Novel non-invasive distribution measurement of texture profile analysis (TPA) in salmon fillet by using visible and near infrared hyperspectral imaging

Abstract

This study developed a pushbroom visible and near-infrared hyperspectral imaging system in the wavelength range of 400–1758nm to determine the spatial distribution of texture profile analysis (TPA) parameters of salmon fillets. Six TPA parameters (hardness, adhesiveness, chewiness, springiness, cohesiveness, and gumminess) were analysed. Five spectral features (mean, standard deviation, skew, energy, and entropy) and 22 image texture features obtained from graylevel co-occurrence matrix (GLCM) were extracted from hyperspectral images. Quantitative models were established with the extracted spectral and image texture signatures of samples based on partial least squares regression (PLSR). The results indicated that spectral features had better ability to predict TPA parameters of salmon samples than image texture features, and Spectral Set I (400–1000nm) performed better than Spectral II (967–1634nm). On the basis of the wavelengths selected by regression coefficients of PLSR models, instrumental optimal wavelengths (IOW) and predictive optimal wavelengths (POW) were further chosen to reduce the high dimensionality of the hyperspectral image data. Our results show that hyperspectral imaging holds promise as a reliable and rapid alternative to traditional universal testing machines for measuring the spatial distribution of TPA parameters.