Discrimination of freshwater fish varieties based on near-infrared spectra

作者：Xu Wenjie; Liu Ru; Hong Xiangsheng; Xiong Shanbai^*

来源：Transactions of the Chinese Society of Agricultural Engineering, 2014, 30(1): 253-261.

DOI：10.3969/j.issn.1002-6819.2014.01.032

• Back propagation artificial neural network (BPANN)
• Model calibration and validation
• Multiplicative scatter correction
• Orthogonal signal correction
• Partial least square (PLS)
• Partial least-square regression
• Standard error of calibrations
• Variety

A new method of discriminating varieties of freshwater fish was developed based on the near infrared spectroscopy technology. 665 freshwater fishes (silver carp 100, herring 100, grass carp 100, bighead carp 100, gurnard 76, carp 89, crucian carp 100) were collected to calibrate the model from different sites in Hubei province of China. The freshwater fishes in this study were obtained from pedlars'market, specialized aquatic research laboratory and aquaculture base. Some of samples were bred in ecological circulating water. Others were bred in non-recycled pool. In addition, freshwater fishes were collected for different seasons. A variety of samples were collected for the model calibration. The near infrared spectra of seven different varieties of 665 freshwater fish samples were analyzed. The discrimination of freshwater fish was conducted with near infrared spectral technology combined with chemometric method. The preprocessing of the spectra can reduce error of prediction, background optical noise and light scattering effects. The development of mathematical models is critical for near infrared spectroscopy. The preprocessing methods used in this study included multiplicative scatter correction (MSC), savitzky-golay (S-G) smoothing, savitzky-golay (S-G) derivative, differential derivation (DD), no spectral data preprocessing (NONE), standard normal variate (SNV), standardization (S), net analyte signal (NAS), orthogonal signal correction (OSC), de-trend (DT). The optimal spectra bands were chosen once determining the preprocessing method. All spectra statistical calculations were conducted using the RIMP software. Samples were randomly divided into two groups, 532 samples for the preparation of the model calibration and the remaining samples for the model validation. The near infrared discrimination models were developed based on partial least square (PLS) regression, principal component analysis (PCA) combined with back propagation artificial neural network (BP-ANN), partial least square combined with back propagation artificial neural network with 1000~1799nm, respectively. The model performance was evaluated by coefficient of determination of calibration (RC), standard error of calibration (SEC), correlation coefficient of validation (RP), the bias-corrected standard error prediction (SEP) and the standard error of cross validation (SECV). The lower typical errors (SEC, SECV and SEP), and higher correlation coefficient indicates that the model is robust. The results showed that discrimination model obtained by PCA combined with BP-ANN was the best. The standardization was the best preprocessing method for discrimination model. The best spectral bands were 1000~1200 and 1300~1450 nm of model for the optimal preprocessing method. The linear equation of the model was y = 0.922637x + 0.333081. This equation represents the relationship between NIR spectroscopy predicted and observed values. High correlation coefficients were reached. The accuracy of discrimination rates of samples were 0.9644 and 0.9553 for the model calibration and validation set, respectively. The model performance demonstrates acceptable accuracy and predictive ability. The results indicate that the near infrared spectroscopy has the potential to become a valuable rapid screening method for discriminating freshwater fish.

• 出版日期2014
• 单位华中农业大学