1
Prediction of sensory characteristics of lamb meat samples by
2
near infrared reflectance spectroscopy
3
S. Andrés1,*, I. Murray1, E.A. Navajas2, A.V. Fisher,3 N.R. Lambe2, and L. Bünger2
4
1
5
AB21 9YA, UK
6
2
7
Edinburgh EH9 3JG
8
3
9
Langford, Bristol BS40 5DU
SAC Life Science Group, Ferguson Building, Craibstone Estate, Bucksburn, Aberdeen,
Sustainable Livestock Systems Group, Scottish Agricultural College, King’s Buildings,
Division of Farm Animal Science, School of Veterinary Science, University of Bristol,
10
11
12
13
*
14
Bucksburn, ABERDEEN, AB21 9YA, Tel. +44 (0) 1224 711 201, Fax. +44 (0) 1224
15
711 292. Present address: Sonia Andrés, Departamento de Producción Animal,
16
Universidad de León, E-24071 León, Spain. Tel. +34 987 291 235 Fax +34 987 291 311
17
E-mail: sandres@unileon.es
Corresponding author: Sonia Andrés, SAC, Ferguson Building, Craibstone Estate,
18
1
19
ABSTRACT
20
This study was implemented to evaluate the potential of visible and near infrared
21
reflectance (NIR) spectroscopy to predict sensory characteristics related to the eating
22
quality of lamb meat samples. A total of 232 muscle samples from Texel and Scottish
23
Blackface lambs was analyzed by chemical procedures and scored by assessors in a
24
taste panel (TP). Then, these parameters were predicted from Vis/NIR spectra. The
25
prediction equations showed that the absorbance data could explain a significant but
26
relatively low proportion of the variability (R2 < 0.40) in the taste panel traits (texture,
27
juiciness, flavour, abnormal flavour and overall liking) of the lamb meat samples.
28
However, a top-tail approach, looking at the spectra of the 25 best and worst samples as
29
judged by TP assessors, provided more meaningful results. This approach suggests that
30
the assessors and the spectrophotometer were able to discriminate between the most
31
extreme samples. This may have practical implications for sorting meat into a high
32
quality class, which could be branded, and into a low quality class sold for a lower price
33
for less demanding food use.
34
Regarding the chemical parameters, both intramuscular fat and water could be more
35
accurately predicted by Vis/NIR spectra (R2 = 0.841 and 0.674, respectively) than
36
sensory characteristics. In addition, the results obtained in the present study suggest that
37
the more important regions of the spectra to estimate the sensory characteristics are
38
related to the absorbance of these two chemical components in meat samples.
39
40
KEYWORDS: NIR spectroscopy, meat quality, lamb, sensory characteristics,
chemical composition, taste panel
41
42
2
43
INTRODUCTION
44
Lamb production is an important part of UK agriculture, contributing more than 10% of
45
total livestock output (Defra, 2005). It also makes a very important contribution to
46
maintaining employment and infrastructure in rural communities and helps manage and
47
enhance landscape and biodiversity, especially in less favoured areas. For the UK sheep
48
industry to continue as a major producer and exporter of lamb it is essential that its
49
economic viability is improved. To do so, it has to provide carcasses that better meet
50
market requirements, since currently only ca. 55% of UK lambs meet target
51
specifications (MLC, 2002).
52
There is also an urgent need to refocus the sheep industry on the customer’s needs.
53
The Dewar-Durie review of the Scottish Sheep Industry (Dewar-Durie, 2001) provided
54
telling arguments to justify this and the recommendations are entirely appropriate for
55
the whole of the UK. Too often sheep meat is of variable quality, prejudicing its
56
competitiveness in the marketplace.
57
The sheep industry therefore needs a correctly functioning, transparent value-based
58
marketing system, which has to send clear and accurate market signals from the
59
consumer backward through the whole supply chain to primary producers. It must have
60
the means to identify the value of individual animals or carcasses based on carcass and
61
meat eating quality (Cross & Whittaker, 1992). This requires an objective, accurate
62
method of predicting lean meat yield and monetary value of individual lamb carcasses
63
so that pricing structures can be based on individual carcass merit. New systems based
64
on video image scanning and analysis technology (VISA, e.g. Stanford, Richmond,
65
Jones, Robertson, Prince & Gordon, 1998; Hopkins, Safari, Thompson & Smith, 2004)
66
provide a non-invasive method operating at normal abattoir chain speeds and enable
3
67
automatic acquisition of data on carcasses from the side and back view. However these
68
systems cannot classify on the basis of meat quality and need therefore to be augmented
69
with other suitable systems to measure meat eating quality traits.
70
Near Infrared (NIR) technology provides complete information about the molecular
71
bonds and chemical constituents in a sample scanned, so it is a convenient tool not only
72
for characterising foods, but also for quality measurements and process control. Optical
73
devices coupled to computers offer potentially very fast data acquisition that may permit
74
decision-making on meat eating quality; albeit from a selected small surface area only.
75
For example Shackelford et al. (2004) could scan 348 carcasses per hour by averaging
76
20 spectra from an area of 19.6 cm2 of the major spinal muscle longissimus thoracis.
77
Therefore it is not surprising that there is substantial interest to use NIR on-line to
78
predict chemical parameters in the meat industry and to augment existing VISA systems
79
(Schwarze, 1996). Although distances from the production line and errors introduced by
80
sampling have been major obstacles, efforts are being made to improve NIR on-line
81
application owing to the meaningful spectra provided by this procedure. Previous
82
studies have shown that, in addition to chemical information, physical and technological
83
characteristics of meat have been measured using NIR (Mitsumoto, Maeda, Mitsuhashi,
84
& Ozawa, 1991; Swatland, 1995; Geesink, Schreutelkamp, Frankhuizen, Vedder, Faber,
85
Kranen, & Gerritzen, 2003; Leroy, Lambotte, Dotreppe, Lecocq, Istasse, & Clinquart,
86
2003; Cozzolino & Murray, 2004).
87
Eating quality and palatability are related to the interactions of many chemical and
88
physical properties of the meat. For example, it is well known that juiciness and
89
tenderness depend not only on fat content but also on the water retention in a meat
90
product as well as on structural characteristics such as the contractile state of muscle
4
91
fibres. Moreover consumers do not perceive attributes such as tenderness in isolation
92
but in relation to other aspects of eating quality such as juiciness (Warriss, 2000). If all
93
or some of these attributes could be predicted using the overall information contained in
94
the Vis/NIR spectra, this could be the first step towards on-line implementation of
95
NIRS.
96
The aims of this study were to investigate the association between chemical
97
composition and meat quality traits scored by a trained sensory panel and absorbance
98
data obtained from NIR spectroscopy, using meat samples from two contrasting breeds
99
of lamb.
100
101
MATERIALS AND METHODS
102
Experimental animals: Ewe and entire ram lambs of two divergent breeds, Texel
103
(TEX) and Scottish Blackface (SBF), were included in this two year study. All lambs
104
were grazed in single sex, mixed-breed groups at the SAC sheep unit, from weaning
105
until reaching a target condition score of 3 (on a scale of 0 to 5) and thresholds for live
106
weight (35 and 32 kg in 2003 and 2004, respectively, due to slower growth rates in the
107
second year). Age at slaughter ranged from 130 to 234 days old, with an average age of
108
171d. More details on the experiment have been given elsewhere (Navajas, Glasbey,
109
McLean, Fisher, Charteris, Lambe, Bünger & Simm, 2006).
110
Meat samples: Meat samples were obtained from 232 animals, with approximately
111
equal numbers from each breed. For NIR analysis, a 10mm thick cross-section of M.
112
longissimus thoracis was removed from the 12/13 rib region of the right sides of all
113
carcasses, within one hour of slaughter, vacuum packed and frozen at -20 ºC. They were
114
then transported for approximately 24 hours on solid CO2 and kept at –20 °C until they
5
115
were used for Vis/NIR spectroscopy. For the taste panel assessment, the M. longissimus
116
lumborum was removed from all right sides the day after slaughter, vacuum packed,
117
aged for seven days (slaughter day = day 0) at 2-4 ºC and frozen at -20 ºC.
118
Chemical analysis: The moisture and intramuscular fat contents of the longissimus
119
lumborum muscle were determined. A frozen cross-sectional slice of the muscle was
120
removed from its cranial end, vacuum packed and frozen. After thawing each sample
121
was blended to a fine paste using a laboratory blender. Sub-samples (25 mg) were
122
weighed into pre-dried and weighed plastic pots, frozen, and freeze dried (72 hrs) using
123
an Edwards Modulyo Unit (BOC Edwards, Wivelsfield Green, West Sussex, UK). Each
124
sample was then removed and further dried in a Townson and Mercer Vacuum Oven for
125
a further 5 hours, removed desiccated, cooled and re-weighed. The moisture content
126
was reported as the loss in weight of each sub-sample.
127
The intramuscular fat was extracted from each of the dried and crushed samples using
128
petroleum ether (B.P. 40-60 degrees C) as the solvent in a modified Soxhlet extraction
129
using an automatic Gerhardt Soxtherm 2000 unit (Gerhardt Gmbh, Koningswinter, DE).
130
pH measurement: pH was measured in the longissimus lumborum (loin) muscle in
131
all carcasses at 45min., 3h. and 24h post slaughter by direct probing using a Testo 230
132
pH meter with a glass electrode.
133
Taste panel preparation and assessment: Frozen samples were thawed at 4 ºC
134
overnight and then cut into 2cm thick steaks. They were cooked in a contact grill
135
(George Forman Double Knockout grill, model 10635) until the internal temperature
136
reached 75 ºC, measured by a thermocouple inserted into the geometric centre of the
137
sample. Between 6 and 10 assessors rated 2cm3 samples of each muscle. The assessors
138
used 8-point category scales to evaluate the following traits:
6
139
texture (1 – extremely tough, 8 – extremely tender); juiciness (1 – extremely dry, 8 –
140
extremely juicy); lamb flavour intensity (1 – extremely weak, 8 – extremely strong)
141
abnormal flavour intensity (1 – extremely weak, 8 – extremely strong) and overall liking
142
(1 – dislike very much, 8 – like very much).
143
144
Spectra collection: The meat samples were thawed in a fridge for 24 hours and then
145
taken out, stored in plastic bags to prevent water evaporation from the surface and left at
146
least one hour at ambient temperature. The surface temperature was recorded by an IR
147
'gun' (IRtec, Miniray 100, Eurotron) and digital colour photos of each sample were
148
taken. Afterwards each sample was trimmed to eliminate connective tissue and a piece
149
of intact meat of 35 mm diameter was cut parallel to the longitudinal orientation of the
150
muscle fibres (Cozzolino & Murray, 2004) and put inside 35 mm diameter quartz
151
cuvettes with aluminium foil backing. The samples were photographed twice inside the
152
cuvette and finally NIR-scanned on both sides in order to obtain a mean spectrum per
153
sample. The diffuse reflectance spectra were collected at 2 nm intervals from 400 to
154
2498 nm (1050 data points per sample at 16-bit precision) using a NIRSystems 6500
155
scanning spectrophotometer (FOSS NIRSystems, Silver Spring, MD, USA) equipped
156
with a spinning module to keep the sample rotating during the scanning time. Thus, the
157
area of the sample scanned could be increased and thereby the sampling error reduced
158
(Downey & Hildrum, 2004). Absorbance data were stored as log 1/R, R being the
159
reflectance. The instrument was operated by the software package NIRS2 version 3.01
160
(InfraSoft International, State College, PA, USA).
161
Data analysis: Calibration development and validation were performed using the
162
software WinISI II version 1.02 (InfraSoft International, State College, PA, USA).
7
163
Spectral data pre-treatments such as standard normal variate and detrending (SNV-D),
164
multiplicative scatter correction (MSC) and first or second order derivatives were
165
applied to the spectra with the purpose of reducing noise and scattering effects. Partial
166
least-squares regression (PLSR) was used for predicting sensory properties using
167
Vis/NIR spectra as independent variables.
168
Two types of outliers can be distinguished; T statistic outliers, whose predicted value
169
exceed 2.5 times the standard error of estimate, and H statistic outliers whose spectra
170
are atypical. The T statistic outliers suggest the reference data are suspect while H
171
statistic outliers suggest their spectra are peculiar for any reason (InfraSoft International,
172
State College, PA, USA). In order to avoid these samples, two passes of elimination of
173
outliers were allowed.
174
Internal full cross validation was performed in order to avoid over-fitting the PLSR
175
equations. Thus, the optimal number of factors in each equation was determined as the
176
number of factors after which the standard error of cross validation (SECV) no longer
177
decreased substantially. The accuracy of prediction is given by SECV (see formula 1).
178
This error is calculated by repeating the PLSR as many times as samples are kept in the
179
calibration set, but always withholding a sample from the modelling process so that the
180
model may be optimised by blind prediction of the sample withheld. When finished,
181
every sample has been treated as an “unknown” test object, and the SECV is
182
mathematically expressed as:
183
SECV 
1 I
 yi  yˆ i 2

I i 1
(1)
184
185
RESULTS AND DISCUSSION
186
Taste panel and chemical data
8
187
The range, simple statistical means and their standard deviation of the sensory
188
properties (texture, juiciness, flavour, abnormal flavour and overall liking) given by the
189
assessors in the taste panels and of the chemical (intramuscular fat and water) and pH of
190
the lamb meat samples are summarised in Table 1. The coefficient of variation was
191
lowest for water (1.4%) and highest for IMF (45%) whereas all sensory traits showed
192
intermediate variability (11 to 18%) and the values for pH showed a relative low
193
variability as well (3-4%).
194
Table 2 shows the correlation among all the chemical and sensory characteristics
195
scored by the taste panel. The main traits influencing the overall liking of the meat were
196
texture (tenderness), flavour and abnormal flavour with correlations between –0.57 and
197
0.65. The chemical parameters (IMF and water content) were positively and negatively
198
correlated, respectively, with texture, flavour and overall liking.
199
200
Study of the spectra
201
The spectra of the 232 lamb meat samples showed excessive noise beyond 1900 nm
202
due to the high absorbance and short penetration path-length at long wavelengths, so
203
this area was avoided in the study. Shorter wavelengths (VIS & Herschel NIR; 400-
204
1400nm) seemed to be more analytically useful.
205
Variation in intramuscular architecture will affect not only tenderness but also the
206
light penetration path length into the muscle. This results in changed IR absorbance of
207
the more tender/tough samples. Figure 1 shows the average spectra (400-1900 nm) of
208
the 5 most tender meat samples vs. the average spectra of the 5 least tender meat
209
samples according to the TP assessors. As can be observed, the most tender samples
210
showed higher absorbance in the visible (400-950 nm) and lower absorbance in the NIR
9
211
region (1100-2498 nm) of the spectra. This is in accordance with results reported by
212
Hildrum, Nielsen, Mielnik, & Næs (1994), who observed that tougher beef samples had
213
overall higher absorption in the NIR region than the tender specimens and in particular
214
between 1150 and 1300 nm. Later work confirmed this fact (Hildrum, Isaksson, Næs,
215
Nielsen, Rødbotten, & Lea, 1995; Park, Chen, Hruschka, Shackelford, & Koohmaraie,
216
1998; Rødbotten, Mevik, & Hildrum, 2001).
217
These variations could be explained by differences in the cold-shortening process
218
(NIR scanned samples were frozen pre-rigor), which could affect the path-length when
219
the sample was irradiated to record its spectrum. According to the Beer Lambert law the
220
absorbance will increase multiplicatively with increasing sample mean penetration path-
221
length (Hildrum et al., 1995). Therefore a possible explanation for the higher
222
absorbance values in the NIR region of the tougher samples (very shortened
223
sarcomeres) can be a deeper penetration path-length than samples with longer
224
sarcomere length (Rødbotten et al., 2001). This is especially important because the meat
225
samples were scanned in their intact form, that is to say, not minced or ground.
226
Nonetheless, this fact does not explain why the absorbance data for the tough
227
samples in the present study were lower in the visible region. Maybe a possible reason
228
can be found in the different ability of the meat to scatter the light depending on the pH.
229
At low pH the meat appears pale and has a greater ability to scatter the light so that the
230
amount of absorbed light is low (Warriss, 2000). Low ultimate pH also increases
231
oxidation of haem pigments, purple or red myoglobin and oxymyoglobin changing to
232
brown metmyoglobin. The ability of the pigment to selectively absorb green light, and
233
therefore appearing red, is consequently reduced (Warriss, 2000).
10
234
In any case, extreme samples show spectra with very different tendencies (Figure 1)
235
so it is possible that information related to the organoleptic characteristics of the meat
236
samples could be explained using the absorbance data. In order to check this possibility
237
prediction equations were developed.
238
239
Prediction of the sensory and chemical characteristics
240
Prediction of organoleptic quality parameters using Vis/NIR spectra as independent
241
variables was explored initially using the whole sample set (n = 232). Table 3
242
summarises the statistics regarding these mathematical models. As can be observed
243
(Table 3), none of these equations showed coefficients of determination higher than 0.4,
244
so the predictability of the sensory characteristics from Vis/NIR spectra seems relatively
245
poor. A validation of these novel findings with literature data was not possible as no
246
studies concerning the estimation of sensory properties using Vis/NIR spectra of lamb
247
meat samples have been found. Nevertheless, our results are in broad agreement with
248
those found for beef samples by other authors trying to predict juiciness, flavour and
249
overall liking (Hildrum et al., 1994; Byrne, Downey, Troy, & Buckley, 1998;
250
Rødbotten, Nielsen, & Hildrum, 2000; Venel, Mullen, Downey, & Troy, 2001).
251
However, in most of these studies (Hildrum et al., 1994; Byrne et al., 1998; Rødbotten
252
et al., 2000; Liu, Lyon, Windham, Realini, Pringle, & Duckett, 2003) tenderness could
253
be predicted with a higher degree of accuracy (R2 = 0.45-0.90) compared with our
254
study (R2  0.40). In this respect it is of note that all these authors used meat samples
255
with different ageing times, so the variability related to the structural properties of the
256
muscle fibres was larger than in our study. On the contrary, we tried to simulate NIR
257
on-line conditions for industry, where all the samples must be scanned immediately
11
258
post-mortem even if it is to predict organoleptic and other quality attributes. With this
259
regard, all the NIR samples were frozen pre-rigor without ageing, whereas all the TP
260
samples were aged for the same duration, so this could be another explanation for the
261
poor results obtained in the present study.
262
As reported by other authors (Hildrum et al., 1994; Byrne et al., 1998; Rødbotten et
263
al., 2000; Venel et al., 2001), the prediction of the sensory characteristics may be
264
improved if the group is segregated in more specific sub-groups. Therefore, criteria such
265
as breed, and sex were used for dividing the initial sample set into sub-groups; however
266
the correlations in these sub-groups were not higher (results not shown).
267
An improvement in R2 was observed when, according to the 'overall liking'-
268
parameter, the 25 best and the 25 worst meat samples were selected (Table 4), however
269
the R2 values were still mostly below 0.8, a value that might be considered as a
270
workable threshold for reliability. Also, there is a risk that the use of a small sample set
271
reduces the accuracy of prediction and may overfit the data.
272
In accordance with our results, Liu et al. (2003) found no improvement when the
273
samples were sub-grouped according to different criteria. The discrepancy between all
274
these studies may be attributed to the use of different meat samples and their intrinsic
275
variability arising from different types of animals, slaughter ages, meat ageing times etc.
276
for the development of prediction equations, the number of samples used for calibration
277
and the precision of the reference values.
278
In the present study the sensory parameters that were found to be more correlated
279
with the spectral data of the 232 lamb meat samples were juiciness, flavour and overall
280
liking, but even in these three cases the associations were relatively low (Table 3). The
281
spectral data predicted intramuscular fat and water much better than any of the sensory
12
282
characteristics. The relationships between intramuscular fat and eating quality traits are
283
themselves low, albeit important. Tenderness, juiciness and flavour are complex traits,
284
being affected by many factors in production and processing and their variation having
285
multiple biological causes. There is also an element of subjectivity in their estimation
286
by the taste panel, however well trained (Warriss, 2000). In agreement with Hoving-
287
Bolink, Vedder, Merks, de Klein, Reimert, Frankhuizen, van den Broek, & Lambooij,
288
en E. (2005), the relationships between spectral data and pH were weak when all
289
animals were considered (Table 3) but there was a high R2 (0.89) involving pH
290
the restricted data set (Table 4), suggesting that early post-mortem drop in pH was
291
associated with changes in absorbance in these extreme examples.
45 min
in
292
It must also be mentioned that the present experiment was designed to examine
293
typical lamb meat samples obtained from normal production systems under grazing
294
conditions. Thus, animals were slaughtered with a normal commercial age range for
295
slaughter lambs, that is to say, age ranged from 4.5 to nearly 8 months for those animals
296
assessed by taste panel. Some were also quite fat at the time of slaughter (several at
297
condition score 4). However, lambs slaughtered at a fixed age, rather than being
298
finished at a fixed live weight and condition score as here, might have shown a higher
299
variability, thus allowing the assessors to find clearer differences among them. In
300
addition, nutrition is a factor that influences IMF and its composition (Varnam &
301
Sutherland, 1995; Wood, Richardson, Nute, Fisher, Campo, Kasapidou, Sheard, &
302
Enser, 2003), but all these animals were grazed on the same pastures. Moreover, all of
303
these samples were aged for one, not several, conditioning period after slaughter.
304
It has also to be noted that the tissue samples that were NIR-scanned (M. longissimus
305
thoracis) were not strictly the same tissue samples tasted by the panellists (M.
13
306
longissimus lumborum), so this is another factor which can partially explain the lower
307
correlation between panel scores and NIRS.
308
However, as stated previously, the spectra corresponding to samples scored with
309
extreme values by the taste panels were indeed totally different. In other words, the
310
assessors could discriminate among the most extreme samples, these differences also
311
being detected by the spectrophotometer. This finding can have practical implications. It
312
would be possible to use visible and NIR spectroscopy to classify these two extremes
313
(Næs & Hildrum, 1997; Park, Chen, Hruschka, Shackelford, & Koohmaraie, 2001).
314
Maybe this is the most important challenge, because it seems improbable that the
315
consumer can detect small changes in any of these sensory parameters when the
316
samples are closer to the mean.
317
A principal component analysis (PCA) of the second derivative spectra of the 50
318
extreme lamb meat samples was performed to detect the main source of variation in the
319
meat samples. Thus, the original absorbance data matrix was reduced to a coordinate
320
axis system (principal components, PCs) where the samples were located according to
321
their scores for each PC. Two different clusters could be observed according to breed,
322
with almost no overlap between Texel (T) and Scottish Blackface (B) meat samples
323
(Figure 2). A possible explanation of this fact is the significant differences for fatty acid
324
composition found for the two breeds (data not shown). Vatansever, Kurt, Enser, Nute,
325
Scollan, Wood, & Richardson (2000) also perceived a clear genetic effect on fatty acid
326
composition of beef from different cattle breeds. In our study, this source of variation
327
might have been detected by Vis/NIR spectra (García-Olmo, De Pedro, Garrido-Varo,
328
Jiménez, Salas, & Santolalla, 2000; García-Olmo, Garrido-Varo, & De Pedro, 2001). It
329
has been demonstrated that changes in fatty acids give rise to variation in flavour, and
14
330
hence in overall acceptability of the meat (Varnam & Sutherland, 1995; Wood et al.,
331
2003). Notwithstanding, the direct effect of breed on eating quality is often very small
332
and confused with other factors such as nutrition, parasite stress and processing
333
procedures (Dundon, Sundstrom, & Gaden, 2000).
334
In order to identify any specific region of the spectra that might be useful to classify
335
samples in the absence of taste panel data, the correlation between overall liking and the
336
absorbance data along the wavelength space was plotted (Figure 3). The region around
337
the third stretching overtone of the C-H bonds of the IMF (950 nm) and the O-H bonds
338
of the water (890 nm) (Shenk, Westerhaus, & Workman, 1992) showed important
339
correlations with the overall liking attribute. Water and IMF are constituents influencing
340
the perception of the overall acceptability of the meat by the assessors (Table 2). This
341
region of the spectra (850-1000 nm) provides information on both from a large depth of
342
penetration, thus avoiding longer wavelengths that might be less repeatable due to
343
greater absorption (Shackelford, Wheeler, & Koohmaraie, 2004). So even if these
344
wavelengths (850-1000 nm) only explain approximately 30% of the variance in the
345
extreme specimens (that is to say, the coefficient of correlation R = ± 0.55, Figure 3),
346
this region of the NIR spectrum may have practical value to identify lamb eating quality
347
in the absence of data from a taste panel.
348
15
349
CONCLUSIONS
350
In light of the present results it seems possible to use spectroscopy to discriminate
351
between samples with extreme scores for sensory eating quality attributes. The region of
352
the spectrum between 890 and 1000 nm seems to be particularly useful for this purpose,
353
as it significantly related to the water and IMF content of meat samples. This could have
354
practical advantages when meat cuts with low or high overall quality need to be
355
identified for sorting into classes or grades for branding or downgrading for less
356
demanding food use. However the procedure has to be improved for on-line application.
357
358
ACKNOWLEDGEMENTS
359
The financial support of the Department for Environment, Food and Rural Affairs
360
(Defra) and the Scottish Executive Environment and Rural Affairs Department
361
(SEERAD) are gratefully acknowledged. S. Andrés is grateful to the “Fundación
362
Alfonso Martín Escudero” (Spain) for financial assistance under a post-doctoral grant.
363
364
REFERENCES
365
Byrne, C. E., Downey, G., Troy, D. J., & Buckley, D. J. (1998). Non-destructive
366
prediction of selected quality attributes of beef by near-infrared reflectance
367
spectroscopy between 750 and 1098 nm. Meat Science, 49(4), 399-409.
368
Cozzolino, D., & Murray, I. (2004). Identification of animal meat muscles by visible
369
and near infrared reflectance spectroscopy. Lebensmittel-Wissenschaft und-
370
Technologie, 37(4), 447-452.
371
372
373
374
Cross, H. R., & Whittaker, A. D. (1992). The role of instrument grading in a beef valuebased marketing system. Journal of Animal Science, 70, 984-989.
Defra
(2005).
Agricultural
http://statistics.defra.gov.uk/esg/quick/agri.asp.
16
Quick
Statistics.
375
Dewar-Durie, A. (2001). The Scottish Sheep Industry: a Way forward. Report to the
376
Minister of Rural Affairs, Ross Finnie, http://www.scotland.gov.uk/library3/agri/tssi-
377
01.asp.
378
Downey, G., & Hildrum, K. I. (2004). Analysis of meat. In C. A. Roberts, J. Workman
379
Jr., & J. B. Reeves III, Near Infrared Spectroscopy in Agriculture, Agronomy No. 44.
380
(pp. 599-632). Madison, WI, USA: American Society of Agronomy, Inc., Crop
381
Science Society of America, Inc., Soil Science Society of America, Inc.
382
Dundon, P., Sundstrom, B., & Gaden, R. (2000). Producing and Processing Quality
383
Beef from Australian Cattle Herds: Industry Outcomes. Cooperative Research Centre
384
for the Cattle and Beef Industry (Meat Quality).
385
García-Olmo, J., De Pedro, E., Garrido-Varo, A., Jiménez, A., Salas, J. & Santolalla, M.
386
(2000). Fatty acids analysis of Iberian pig fat by near infrared spectroscopy (NIRS).
387
In J.A. Alfonso de Almeida, & J.L. Tirapicos Nunes, Tradition and innovation in
388
Mediterranean pig production. International Symposium on Mediterranean Pig (pp.
389
191-195), 26-28 November 1998, Evora, (Portugal).
390
García-Olmo, J., Garrido-Varo, A., & De Pedro, E. (2001). The transfer of fatty acid
391
calibration equations using four sets of unsealed liquid standardisation samples.
392
Journal of Near Infrared Spectroscopy, 9(1), 49-62.
393
Geesink, G. H., Schreutelkamp, F. H., Frankhuizen, R., Vedder, H. W., Faber, N. M.,
394
Kranen, R. W., & Gerritzen, M. A. (2003). Prediction of pork quality attributes from
395
near infrared reflectance spectra. Meat Science, 65(1), 661-668.
396
Hildrum, K. I., Nielsen, B. N., Mielnik, M., & Næs, T. (1994). Prediction of sensory
397
characteristics of beef by near-infrared spectroscopy. Meat Science, 38(1), 67-80.
398
Hildrum, K. I., Isaksson, T., Næs, T., Nielsen, B. N., Rødbotten, M., & Lea, P. (1995).
399
Near infrared reflectance spectroscopy in the prediction of sensory properties of
400
meat. Journal of Near Infrared Spectroscopy, 3, 81-87.
401
Hopkins, D. L., Safari, E., Thompson, J. M. & Smith, C. R. (2004). Video image
402
analysis in the Australian meat industry - precision and accuracy of predicting lean
403
meat yield in lamb carcasses. Meat Science, 67, 269-274.
17
404
Hoving-Bolink, A. H., Vedder, H. W., Merks, J. W. M., de Klein, W. J. H., Reimert, H.
405
G. M., Frankhuizen R., van den Broek W. H. A. M., & Lambooij, en E. (2005).
406
Perspective of NIRS measurements early post mortem for prediction of pork quality.
407
Meat Science, 69(3), 417-423.
408
Leroy, B., Lambotte, S., Dotreppe, O., Lecocq, H., Istasse, L., & Clinquart, A. (2003).
409
Prediction of technological and organoleptic properties of beef longissimus thoracis
410
from near-infrared reflectance and transmission spectra. Meat Science, 66(1), 45-54.
411
Liu, Y. L., Lyon, B. G., Windham, W. R., Realini, C. E., Pringle, T. D. D., & Duckett,
412
S.K. (2003). Prediction of color, texture, and sensory characteristics of beef steaks by
413
visible and near infrared reflectance spectroscopy. A feasibility study. Meat Science,
414
65(3), 1107-1115.
415
Mitsumoto, M., Maeda, S., Mitsuhashi, T., & Ozawa, S. (1991). Near-infrared
416
spectroscopy determination of physical and chemical characteristics in beef cuts.
417
Journal of Food Science, 56(6), 1493-1496.
418
MLC (2002). Sheep Yearbook 2002.MLC, Milton Keynes, Buckinghamshire.
419
Næs, T., & Hildrum, K. I. (1997). A comparison of multivariate calibration and
420
discriminant analysis for determining tenderness of meat by NIR spectroscopy.
421
Applied Spectroscopy, 51, 350-357.
422
Navajas, E., Glasbey, C. A., McLean, K. A., Fisher, A. V., Charteris, A. J. L., Lambe,
423
N. R., Simm, G., & Bünger, L. (2005). In vivo measurements of muscle volume by
424
automatic image analysis of spiral computed tomography scans. Animal Science, 82,
425
545-553.
426
Park B., Chen Y. R., Hruschka W. R., Shackelford, S. D., & Koohmaraie, M. (1998).
427
Near infrared reflectance analysis for predicting beef longissimus tenderness. Journal
428
of Animal Science, 76, 2115–2120.
429
Park B., Chen Y. R., Hruschka W. R., Shackelford, S. D., & Koohmaraie, M. (2001).
430
Principal components regression of near infrared reflectance spectra for beef
431
tenderness prediction. Transactions of the American Society of Agricultural
432
Engineers, 43(3), 609-615.
18
433
Rødbotten, R., Nielsen, B. N., & Hildrum, K. I. (2000). Prediction of beef quality
434
attributes from early post mortem near infrared reflectance spectra. Food Chemistry,
435
69, 427-436.
436
Rødbotten, R., Mevik, B. H., & Hildrum, K. I. (2001). Prediction and classification of
437
tenderness in beef from non-invasive diode array detected NIR spectra. Journal of
438
Near Infrared Spectroscopy, 9, 199-210.
439
440
SAS, (1989). SAS/STAT User’s Guide, release 6.03 ed. Ed by Statistical Analysis
System Institute Inc. Cary, NC, USA.
441
Schwarze, H., (1996). Report no. 996-10-1. Third European Symposium on Near
442
Infrared (NIR) Spectroscopy: Continuous fat analysis in the meat industry, pp. 43-49.
443
Shackelford, S.D, Wheeler, T.L., & Koohmaraie, M. (2004). Development of optimal
444
protocol for visible and near-infrared reflectance spectroscopic evaluation of meat
445
quality. Meat Science, 68, 371-381.
446
Shenk, J. S., Westerhaus, M. O., & Workman, J. J. (1992). Application of NIR
447
spectroscopy to agricultural products, In D.A. Burns, & E.W. Ciurczak, Handbook of
448
Near Infrared Analysis (pp. 383-431). New York: Marcel Dekker, Inc.
449
Stanford, K., Richmond, R. J., Jones, S. D. M., Robertson, W. M., Prince, M. A., &
450
Gordon, A. J. (1998). Video image analysis for on-line classification of lamb
451
carcasses. Animal Science, 67, 311-316.
452
453
454
455
Swatland, H. J. (1995). On-line evaluation of meat. Pennsylvania: Technomic
Publishing Company.
Varnam, A. H., & Sutherland, J. P. (1995). Meat and Meat Products: Technology,
Chemistry, and Microbiology. London: Chapman & Hall.
456
Vatansever, L., Kurt, E., Enser, M., Nute, G. R., Scollan, N. D., Wood, J. D., &
457
Richardson, R. I. (2000). Shelf life and eating quality of beef form cattle of different
458
breeds given diets differing in n-3 polyunsaturated fatty acid composition. Animal
459
Science, 71, 471-482.
19
460
Venel, C., Mullen, A. M., Downey, G., & Troy, D. J. (2001). Prediction of tenderness
461
and other quality attributes of beef by near infrared reflectance spectroscopy between
462
750 and 1100 nm; further studies. Journal of Near Infrared Spectroscopy, 9, 185-
463
198.
464
465
Warriss, P. D. (2000). Meat Science: An Introductory Text. Wallingford: CABI
Publishing.
466
Wood, J. D., Richardson, R. I., Nute, G. R., Fisher, A. V., Campo, M. M., Kasapidou,
467
E., Sheard, P. R., & Enser, M. (2003). Effects of fatty acids on meat quality: a
468
review. Meat Science, 66, 21-32.
469
20
470
Figure 1. Original spectra as log 1/R (400-1900 nm) corresponding to the average
471
spectra of the 5 most tender and the 5 least tender lamb meat samples
472
Figure 2. Principal components analysis (PCA) of the second derivative spectra (400-
473
1900 nm) of the 50 extreme lamb meat samples (T = Texel; B = Black face)
474
Figure 3. Correlation of the second derivative (2, 12, 2, 2) absorbance data vs "overall
475
liking” corresponding to the 50 extreme lamb meat samples
476
21