. RPD values of them are both greater than 5, which mean they are available to application. Analyzing starch and Lepidotrigla microptera content of surmi products with NIRS is workable and it is more accurate than detecting natural samples.4. Different pretreatments of spectral data are carried out in MATAB language and environment including the two most important arithmetics in the experiment: PLS and GA. PLS culd distill the useful information from spectral data step by step and avoid multicollinearity of NIRS models. GA could offer a robust search method in complex space and it possesses ability of global search which plays an important role in NIRS. At the same time, some MATLAB programs are coded to implement the arithmetics, and some source codes are listed for someone who needs.Both theoretical and practical researches of NRIS methods for aquatic products detection are studied in this article. Effects of different data pretreatments in analyzing high moisture complex samples are compared in detail. All the tasks will improve the application of NIRS in aquatic field. It is significant for the research on on-line analysis in aquatic processing.