Hyperspectral imaging for real-time monitoring of water holding capacity in red meat

• Published in: Food science & technology Vol. 66; pp. 685 - 691
• Main Authors: Kamruzzaman, Mohammed, Makino, Yoshio, Oshita, Seiichi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2016
• Subjects: Hyperspectral imaging; Multispectral imaging; Multivariate analysis; Red meat; Water holding capacity; Red meat; Multispectral imaging; Multivariate analysis; Water holding capacity; Hyperspectral imaging
• ISSN: 0023-6438; 1096-1127
• DOI: 10.1016/j.lwt.2015.11.021

A hyperspectral imaging system was investigated for determination of feature wavelengths to be used in a design of a multispectral system for real-time monitoring of water holding capacity (WHC) in red meat. Hyperspectral images of different red meat samples were acquired in the spectral range of 400–1000 nm and partial least-squares regression (PLSR) and least square support vector machine (LS-SVM) models were developed. Feature wavelengths were selected using regression coefficients (RCs) and competitive adaptive reweighted sampling (CARS). The best set of feature wavelengths was determined using RCs and the best calibration model obtained was based on RCs-LS-SVM. The model obtained an R2p of 0.93 and RPD of 4.09, indicating that the model is adequate for analytical purposes. An image processing algorithm was developed to transfer this model to each pixel in the image. The results showed that instead of selecting different sets of wavelengths for beef, lamb, and pork, a subset of feature wavelengths can be used for convenient industrial application for the determination of WHC in red meat. The pixel wise visualization of WHC obtained with the aid of image processing was another advantage of using hyperspectral imaging that cannot be obtained with either imaging or conventional spectroscopy. •Hyperspectral imaging was used for real time monitoring of WHC in red meat.•Calibration models were developed using PLSR, and LS-SVM.•Feature wavelengths were selected using RCs and CARS.•The best calibration model was obtained based on RCs-LS-SVM.•Pixel wise prediction maps were created with the aid of image processing.