Visible and Near Infrared Hyperspectral Imaging for Non-Destructive Grading and Classification of Chicken Breast Fillets

This study investigated the potential of visible and near infrared (Vis/NIR) hyperspectral imaging (HSI) for grading and classification of pale, soft, and exudative (PSE), dark, firm, and dry (DFD), and normal chicken breast fillets. Hyperspectral images of boneless and skinless chicken breast samples were acquired with spectra in the wavelengths between 400 and 1000 nm. All samples were divided into PSE, normal, and DFD categories based on their color and pH values. Spectral pre-processing algorithms of Savitzky-Golay (S-G) smoothing, S-G first and second derivative processing, and standard normal variate (SNV) were applied to the spectral data obtained from region of interest (ROI) to reduce noises and enhance the performance of partial least square-discriminant analysis (PLS-DA) models. Full-wavelength model based on the second derivative processed spectra obtained the highest correct classification rate (CCR) of prediction set with value of 84.62 %. Twelve wavelengths were selected from full wavelengths by using Successive projection algorithm (SPA) to build new PLS-DA classification model. CCR value of prediction set was 84.62 % for the simplified model, the same as that for the full-wavelength model. Results suggest that Vis/NIR HSI can be used as a useful tool to grade and classify chicken breast meat.