Application of VNIR hyperspectral imaging for non-destructive prediction of pH, color and drip loss of chicken breast fillets

Abstract. Non-destructive and rapid prediction of quality attributes of chicken breast fillets using visible and near-infrared (VNIR) hyperspectral imaging (400-1000 nm) was carried out in this work. All hyperspectral images were acquired for bone (dorsal) side of chicken breast. A forward principal component analysis (PCA) and its reverse rotation was firstly conducted to reduce noises and multicollinearity. A band threshold method was adopted on PC₁ score image to get the region of interest (ROI) of each sample, then the average reflective spectra of ROI of each image were acquired by reverse PCA rotation. Partial least square regression (PLSR) was utilized to correlate the spectra with measured pH, L* and drip loss values. Informative wavelengths were selected using competitive adaptive reweighed sampling (CARS) to build new PLSR models. Better results were acquired with determination coefficient of prediction R_p^2/RPD of 0.75/1.86, 0.85/2.52 and 0.68/1.69 for pH, L* and drip loss, respectively. Distribution maps of pH, L* and drip loss were generated based on the improved PLSR models. The results demonstrated that VNIR hyperspectral imaging technique can be used to predict quality attributes of chicken breast fillets.