Beef quality parameters estimation using ultrasound and color images
- Jose Luis Nunes†1,
- Martín Piquerez†1,
- Leonardo Pujadas†1,
- Eileen Armstrong2,
- Alicia Fernández1 and
- Federico Lecumberry1Email author
© Nunes et al.; licensee BioMed Central Ltd. 2015
Published: 23 February 2015
Beef quality measurement is a complex task with high economic impact. There is high interest in obtaining an automatic quality parameters estimation in live cattle or post mortem. In this paper we set out to obtain beef quality estimates from the analysis of ultrasound (in vivo) and color images (post mortem), with the measurement of various parameters related to tenderness and amount of meat: rib eye area, percentage of intramuscular fat and backfat thickness or subcutaneous fat.
An algorithm based on curve evolution is implemented to calculate the rib eye area. The backfat thickness is estimated from the profile of distances between two curves that limit the steak and the rib eye, previously detected. A model base in Support Vector Regression (SVR) is trained to estimate the intramuscular fat percentage. A series of features extracted on a region of interest, previously detected in both ultrasound and color images, were proposed. In all cases, a complete evaluation was performed with different databases including: color and ultrasound images acquired by a beef industry expert, intramuscular fat estimation obtained by an expert using a commercial software, and chemical analysis.
The proposed algorithms show good results to calculate the rib eye area and the backfat thickness measure and profile. They are also promising in predicting the percentage of intramuscular fat.
Such procedures are done in inhospitable environments and consist of repetitive tasks that are tedious for the expert, with high error rates linked to fatigue and inspector's mood. In [3, 4], methods for automatic segmentation of the rib eye in color images are proposed. These methods separate meat from "non-meat" (fat and bones). However, this method treats equally all the meat in the image, since the rib eye is not always surrounded by "non-meat" and includes other adjacent muscles in detection. Some works , avoid this problem removing other regions of the steak leaving only the rib eye beef; clearly this method is not suited for evaluating carcass at the slaughterhouse. In this work we propose a method based on curve evolution both for rib eye area and subcutaneous fat thickness measurements.
There are several previous work in this kind of applications, such as [7, 8] addressing the estimation of the IMF% in ultrasound images for livestock. In  the rib area was used as a determinant factor in the estimation of beef quality.
The production method used in Latin America usually includes a high component of extensive farming (although feedlot is used too) impacting in the amount of IMF%, while in other regions the feedlot production is preferred [10, 6, 11]. Therefore, IMF% in animals analysed in previous works such as [12, 6, 7, 13] might be different from the animals analysed in the present work and predictors should be adjusted in this case.
In this work we present new methods to automatically measure three beef quality parameters, IMF% in ultrasound images, rib eye area and backfat thickness in color images in slaughtered steers.
Automatic measurement methods
This section describes the proposed methods for ribeye detection and area estimation, then we present the method for subcutaneous fat detection which allows to estimate the thickness and to extract the fat profile. Finally the proposed strategy for IMF% is presented.
Rib eye detection and area estimation
The rib eye area estimation is used as a quality indicator and is also used for carcass categorizing the animal. Database images show a cross-sectional area of the longissimus dorsi muscle (dorsal length) at the level of the 12th intercostal space. The objective is to estimate the area of the muscle, commonly measured manually by an expert [5, 2].
This works proposes a curve evolution algorithm, more precisely an adaptation of the algorithm Distance Regularized Level Set Evolution (DRLSE). This method is a recent improvement of the level set techniques [15, 16]. The DRLSE method adds an intrinsic capability to the level set framework for maintaining regularity of the level set function, particularly maintaining the property of signed distance function around the zero level set. This method is used in an edge-based active contour segmentation that modifies the geodesic active contour proposed by Caselles . Therefore, edge information computed from the image is used to create the energy functional that govern the level set evolution.
where G σ is the Gaussian kernel with the standard deviation σ.
In general, the adjustment of the curve to the rib eye edges achieved in these two steps is quite accurate. However, a third stage of evolution adjustment is applied, performing ten more iterations in order to achieve finer adjustment.
Measurement of the backfat thickness in color images is made on the cross section at the level of twelfth intercostal space, perpendicular to the outer edge of the fat and up to a quarter of the end of the muscle longissimus dorsi respect to the backbone. Fat provides desirable attributes, a well distributed coverage associated with a creamy-white color, are considered ideal; however excessive amounts of fat must be removed (industrial process called "conditioning"), which significantly decreases meat yield.
The proposed method automatically determines the axis of the rib eye's first moment and measures the thickness in the corresponding points. Moreover, the method computes a full profile of thickness for all the backfat, allowing to measure the thickness along it. The algorithm is based in a curve evolution from an initial curve to a target curve, assigning a thickness proportional to the iterations needed in each point to reach the target curve.
It needs two closed curves whose intersection is empty and the backfat is between. These curves will be the input of the curve evolution algorithm. To provide good detection, the initial curve should match the inner boundary of the backfat and a section of the target curve must match the outer limit of the backfat.
As said before, backfat should be measured in two points given by the perpendicular projection of half and 3/4 lengths of rib eyes's main axis. This point are geometrically located from the rib eye segmentation, computing its main axis.
Intramuscular fat percentage estimation
Intramuscular fat percentage (IMF%) is the proportion of fat between the muscle fibers of the rib eye. This quality measure can be performed in color images in slaughtered animals, and in ultrasound images in live animals. It has been shown that intramuscular fat percentage is highly correlated with tenderness [20, 2], which is highlighted for consumers as one of the most determinant factors in beef quality. Therefore an automatic system for its measurement results fundamental.
In this work we propose its measure through two types of images, color images and ultrasound images.
The algorithm proposed includes three stages: region of interest detection, features extraction and selection and modelling for the estimation.
Region of interest detection
Features extraction and selection
Features extracted from the US images
- mean µ*
- std deviation σ*
- contrast ratio
- percentile (each 20%)
- variance coefficient
- power percentile (× 5)
Local Binary Pattern (LBP)
As a result of the feature acquisition stage we obtain a 42-dimension feature space. To reduce the space dimension in order to improve computational performance a feature selection stage based on Principal Components Analysis (PCA) was done, finding that 99% of the variance is accumulated in the first ten components. As a result of the PCA a new space of ten new features, linear combinations of the original ones, was used to do the IMF% estimation model.
Modeling for the estimation
Support Vector Regression (SVR) is a variant of the classic Support Vector Machine algorithm. The basic idea of SVR consist in mapping the training data, , into a larger dimensional space via a nonlinear mapping Φ: , where a linear regression can be performed. For more details on SVR see [21, 22].
In this work, a radial basis function ) was used as kernel. Parameters γ and tolerance of termination criterion were optimized based on the data train set.
The same strategy was applied for the post mortem color images, using the rib eye area as the ROI. Similar performance was obtained using specific descriptors for color data (intensity mean in different color channels, Fourier coefficients, number of pixels in each channel).
Experiments and results
We worked with a database of 153 steers acquired at the slaughter using a special hardware controlling distance and lighting. Only for 71 of them having ultrasound images. Ultrasound images were collected at a cattle ranch in Uruguay.
A measure of the rib eye area performed by an expert for all the database was used for the validation process, as well as a measure of backfat 1/2 and 3/4 for 51 steers.
IMF% was measured by chemical gas chromatography analysis and used as ground truth to validate the regression results. The lipid extraction protocols used are described in ; its margin of error in the measurement is less than 0.3%. An estimation of the IMF% from ultrasound images was performed by an expert from the meat industry using a commercial software. 283 ultrasound images were acquired, four images were taken per animal. The ultrasound hardware used was the Aquila Pro Vet, an industry standard equipment.
Rib eye area
The database was divided into two sets randomly drawn, one to train the algorithm and the parameters optimization (103 images, 2/3 of the dataset) and the other to test it (50 images, 1/3 of the dataset).
Two performance indicators were used,
Relative area error:(2)
where A union = R auto ∪ R manual y A inter = R auto ∩ R manual . With R auto rib eye area automatically detected by the implement algorithm and R manual rib eye area measured by an expert manually.
Results for the rib eye area estimation for images with errors less than 10%.
Results for the rib eye area estimation for images with errors less than 15%.
Results for the backfat thickness estimation.
Intramuscular fat percentage
The database of ultrasoud images was divided into two sets randomly drawn, one to train the algorithm and compute the linear regression coefficients (184 images, 2/3 of the dataset) and the other to test it (92 images, 1/3 of the dataset).
Results for the IMF% estimation.
Conclusions and future work
New automatic methods to measure different beef quality parameters from the analysis of ultrasound (in vivo) and color images (post mortem) are presented. The parameters estimated are related to meat quality, meat yield and consumer's health: eye area steak, percentage of intramuscular fat and backfat thickness or subcutaneous fat.
First of all, we propose to measure the rib eye area, using an algorithm based on curve evolution. The results are satisfactory in over 77% of the processed images (seven different data sets with different acquisition conditions). The method is very robust and general, making it suitable for the segmentation of other muscle regions. In controlled conditions results are even better (more than 89%). Second, we propose to measure backfat thickness from the profile of distances between a curve that limits the steak and one that limits the rib eye previously detected. The automatic distance measurement achieves a very precise estimate using a novel evolution strategy. Beside this results, the algorithm gives the backfat thickness profiles, which provides information on the uniformity of the subcutaneous fat which is an important carcass quality information.
Last, we propose a procedure for estimating the intramuscular fat percentage based in Support Vector Regression in a region of interest automatically determined for ultrasound images (rectangle) and for the color images (rib eye). The performance of the automatic selection of the ROI for ultrasound was highly satisfactory, more than 96% of the database were well detected, in the remaining 4% where the ROI was wrongly detected, the software gives an alert and allows for a manual definition. The prediction of intramuscular fat showed a better adjustment only in the middle range of fat percentages (3%-5%). However this error in our approach is lower than the error in the expert's estimation. The overall performance is promising, clearly a deeper analysis of the features considered is needed. In the color images the main difference is in ROI determination and the set of features used, but we obtain similar performance results.
In future work we propose to analyse new strategies in the feature extraction and selection stages, both for ultrasound and color images, for intramuscular fat percentage estimation. We also want to explore the impact of different parameters in the estimation, such as the ROI's area and location.
This work was partially supported by ANII grant FMV 2_2011_1_7376 and Project ANII PR_FSA_2009_1_1383. The authors would like to thanks, Gessy Druillet, Marcela Eugster for their contribution on the database acquisition and analysis, Alvaro Gómez, Giovanni Gnemmi, Gregory Randall for their expertise support and Matías Tailanián for poviding the code of the curve evolution algorithm used in the backfat measurement.
Publication costs for this article were funded by the Department of Signal Processing, Facultad de Ingeniería, Universidad de la República, Montevideo, Uruguay.
This article has been published as part of BMC Bioinformatics Volume 16 Supplement 4, 2015: Selected articles from the 9th IAPR conference on Pattern Recognition in Bioinformatics. The full contents of the supplement are available online at http://www.biomedcentral.com/bmcbioinformatics/supplements/16/S4.
- J K, D L: Trends in Meat Consumption and the Need for Fresh Meat and Meat Products of Improved Quality. Improving the Sensory and Nutritional Quality of Fresh Meat. 2009, Woodhead Publishing Ltd, Sawston, Cambridge, UKGoogle Scholar
- USDA L: Agricultural Marketing Service, Division, S.: United States Standards for Grades of Carcass Beef. 1997Google Scholar
- Jackman P, Sun D-W, Du C-J, Allen P: Prediction of beef eating qualities from colour, marbling and wavelet surface texture features using homogenous carcass treatment. Pattern Recognition. 2009, 42 (5): 751-763. 10.1016/j.patcog.2008.09.009.View ArticleGoogle Scholar
- Jackman P, Sun D-W, Allen P: Automatic segmentation of beef longissimus dorsi muscle and marbling by an adaptable algorithm. Meat Science. 2009, 83 (2): 187-194. 10.1016/j.meatsci.2009.03.010.View ArticlePubMedGoogle Scholar
- Hale DS, Goodson K, Savell JW: USDA Beef Quality and Yield Grades. Department of Animal Science Texas A&M AgriLife Extension Service College Station TX 77843-2471. 2013, Accessed: 2014-05-23, [http://meat.tamu.edu/beefgrading/]Google Scholar
- Harron W, Dony R: Predicting quality measures in beef cattle using ultrasound imaging. Computational Intelligence for Image Processing, 2009. CIIP '09. IEEE Symposium. 2009, 96-103.View ArticleGoogle Scholar
- Li C, Zheng Y, Kwabena A: Prediction of imf percentage of live cattle by using ultrasound technologies with high accuracies. Information Engineering, International Conference. 2009, 2: 474-478.Google Scholar
- Greiner SP, Rouse GH, Wilson DE, Cundiff LV, Wheeler TL: Prediction of retail product weight and percentage using ultrasound and carcass measurements in beef cattle. Journal of Animal Science. 2003, 81 (7): 1736-1742.PubMedGoogle Scholar
- Pinheiro TR, Mercadante MEZ, Albuquerque LG, Cyrillo JNSG, Branco RH: Phenotypic and genetic parameters compared during repeated measures of longissimus muscle area and subcutaneous fat thickness in nelore cattle. Genet Mol Res. 2011, 10 (4): 2944-52. 10.4238/2011.November.29.5.View ArticlePubMedGoogle Scholar
- Gresham JD: International study guide real time ultrasound beef cattle applications. University Scholars Designation at the Universtiy of Tennesse at Martin. 2006Google Scholar
- Aass L, Fristedt C-G, Gresham JD: Ultrasound prediction of intramuscular fat content in lean cattle. Livestock Science. 2009, 125 (2-3): 177-186. 10.1016/j.livsci.2009.04.005.View ArticleGoogle Scholar
- Brethour JR: Using receiver operating characteristic analysis to evaluate the accuracy in predicting future quality grade from ultrasound marbling estimates on beef calves. J Anim Sci. 2000, 78 (9): 2263-8.PubMedGoogle Scholar
- Li J, Tan J, Martz FA, Heymann H: Image texture features as indicators of beef tenderness. Meat Science. 1999, 53 (1): 17-22. 10.1016/S0309-1740(99)00031-5.View ArticlePubMedGoogle Scholar
- Li C, Xu C, Gui C, Fox MD: Distance regularized level set evolution and its application to image segmentation. IEEE Trans Image Process. 2010, 19 (12): 3243-3254.View ArticlePubMedGoogle Scholar
- Osher S, Sethian JA: Fronts propagating with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations. Journal of Computational Physics. 1988, 79: 12-49. 10.1016/0021-9991(88)90002-2.View ArticleGoogle Scholar
- Sapiro G: Geometric Partial Differential Equations and Image Analysis. 2001, Cambridge University Press, New York, NY, USAView ArticleGoogle Scholar
- Caselles V, Kimmel R, Sapiro G: Geodesic active contours. Int J Comput Vision. 1997, 22 (1): 61-79. 10.1023/A:1007979827043.View ArticleGoogle Scholar
- Otsu N: A Threshold Selection Method from Gray-level Histograms. IEEE Transactions on Systems, Man and Cybernetics. 1979, 9 (1): 62-66.View ArticleGoogle Scholar
- Tailanian M, Lecumberry F, Gnemmi G: Measuring thickness profile between closed curves applied to endometritis detection on in dairy cattle. Technical report, Facultad de Ingeniería, IIE - Projecto ANII FMV 2 2011. 2014, 1: 73-76. [http://iie.fing.edu.uy/publicaciones/2014/TLG14]Google Scholar
- Association, NCB: An evolving industry. 2013Google Scholar
- Drucker H, Burges CJC, Kaufman L, Smola A, Vapnik V: Support vector regression machines. Advances in Neural Information Processing Systems 9. Edited by: Mozer M, Jordan M, Petsche T. 1997, MIT Press, Cambridge, MAGoogle Scholar
- Smola A, Schölkopf B: A tutorial on support vector regression. Statistics and Computing. Invited paper. 1998Google Scholar
- Folch J, Lees M, Stanley GHS: A simple method for the isolation and purification of total lipides from animal tissues. Journal of Biological Chemistry. 1957, 226 (1): 497-509.PubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.