Predicting Aged Beef Tenderness with a Hyperspectral Imaging System

1
Predicting Aged Beef Tenderness with a
Hyperspectral Imaging System

• L. M. Grimesa, G. Konda Naganathanb, J. Subbiahb,
  C. R. Calkinsa
• aDepartment of Animal Science
• bDepartment of Biological Systems Engineering
• University of Nebraska, Lincoln
•  

2
Tenderness Consumers

• Consumers have shown a willingness to pay a
  premium for guaranteed tender steaks (Boleman et
  al., 1997, Shackelford et al., 2001)
• The industry currently sells 15 to 20 tough
  steaks to consumers (Miller et al., 2001)
• Industry Goal To devise a reliable,
  non-destructive on-line method of tenderness
  prediction

3
Hyperspectral Imaging Uses

• Application/Research
• Remote Sensing by NASA and military
• Nutrient deficiencies in plants
• Fecal contamination in chicken
• Fungal/bacterial contamination in fruits

4
Hyperspectral Imagery

• Combines both biochemical and muscle structure
  data
• Data consists of dozens of narrow, adjacent,
  spectral bands, creating a continuous spectrum
  for each image cell
• Pixel-by-pixel basis, rather than averaging the
  data across the entire image

5
Individual Pixels
6
2006 Research

• 96.4 prediction accuracy of 14 d
• tenderness from scans of steaks
• at 14 d postmortem

7
Objective

• Develop a hyperspectral imaging
• system that accurately predicts
• 14 d aged beef tenderness
• from 2 d postmortem scans

8
Materials and Methods

• An InGaAs digital video camera (1000-1700 nm) and
  spectrograph
•  
• Diffuse-flood lighting system with
  tungsten-halogen lamps and a dome with a white
  reflectance coating
• Steaks were placed on a moving platform
• Images were taken in 2 nm wavelength intervals
•  

9
Imaging Machine
10
Imaging Spectrometer
11
Materials and Methods

• USDA Choice and Select grade longissimus steaks
  (n314) from between the 12th and 13th ribs
• Scanned at 2 d postmortem, vacuum packaged and
  aged to 14 d, then frozen
• Thawed overnight, cooked on an impingement oven
  and slice shear force values were obtained 

12
Statistical Analysis

• A region-of-interest (ROI) was selected
•  
• A mean spectrum of the ROI was obtained, and
  analyzed
• with partial least squares (PLS) regression to
  obtain the
• loading vector
•  
• PLS bands were generated and textural
  co-occurrence
• matrix analysis was conducted to develop the
• canonical discriminant model
•  
• Leave-one-out cross validation for prediction
• SSF Categories Tender 21.0 kg Intermediate
  21.1-25.9 kg, Tough 26.0 kg
•  

13
Results

• Tender 78.13 Accuracy
• Intermediate 76.19 Accuracy
• Tough 62.5 Accuracy
• Total system accuracy of 77.07

14
Conclusions

• This hyperspectral imaging system
• was effective in predicting
• 14 d beef tenderness from scans at
• 2 d postmortem.
•  

15
ANY QUESTIONS?

• Questions
• And
• Suggestions