High-speed assessment of fat and water content distribution in fish fillets using online imaging spectroscopy.

A nondestructive method using online spectral imaging has been developed for quantitative measurements of moisture and fat distribution in six species of fish fillets: Atlantic halibut (Hippoglossus hippoglossus), catfish (Icatalurus punctatus), cod (Gadus morhua), mackerel (Scomber japonicus), herring (Clupea harengus), and saithe (Pollachius virens). A spectral image cube was acquired for each fish fillet, and a subsampling approach for relating spectral and chemical features was applied. Spectral data was first analyzed by partial least-squares regression (PLSR), and then the regression coefficients were applied pixel-wise to convert the pixel spectra to a meaningful distribution map of moisture and fat contents. The resulting images are called "chemical images", which illustrate the distribution of fat and/or water content in the fillets. The pixel-wise prediction models for water and fat content had a correlation value of 0.94 with root-mean-square error estimated by a cross-validation (RMSECV) of 2.73% and a correlation value of 0.91 with RMSECV of 2.99%, respectively. This technique is suitable for high-speed assessment of quality parameters of biomaterials and should thus be implemented in industrial applications. The product could comprehensively be defined not only in terms of its external features such as size, shape, and color but also in terms of its chemical composition and its spatial distribution.