Non-destructive assessment of final color and pH attributes of broiler breast fillets using visible and near-infrared hyperspectral imaging: A preliminary study

• Published in: Infrared physics & technology 2018-08, Vol.92, p.309-317
• Main Authors: Jiang, Hongzhe, Yoon, Seung-Chul, Zhuang, Hong, Wang, Wei, Li, Yufeng, Lu, Chengjun, Li, Ning
• Format: Article
• Language: English
• Subjects: Broiler breast fillet; Deboning time; Final quality attribute; Hyperspectral imaging; PLSR
• ISSN: 1350-4495; 1879-0275
• DOI: 10.1016/j.infrared.2018.06.025

•Sensible color24 and pH24 of broiler breast fillets were predicted by hyperspectral imaging.•ROIs and spectra were accurately extracted by 2-D scatter pixels plots of first two PC score images.•Spectral analysis indicated that color24 was affected by deboning time while pH24 was not.•Good results were obtained for L∗24, a∗24 and b∗24 by developed PLSR and simplified RC-PLSR models.•Distribution maps were built to visualize fillet meat color24 and pH24. In poultry industry, the consensus is the final quality of chicken meat should be assessed at 24 h postmortem (PM). Visible and near-infrared (Vis/NIR, 400–1000 nm) hyperspectral imaging (HSI) was adopted to non-destructively assess final color (color24) and pH (pH24) of broiler breast fillets (pectoralis major). 25 fillets of the collected 75 broiler carcasses were deboned at each of three PM times (2, 4 or 24 h). To obtain representative spectra, regions of interest (ROIs) were extracted from hyperspectral images based on pixels selected from the 2-D scatter pixel plots of the first two principal component (PC) score images. Linear discriminant analysis (LDA) showed that color24 was affected by deboning time. Predictive models built with partial least squares regressions (PLSR) performed well for either a∗24 or b∗24 (Rp ≥ 0.87; RPD ≥ 2.02; RER ≥ 7.91), moderately for L∗24 (Rp = 0.75; RPD = 1.45; RER = 5.74), but unsatisfactorily for pH24 which was mainly due to its narrow value range (0.52). Simplified models based on optimal wavelengths selected by regression coefficients (RC) presented better predictive performances for a∗24 and b∗24 while slightly worse results for L∗24 and pH24. Distribution maps were created by pixels prediction in images, and color24 and pH24 within each broiler breast fillet were readily discernible. Overall, Vis/NIR HSI has a good potential to assess color24 and pH24 of chicken meat, but additional sample sizes should be further included to further enhance the prediction capability.