Early identification of mushy Halibut syndrome with hyperspectral image analysis

Abstract

Mushy Halibut Syndrome (MHS) is a condition that appears in Greenland halibut and manifests itself as abnormally opaque, flaccid and jelly-like flesh. Fish affected by this syndrome show poor meat quality, which results in negative consequences for the fish industry. The research community has not carefully investigated this condition, nor novel technologies for MHS detection have been proposed. In this research work, we propose using hyperspectral imaging to detect MHS. After collecting a dataset of hyperspectral images of halibut affected by MHS, two different goals were targeted. Firstly, the estimation of the chemical composition of the samples (specifically fat and water content) from their spectral data by using constrained spectral unmixing. Secondly, supervised classification using partial least squares discriminant analysis (PLS-DA) was evaluated to identify specimens affected by MHS. The outcomes of our study suggest that the prediction of fat from the spectral data is possible, but the prediction of the water content was not found to be accurate. However, the detection of MHS using PLS-DA was precise for hyperspectral images from both fillets and whole fish, with lower bounds of 75% and 83% for precision and recall, respectively. Our findings suggest hyperspectral imaging as a suitable technology for the early screening of MHS.