Visualisation of Fat and Fatty Acid Distribution in Beef Using a Set of Filters Based on near Infrared Spectroscopy

Food quality is strongly affected by its components and their spatial distributions. Recently, spectroscopic methods have been widely applied as a non-destructive and rapid method to measure food quality. Although it is a versatile technique, the measurement system is extremely costly for practical use. In this paper, we propose a simple measurement system using a small set of band-pass filters. A food constituent was predicted using output from the band-pass filters as input for a multiple linear regression model, and the bands were designed to obtain high prediction accuracy characterised by the determination coefficient, using hyperspectral data by the optimisation approach. We designed three sets of filters to separately determine contents such as oleic acid, total unsaturated fatty acid and fat content in raw beef using NIR hyperspectral data, and then we implemented these designs as real optical filters. By mounting the filter in front of the lens of an NIR monochrome camera, we captured a set of filtered images. We then performed a pixel-by-pixel prediction of the content to enable the spatial distribution to be visualised. The determination coefficient (R²) and prediction error, which we characterised by the root mean square error of cross-validation (RMSECV), of this filtering method (R² = 0.638–0.739, RMSECV = 3.13–5.15) were superior to those obtained with partial least squares (PLS) regression using hyperspectral measurements (R ² = 0.610–0.643, RMSECV= 3.70–6.12). Our method, therefore, facilitates the application of a hyperspectral technique for practical use.