Potential Of Hyperspectral Imaging For Evaluating Prawns Freshness

Prawn, a kind of nutrient-rich and delicious foods, has always been regarded as one of the most commercially important aquatic products as well as the most favored seafood in China. Due to the nature of easy deterioration, it is very important to develop a rapid and accurate method for freshness evaluation in prawn during the process of transport, storage and sales. However, the traditional methods for freshness evaluation(such as physical and chemical methods) are time-consuming, laborious and destructive, which are unable to meet the requirements of non-destructive, real-time and accurate testing in the modern food industry. This study was committed to investigate the potential of using hyperspectral imaging in the wavelength region of visible and short-wave near infrared range(400-1000 nm) for the rapid and non-destructive evaluation of prawn freshness, which could provide technical support for the future of food monitoring system. The main contents and results are as follows:1.The texture profile analysis(TPA) properties and color properties of Metapenaeusensis were determined based on visible and short-wave near infrared(VIS/NIR) hyperspectal imaging system. Six TPA properties(hardness, springiness, cohesiveness, gumminess, chewiness and resilience) and three color attributes(L*、a* and b*) were measured to representthe texture and color variations of prawn in cold storage, respectively. With proper pre-treatment, SPA algorithm was used to choose the most important wavelengths from hyperspectrum dataset. Based on the selected important wavelengths, prediction models were then established for TPA properties and color properties using partial least square regression(PLSR), RBF artificial neural network(RBF-NN) and least-square support vector machine(LS-SVM). For TPA properties, the SPA-LS-SVM model was considered as the best predictive model to detect the hardness, gumminess and chewiness values of prawn with correlation coefficient of prediction(Rp) of 0.81, 0.80 and 0.84, and root mean square error of prediction(RMSEP) of 0.402, 0.163 and 0.174 uint, respectively. In combination with LS-SVM, acceptable results were obtained for L* and b* with Rp of 0.88 and 0.85, RMSEP of0.716 and 0.685, respectively. The best predicting models were then applied to generate distribution maps of TPA properties and color properties in each pixel of the hyperspectral image for prawn.2.A VIS/NIR hyperspectal imaging system coupled with wavelet analysis was implemented to determine the total volatile basic nitrogen(TVB-N) contents of prawn during cold storage. Three wavelet features, namely energy, entropy, modulus maxima, were extracted from the uninformative variable elimination(UVE) pre-treated spectrum. Quantitative models were then established between the wavelet features and the reference measured TVB-N contents by using three regression algorithms, including PLSR, RBF-NN and LS-SVM. The LS-SVM model based on modulus maxima features was considered as the best model for determining the TVB-N contents of prawns, with an excellent Rc of 0.99, Rp of 0.98, RMSEC of 0.256 mg N/100 g, RMSEP of 0.712 mg N/100 g. The best predicting model was then applied to generate a visualization map of TVB-N distribution in each pixel of the hyperspectral image for prawn.3.The total viable counts(TVC) of prawn were determined based on visible and short-wave near infrared hyperspectal imaging system. With MSC pre-treatment, SPA was conducted to choose the most efficient wavelengths. Meanwhile, eight texture variables were extracted from the first three PC images using gray-level co-occurrence matrix(GLCM). Based on the important wavelengths, texture variables and important wavelengths + texture variables, quantitative models were established for predicting TVC in prawns using PLSR, RBF-NN and LS-SVM, respectively. As a result, similar results were obtained when important wavelengths and important wavelengths + image texture feature were used as the input dataset. The LS-SVM model based on important wavelength +image texture feature was considered as the best model for the TVC determination, with Rc of 0.97, RMSEC of 0.045, Rp of 0.95, RMSEP of 0.437. Considering for the simplicity, the important wavelength in combination with LS-SVM was used to generate a visualization map of TVC distribution in each pixel of the hyperspectral image for prawn.