An improved nondestructive measurement method for salmon freshness based on spectral and image information fusion

Abstract

The freshness of salmon is one of the important qualities that consumers care about. This study found that not only spectral data, but also the image information was effective in predicting the freshness of salmon. Therefore, this paper proposed a novel method for evaluating the freshness of salmon by fusing spectra and image information. Salmon RGB images of different storage times were dimensionally reduced by principal component analysis (PCA) algorithm and integrated with 400–700 nm spectral data. Then a neural network model was built to extract features of the fused data and used to predict the total viable counts (TVC) and total volatile basic nitrogen (TVB-N) values of the salmon. The results show that 92.3% prediction accuracy could be achieved when predicting the storage time of the test sets. When predicting the values of TVC and TVB-N, the RMSEP could reach 0.36 lg cfu/g and 1.78 mg/100 g, respectively, and both of the determination coefficients (R2p) could reach 0.92, which were all better than using only spectral data or image data. Thus the results indicated that the novel method could effectively improve the accuracy and model performance when predicting the freshness of salmon.