Abstract:
Abstract: The identification of geographical origin of camellia oil is very significant in food market to maintain the market order and promote fair competition. The objective of this research was to evaluate the feasibility of a high efficient and nondestructive detection of the geographical origin of camellia oil by using near infrared spectroscopy combined with chemometrics methods. In this paper, 4 kinds of camellia oil samples obtained from Hunan, Jiangxi, Anhui and Zhejiang Province were tested with 39 samples for each kind. The samples were randomly divided into 2 groups, i.e. calibration set (30, 27, 27 and 24 were respectively for Hunan, Jiangxi, Anhui and Zhejiang) and validation set (9, 12, 12 and 15 were respectively for Hunan, Jiangxi, Anhui and Zhejiang). Use the FTIR (Fourier transform infrared spectroscopy) spectrometer to gather the samples' near infrared spectrum information. The raw spectral data contain not only chemical information of camellia oil samples, but also some interferential information produced by environmental factors, so reasonable preprocessing method was adopted to eliminate the influence of these factors, which helped to improve the accuracy and reliability of modeling and prediction. Four spectral preprocessing methods were proposed, including the Savitzky-Golay (SG), the multiplicative scattering correction (MSC), the first derivation (FD) and the vector normalization (VN). These methods were conducted by using the Unscrambler 10.3 software. The partial least squares (PLS) was proposed to establish the identification model. Meanwhile, the principal component analysis (PCA) and PLS were used to extract the best principal components (PCs), which were used as input variables for the back propagation artificial neural network (BPANN) to establish the PCA-BPANN and PLS-BPANN identification models. Then, the results of prediction were compared according to the correlation coefficient of validation (RP) and the root mean square error of validation (RMSEP). The results of experiment showed that the identification of the geographical origin of camellia oil was ideal using the SG-PLS-DA and SG-PLS-BPANN-DA model, and the correct recognition rates of calibration set and validation set were more than 90%. In calibration set, one sample of Hunan was mistakenly identified as the sample of Jiangxi, one sample of Anhui was mistakenly identified as the sample of Jiangxi, 3 samples of Jiangxi were mistakenly identified as the samples of Hunan, and one sample of Jiangxi was identified as the sample of Anhui using the SG-PLS-DA method; and the correct recognition rate was 94.44%. And in validation set, 2 samples of Jiangxi were mistakenly identified as the samples of Hunan, and 2 samples of Anhui were mistakenly identified as the samples of Jiangxi; the correct recognition rate was 91.67%. The correlation coefficient of calibration (RC) was 0.942 and the root mean square error of calibration (RMSEC) was 0.189, and the RP and the RMSEP were 0.932 and 0.192, respectively. There was one sample of Hunan mistakenly identified as the sample of Jiangxi in calibration set, and the validation set was the same. There was one sample of Jiangxi mistakenly identified as the sample of Anhui in calibration set, and there was one sample of Anhui mistakenly identified as the sample of Jiangxi in validation set using the SG-PLS-BPANN-DA model; the correct recognition rate of validation set was 95.83%. Therefore, the identification model established by the SG-PLS-BPANN-DA method was better. The RC and the RMSEC were respectively 0.974 and 0.170, and the RP and the RMSEP were respectively 0.972 and 0.172. Consequently, the SG-PLS-BPANN-DA model can more accurately identify the origin of camellia oil, and provide technical support for quickly, non-destructively and accurately identifying the geographical origin of camellia oil.