Hyperspectral Imaging for Evaluation of Herring Stocks caught around Iceland

Abstract

The correct identification of herring stocks is significant for the sustainable fisheries management and the maintenance of the ecosystem balance in Icelandic waters. Norwegian spring-spawning herring (NSSH) and Icelandic summer-spawning herring (ISSH) are morphological alike and share common feeding grounds, therefore, their distinction can be challenging. This research study focuses on the application of hyperspectral imaging (HSI) combined with multivariate analysis to identify these herring stocks non-destructively and to predict their water content. In total were 80 herring samples examined using a Maritech Eye® hyperspectral imaging system equipped with a HySpex camera (485–960 nm). Among these samples, 54 were NSSH and 26 were ISSH. Partial Least Squares Discriminant Analysis (PLS-DA) was utilized for the stock classification. The water content of the fish was predicted through the Partial Least Squares Regression (PLS-R) of the spectral data, compared to traditional analytical results. According to the results, the accuracy of the PLS-DA model while using the whole fish ROI approach was 93.7 % and it outperformed the method of using a smaller rectangular ROI having an accuracy of 87.5%. The water content for the NSSH was on average 60.7% ± 2.2 g/100 g sample, while the average water content for the ISSH was 63.2% ± 2.7 g/100 g sample. Satisfactory predictive results were demonstrated by the PLSR modelling by revealing a correlation coefficient R² of 0.81 during training, even though the performance declined slightly during cross-validation (R² = 0.72) and prediction (R² = 0.65). HSI together with chemometrics is thus an effective approach to differentiate between herring stocks, and to evaluate their water content. This study thus supports the fact that HSI is a fast, reliable and non-destructive analytical technique.