Nondestructive Determination of the Total Volatile Basic Nitrogen (TVB-N) Content Using hyperspectral Imaging in Japanese Threadfin Bream (Nemipterusjaponicus) Fillet

Background and Objectives: Considering the importance of safety evaluation of fish and seafood from capture to purchase, rapid and nondestructive methods are in urgent need for seafood industry. This study aimed to assess the application of hyperspectral imaging (HSI: 430-1010 nm) for prediction of total volatile basic nitrogen (TVB-N) in Japanese-threadfin bream (Nemipterusjaponicus) fillets, as a marine fish, during 8 days of cold storage (4±2°C).  Materials & Methods: Hyperspectral imaging data and TVB-N value of the fillets were obtained in the laboratory. The basic prediction model was established based on Back-propagation artificial neural network (BP-ANN). To simplify the calibration models, 10 wavelengths were selected based on regression-coefficient (RC). Multiple-linear regression (MLR) and BP-ANN models were established based on the selected wavelengths. Results: In full spectral range, the BP-ANN models exhibited relatively weak prediction performance (R2P=0.76 and RMSEP=4.45). After selecting 10 wavebands, the capability of the simplified models was better than that of the full-wavebands. The predictive power of simplified BP-ANN was better than that of MLR model (R2P(RC-BP-ANN)=0.820; RMSEPRC-BP-ANN=3.79 and R2P(RC-MLR)=0.794 and RMSEPRC-MLR=4.25). Therefore, r RC-BP-ANN model showed  more acceptable predictive performance (0.82 R2P 0.90). Conclusion: Although the effectiveness of the developed simple multispectral imaging system based on BP-ANN model showed promising results to predict the TVB-N values of fillets, it did not show a strong prediction power of TVB-N values during storage. Therefore, further researches are required to enhance the prediction power and suitability of HSI method to evaluate TVB-N value in Japanese threadfin bream fish.