Pork quality and marbling level assessment using a hyperspectral imaging system

Share Embed


Descrição do Produto

Pork quality and marbling level assessment using a hyperspectral imaging system Jun Qiao c, Michael O. Ngadi a,*, Ning Wang a, Claude Garie´py b, Shiv.O. Prasher a a

Department of Bioresource Engineering, McGill University, Macdonald Campus, Ste-Anne-de-Bellevue, Que., Canada H9X 3V9 b Agriculture and Agri-food Canada, Saint-Hyacinthe, Que., Canada J2S 8E3 c China Agricultural University, Beijing 100083, PR China

Abstract Pork quality is usually evaluated subjectively based on color, texture and exudation characteristics of the meat. In this study, a hyperspectral imaging-based technique was evaluated for rapid, accurate and objective assessment of pork quality. In addition, marbling level was also automatically determined. The system was able to extract spectral characteristics of pork samples. Appropriate spatial features were obtained for marbling distribution in pork meat. Existing marbling standards were scanned, and indices of the marbling scores were formulated by co-occurrence matrix. The principal component analysis (PCA) method was used to compress the entire spectral wavelengths (430–1000 nm) into 5, 10 and 20 principal components (PCs), which were then clustered into quality groups. Artificial neural network was used to classify these groups. Results showed that reddish, firm and non-exudative (RFN) and reddish, soft and exudative (RSE) samples were successfully grouped; the total corrected ratio was 75–80%. The feed-forward neural network model yielded corrected classification as 69% by 5 PCs and 85% by 10 PCs. Angular second moment was successfully used to determine marbling scores excepting the score at 10.0. Forty samples were sorted and the result showed that the samples’ marbling score ranged from 3.0 to 5.0. Ó 2007 Elsevier Ltd. All rights reserved. Keywords: Hyperspectral imaging; Pork quality; Marbling; PCA; Cluster analysis; Neural network

1. Introduction Canada is one of the largest pork exporters in the world. With market expansion and segmentation, the Canadian processing industry needs efficient technologies for assessment of pork quality in order to maintain its leading position. Quality of fresh pork varies greatly. Traditionally, pork quality used to be classified into three categories based on color, texture (firmness) and exudation (drip loss). Pork meat that is classified as RFN (reddish pink, firm and non-exudative) has a desirable color, and normal texture and water-holding capacity (WHC). PSE (pale pinkish gray, very soft and exudative) meat has an undesirable appearance and lack firmness due to excessive drip loss. DFD (dark purplish red, very firm and dry) meats

have a firm and sticky surface with high WHC, very little or no drip loss, and very high pH. Over the years two other categories namely RSE and PFN have arisen. RSE (redish, soft and exudative) has a normal color, but a soft texture and an exudative character similar to PSE (Kauffman, Cassens, Scherer, & Meeker, 1992). PFN is pale, firm and non-exudative (Nam, Du, Jo, & Ahn, 2002). PFN and RSE have been recognized recently as major quality defects in Canada, which account for >13% in all defects compared to PSE (13%) and DFD (10%) (Murray, 2000). According to the author, exudative pork can induce an economic loss of $5 per carcass. An efficient and effective quality assessment system is urgently needed for the meat industry to identify the defects rapidly and objectively. Marbling is the intermingling of fat with lean in a muscle and it is usually assessed subjectively by visual assessment. Marbling scores encompasses size, number and

distribution of fat particles (Jeremiah, 1998). Visual standards grade the marbling in seven scores by the National Pork Board (NPB, 1999). Some research works were reported for assessing beef or pork marbling by machine vision (Faucitano, Huff, Teuscher, Gariepy, & Wegner, 2005; Shiranita, Hayashi, Otsubo, Miyajima, & Takiyama, 2000; Tan, 2004; Toraichi et al., 2002; Yoshikawa et al., 2000). Shiranita, Miyajima, and Takiyama (1998) used a co-occurrence matrix to extract standard texture vectors from beef samples sorted by professional grader and successfully classified the unevaluated samples. Hyperspectral imaging techniques can provide not only spatial information, as regular imaging systems, but also spectral information for each pixel within an image. This information will form a three-dimensional ‘‘hypercube” which can be analyzed to ascertain minor and/or subtle physical and chemical features of an object. Thus, a hyperspectral image can be used to detect physical and geometric characteristics such as color, size, shape, and texture. It can also be used to extract some intrinsic chemical and molecular information (such as water, fat, protein, and other hydrogen-bonded constituent) from a product. Recently, several hyperspectral imaging research were reported on quality assessment for meat, fruit and vegetables. Kim, Chen, and Mehl (2001) introduced a hyperspectral reflectance and fluorescence imaging system for food quality and safety. Cheng et al. (2004) developed a method to inspect damage of cucumber by hyperspectral image. A similar approach was successfully developed to inspect the contamination of chicken carcasses (Kim, Kim, Chen, & Kong, 2004; Yang, Chao, & Chen, 2005). In this study, the potential of hyperspectral imaging techniques was exploited for pork quality and marbling assessment. The specific objectives were to evaluate different groups of pork quality based on spectral information acquired from the hyperspectral imaging system using principle components analysis (PCA) and cluster analysis; develop models for classifying the quality levels by artificial neural network (ANN); and to estimate marbling scores of pork by image texture indices extracted from digitized meat marbling standard.

perature between 20 and 22 °C. Samples were allowed to bloom for 20 min before spectral analyses. 2.2. Spectral image collection and processing 2.2.1. Hyperspectral imaging system The hyperspectral imaging system consisted of a linescan spectrograph (ImSpector, V10E, Spectra Imaging Ltd, Finland), a CMOS camera (BCi4-USB-M40LP, Vector International, Belgium), a DC illuminator (Fiber-Lite PL900-A, Dolan-Jenner Industries Inc, USA), a conveyer (Dorner 2200 series, Donner Mfg. Corp., USA), an enclosure, a data acquisition and preprocessing software (SpectraCube, Auto Vision Inc., CA, USA), and a PC as shown in Fig. 1. The ImSpector collected spectral images in a wavelength range of 400–1000 nm with a spectral resolution of 2.8 nm and a spot radius
Lihat lebih banyak...

Comentários

Copyright © 2017 DADOSPDF Inc.