Shrimp Freshness Analysis By Visible And Near Infrared Spectroscopy

As the research object, the farm shrimp was studied using visible/near-infrared spectroscopy technology from two aspects in this paper. On the one hand, fresh shrimp and thawed shrimp were classified by discriminant analysis and other recognition methods; on the other hand, quantitative models were established based on visible/near-infrared spectroscopy technology combined with chemometrics methods for shrimp color indicators (L*, a*, b*) and freshness indicators total volatile base nitrogen (TVB-N) content.The main research conclusions were listed as follows:(1) Spectra were collected on three detecting points of fresh and thawed shrimps to do discriminant analysis. The discriminant accuracy of models in the range from900to1100nm were much better than the model in the range from670to1100nm. Based on these models, the influence of different spectra pretreatment methods on the discriminant model performance were compared. The correct classifications for fresh and thawed shrimp were both100%using discriminant analysis after Savizky-Golay19smooth pretreatment on the first detecting point. The third detecting point had the same correct rate as the first detecting point. The accuracy of the calibration and prediction set model were98.10%and98.08%using discriminant analysis after Savizky-Golay11smooth pretreatment on the second detecting point.(2) The influence of different pattern recognition methods, including discriminant partial least squares (DPLS), principal component discriminant analysis (PCA) and least square support vector machines (LS-SVM) were compared. The results showed that these methods could achieve the correct classification of all samples. Therefore, it could be concluded that fresh shrimp can be classified from thawed shrimp based on visible/near-infrared spectroscopy.(3) After removing spectra and concentration abnormal samples, PLSR models were established. The influence of different spectra pretreatment methods on the quantitative models for shrimp color indicators (L*, a*, b*) had been discussed. Considering both the model performance and stability, the performance of L*, a*, b*content models established in670-1100nm with multiplicative scatter correction pretreatment were much better. The correlation coefficients of calibration (rc), root mean square error of calibration (RMSEC), the correlation coefficients of prediction (rp), root mean square standard error of prediction (RMSEP) were 0.565,0.976,0.561,0.822,0.067,0.990,0.104,0.833,0.443,0.900,0.398,0.847, respectively.(4) After removing spectra and concentration abnormal samples, PLSR models were established. The influence of different spectra pretreatment methods on the quantitative models for shrimp freshness indicators total volatile base nitrogen content (TVB-N) was disscussed. Considering both the model performance and stability, the performance of PLSR TVB-N content model established in670-1100nm with standard normal variate spectral pretreatment was much better, the correlation coefficients of calibration (rc), root mean square error of calibration (RMSEC), the correlation coefficients of prediction (rp), root mean square standard error of prediction (RMSEP) were3.87mg/100g,0.911,5.46mg/100g,0.796, respectively.