Quality Evaluation Of Tilapia(Oreochromis) Fillets Using Near Infrared Spectroscopy

In recent years, cultivation industry of tilapia(Oreochromis) in China is developing rapidly. China has become the largest producer and exporter of tilapia in the world. More than sixty percent of the exports are in the form of frozen tilapia and frozen tilapia fillets. Due to temperature fluctuation during storage and transportation, the frozen food may undergo frozen-thawed processes which will lead to a rapid deterioration of their freshness and other quality indicators. So it is necessary to monitor the changes of quality at real time during storage and transportation for ensuring the final quality of the products. However, conventional methods are time-consuming, destructive and unsuitable for application. Near infrared spectroscopy(NIRS) as a rapid and non-destructive method has immense potential for quality evaluation of food.Near infrared spectrometer was used to collect the spectra of dorsal and belly flesh before and after minced from tilapia. Through fitting the spectra to the total volatile basic nitrogen(TVB-N) content, quantitative prediction models of TVB-N can be established. The smoothing average 3 points(sa3), smoothing average 9 points(sa9), smoothing savitzky-golay 9 points(sg9), 1st derivative(Db1), normalization by closure(Ncl), standard normal variate(SNV), and multiplicative scatter correction(MSC) were applied to pretreat the spectra in the process of modeling. The results indicated that, models would show the better prediction accuracy and modeling efficiency while combining other pretreatment methods with Db1 to pretreat the spectra. The best wave number region was selected to get rid of the irrelevant information, and the performance of models were further optimized. The determination coefficient of calibration set and validation set for dorsal flesh before minced increased from 0.870 and 0.821 to 0.973 and 0.925. The standard error were 1.032 and 1.581 dropped down from 2.152 and 2.991. By comparing the models performance, the process of mincing was beneficial for modeling. And the model of belly flesh after minced showed the best performance, which got a determination coefficient of 0.984 with a standard error of 0.879 for validation set. But after considering the actual demands of rapid and non-destructive evaluation, the models established by flesh before minced still have apparent advantage. At last, the belly flesh before minced was used for modeling. The calibration set got a determination coefficient of 0.982 with a standard error of 0.962 and the validation set got a determination coefficient of 0.976, with a standard error of 1.006. The relative standard deviation of calibration set and validation set were 6.66% and 6.69% respectively. The relative percent deviation of calibration set and validation set were 6.76 and 6.95 respectively. This method showed enormous potential application in detecting TVB-N content and non-destructively evaluating freshness of tilapia fillets.Near infrared spectrometer was used to collect the spectra of dorsal flesh from tilapia. Through fitting the spectra to the textural parameters, prediction models of texture can be established. The sa3, sa9, sg9, Db1, SNV and MSC were applied to pretreat the spectra in the process of modeling. The results indicated that, models had showed the better prediction accuracy and modeling efficiency while combining other pretreatment methods with Db1 to pretreat the spectra. The best wave number region was selected to get rid of the irrelevant information and the performance of models were further optimized. By comparing the models performance, models of hardness, spinginess and cohesiveness showed better predicting outcomes than those of the other parameters. The determination coefficient of calibration set and validation set were all above 0.85. The relative standard deviation were less than or close to 10%. And the relative percent deviation were all above 2.5. However, models of gumminess, chewiness and resilience showed bad predicting outcomes. The determination coefficient of calibration set and validation set were all below 0.85. The relative standard deviation were more than 10% or the relative percent deviation were below 2.5. Therefore, the first three models can be more suitable for application.The thawing loss, cooking loss, moisture content, TVB-N and textural parameters of tilapia fillets during frozen-thawed cycles were analyzed to reflect the changes of quality and most of them showed trends of deterioration. Near infrared spectrometer was used to collect the spectra of tilapia fillets in different frozen-thawed cycles. Data pretreatments including Ncl, normalization by maxima(Nma), normalization between 0 and 1(N01), normalization to unit length(Nle), SNV and MSC were selected to pretreat the original spectra. Meanwhile, appropriate wavebands were selected for each qualitative model. After applying principal component analysis(PCA) to the pretreated spectra, fresh and once thawed fillets can be discriminated. Then the qualitative models were esablished by Mahalanobis distance based discrimination. The classification accuracy was ranged from 73.33 to 86.67%. In the discrimination of once and repeated frozen-thawed fillets, better discrimination results were obtained in frozen state than those in thawed state. Mahalanobis distance based discrimination models showed that frozen samples were still conducive to the differentiation and got higher classification accuracy ranging from 80 to 93.33%. Dorsal flesh was more beneficial for the differentiation than belly flesh and got the highest accuracy ranging from 86.67 to 93.33% in frozen state. These results showed that the non-destructive and rapid detection of repeated frozen-thawed cycles in frozen tilapia fillets can be accomplished by NIRS that has enormous potential for practical application.