Detection Of Ochratoxin A In Red Wine Based On Three-dimensional Fluorescence Spectrum

Ochratoxin A (OTA), is a mycotoxin produced by several Aspergillus and Penicillium species growing in different agricultural commodities in the field or during storage. OTA has nephrotoxic, teratogenic, carcinogenic, and immunotoxic activity in several animal species. According to the International Agency for Research on Cancer (IARC), OTA has been included in group2B as a possible carcinogen to humans. So it’s important to find a fast detection of OTA in red wine. Currently, the main analytical methods to detect OTA are high pressure liquid chromatography (HPLC), enzyme linked immunosorbent assay (ELISA), and so on. This thesis creatively used three-dimensional fluorescence spectroscopy to detect the concentration of OTA in red wine. Firstly, pre-processed the fluorescence spectroscopy data, and then used support vector regression (SVR) and support tensor regression (STR) to build the mathematic model separately. Finally we could use these models to detect the concentration of OTA in red wine. The main work of this thesis is organized as following.1. Prepared36red wine samples with different concentrations of OTA, and get the fluorescence data of these samples. Using interpolation method to eliminate the Rayleigh scattering light of the original spectrum and then using S-G cubic polynomial fitting method to smooth the data in order to reduce noise and improve the spectral data.2. Researched support vector regression modeling method, and used this method to process the data. We compared the performance of SVR with other vector-based methods such as PCA, PLS. The result showed that the model based on SVR worked best. Two parameters should given when using SVR.3. Researched support tensor regression modeling method, and used this method to process the data. We compared the performance of STR with SVR. The result showed that the performance of STR was better than SVR. But STR calculated slower than SVR, and needed long computation time.4. Derivative spectrum can eliminate baseline drift, enhance feature bands, and overcome bands overlap. So we relatively used SVR and STR modeling methods processing derivative spectrum. Result showed that the modeling performance of derivative spectrum improved greatly.