Directly assessing the amount and yield of meat in an intact carcass is quite difficult. At present, we often do it indirectly: first estimating the proportion of fat and then, assuming the proportion of bone to be constant, estimating the remainder - lean meat. But how do we estimate the proportion of fat? A variety of methods are in use in various countries, as we will consider in this lecture.

Fat-depth probes

The hand-held fat depth probe has many practical advantages: the technology is relatively simple, therefore it tends to be more reliable and robust relative to some other methods. But it is also a decentralised system and, if any one probe malfunctions, spares may be brought into action while repairs are made. As the tip of a fat-depth probe is pushed into the carcass, a flat plate stays on the surface, enabling the depth of the probe in the carcass to be determined electromechanically. The following terms may be used.

Transect. A cut or line made with a meat probe through the tissues of a carcass.

Way-in. Describing a transect made during the initial penetration of a tissue, as a probe first penetrates the meat.

Way-out. Describing a transect made during the withdrawal of a probe from the meat.

Signal. The output from a probe sensor, such as a photometer, made during way-in or way-out transects.

Depth sensor. A device that measures the depth of the probe in the meat.

Depth vector. The signal from the depth sensor during way-in and way-out transects.

Vector diagram. The signal from the primary probe sensor, such as a photometer, on the y-axis compared with the depth vector on the x-axis.

Cut-off. A point on a vector diagram that represents a change from one condition to another, as when passing from fat to muscle.

Threshold. One or more criteria used to locate a cut-off.

For example, the figure above shows the boundary between the backfat and longissimus dorsi in the thoraco-lumbar region of a pork carcass detected by a typical commercially-available diode probe. The signal from the photometer diode along a way-out transect is plotted against the depth vector. How it is processed in commercial probes is proprietary information but, computationally, we may find the separation between fat and muscle with a simple algorithm, using the depth in the carcass at which the signal passes a cut-off value determined from two signal-dependent scalars, the maximum signal from the fat (Fmax) and the minimum signal from the muscle (Mmin), multiplied by an arbitrary threshold of .5

cut-off = ((Fmax - Mmin) × .5) + Mmin

There is considerable variation in signals from conventional diode probes. Reflectance from fat may be irregular if there are traces of muscle or blood, or translucent seams of soft fat with a low reflectance, whereas reflectance from muscle may be irregular if it is marbled. The reflectance of PSE muscle may be so high that the boundary between fat and muscle may be indistinguishable.

Fat-depth probes have had a dramatic impact on the economics of pork carcass grading when used to reward producers of lean carcasses, hence providing a powerful incentive for the general improvement of pork carcasses. Originally two fat-depth measurements were made with a ruler on Canadian pork carcasses, one over the shoulder and one at the posterior end of the loin in split sides. Thus, it was necessary to specify a slight offset to one side in splitting the carcass so that the fat was not lost by irregularity in splitting right from left sides. Two measurements were an insurance against fundamental differences in adipose tissue development between the shoulder and the ham, but they doubled the time for data collection and introduced an extra summation step in computation. Later, the measurement was reduced to a single point for economic reasons. The system was in place to the general advantage of both producers and meat customers, and created a steady increase in leanness.

In the early 1980s, there was still considerable uncertainty as to which direction in which to proceed, using either invasive optoelectrical probes or non-invasive ultrasonics. Electrical conductivity probes for detecting the fat-muscle boundary date back to the Lean-Meter of Andrews and Whatley in the USA. The Danish KS meter (K for meat and S for fat) detected the fat lean boundary from the difference in conductivity between fat and muscle and was used widely in Danish meat plants, but optical probes soon were developed to do the same thing using light reflectance. Essentially, the depth sensor of the conductivity probe was retained, but the boundary sensor was changed from an impedance measurement to a reflectance measurement made between light-emitting and light-sensitive diodes.

The Danish MFA (Meat Fat Automatic) that followed the KS, but which still used the conductivity principle was competitive with optical probes such as the Ulster probe (developed by the Wolfson Opto-Electronics Unit at Queen's University, Belfast) and the Intrascope (developed for the UK Meat and Livestock Commission). The Ulster probe gave similar results to the Hennessy and Chong fat depth indicator (FDI) from Aukland, New Zealand. To add to the confusion, reasonable results also had been attained with ultrasonics on pork carcasses. However, the Hennessy and Chong FDI of the time was found to be more precise than the ultrasonic Renco Lean-meter (Minneapolis), and this tipped the balance in favor of optical probes.

Subsequent testing showed the Danish version of the optical probe, the Fat-O-Meater (SFK Ltd, Hvidovre, Denmark), gave similar results to those given by the Hennessy and Chong FDI. The maximum prediction of lean meat yield from fat-depth measurements was at a position to one side (6.5 cm) of the midline over the last rib. The results from non-invasive electronic meat measuring equipment of the day (EMME, Phoenix, Arizona), which previously looked promising for live pigs, were rather poor for pork carcasses, although still possibly worthwhile for beef carcasses.

Danish Carcass Classification Centre

The Danish carcass classification centre, first implemented in 1990, is by far the most advanced on-line system for meat evaluation yet developed. The totally automated system evaluates overall carcasses, plus separate evaluations of the ham, loin, belly, and forelimb, at line speeds of 360 to 400 carcasses per hour. Carcasses are branded automatically. When carcasses are moved into the system, their dimensions are measured mechanically (carcass length from gambrel to snout, forelimb position, and height of pubis) to enable correct positioning of a head-holder and the probes. There are nine probes using optical diodes and the depth vector has a resolution of 0.5 mm in a total length of 18 cm. Probes are inserted by pneumatic pistons: two probes into the ham, two into the belly, two into the shoulder and three into the loin. All the probes measure fat thickness, while only the loin probes measure meat thickness as well. If a probe hits a bone, the measurement is repeated at a new site 16 mm below the impact site.

As well as being the most advanced system mechanically, the Danish Carcass Classification Center also is the first to exploit neural networks in the analysis of probe signals. Overall fat depth is taken as a step function between the low reflectance of the probe window outside the carcass and the low reflectance from the muscle below the fat.


One of the major attractions of ultrasonics is of being able to make measurements on live animals so that, if antemortem and postmortem measurements are compatible, then animals may be finished to the point required to reach a certain grade. Ultrasound is sonic energy at frequencies beyond the audible range (> 20 kHz). Sound moves by compression waves which, when they pass from one medium to another, may be reflected or refracted. The amount of energy reflected depends on the difference in density between the two media and on the angle of incidence at which the beam of ultrasound encounters the interface between the media. If the density difference between the media is large, then reflection is high. If the angle of incidence is near to 90°, then reflection is high. Thus, a beam of ultrasound perpendicular to a bone or an air space in the meat is nearly all reflected.

Acoustic impedance is the ratio of ultrasonic pressure to the resulting flux but, since it is a complex ratio containing both resistive and reactive components, a simpler ratio is more useful. The characteristic impedance is the product of the density of a medium and the velocity of ultrasound through it. Energy is absorbed as ultrasound moves through a medium so that the flux is attenuated, thus, the attenuation constant is the relative rate of decrease in ultrasound amplitude in the direction of movement. The relevance to on-line meat evaluation, is that attenuation increases as a power of frequency. Thus, a high frequency has a poor penetration. However, low frequencies have poor resolution of particles within the meat. High frequencies are more liable to be scattered than low frequencies, but this loss of resolution is balanced by the fact that high frequencies are more easily focused than low frequencies. In summary, the choice of frequency (usually from 1 to 15 MHz) is a compromise between penetration and resolution, so that it is difficult or impossible to resolve small particles at great depth in the meat.

Ultrasonic imaging involves pulses of ultrasound generated from a piezoelectric crystal, with reception of the echoes by the same or another crystal. The simplest form of imaging is an A-mode (amplitude mode) display on an oscilloscope with a stationary transducer. The oscilloscope sweep is triggered when the pulse is generated so that the length of time for echoes to return may be read as the equivalent distance on the x-axis of the oscilloscope.

The A-mode may be used to measure the fat content of meat. The velocity of ultrasound in meat decreases when marbling levels are high. Using a non-linear regression model, intramuscular fat content of beef may be predicted with 90% accuracy for fat levels over 8% and with 76% accuracy for under 8% fat. However, at present, the technique requires cubes of meat to be removed from the carcass and accurately located between two aluminum plates, and the direction of the muscle fibres also must be controlled. It may be possible to replace the lower aluminium plate reflector by a natural reflector in the live animal (such as a boundary between fat and lean at a known depth), but whether the method can be made to work without accurate control of the depth of tissue remains to be seen. Another approach, is a Fourier analysis of the frequency spectrum of the A-mode echo, which allows predictions at fat levels below 4%.

Instead of the amplitude of the echo being used to drive the y-axis of the oscilloscope, as in the A-mode, the amplitude may be used to make the oscilloscope spot more intense or bright. Thus, if the spot intensity is reduced to the threshold level expected for an echo, when an echo appears it is seen as a bright spot on the screen, rather than as a peak in the y-axis direction. The B-mode (brightness mode) display commonly used to obtain images of fat depth and rib-eye areas from meat animals uses this same principle, but the transducer is moved in steps synchronized with steps in the y-axis the display device. Thus, the bright spots indicating reflective structures in the carcass build up a two dimensional image which, for pork carcasses, can be analyzed automatically by machine vision to predict lean yield. B-mode measurements of fat depth are fairly reliable for lambs.

B-mode images of beef muscle may contain a speckle pattern related to small inclusions in the muscle, mostly marbling. But direct estimation of marbling is difficult even with the higher resolution of pulse echoes using a coloured-image. Autocorrelation analysis of the speckle pattern may provide predictions of marbling, but results are somewhat variable (from R = 0.82 to 0.21). Autocorrelation may be used to detect cyclic activity in a signal. A copy of the signal is delayed, then multiplied by the undelayed signal. After smoothing with a low-pass filter, meaningful signal peaks from marbling persists, while noise is lost. Another approach is to use video pattern recognition coupled with a neural network.

The Danish Autofom system is a very neat solution to the problem of how to move the ultrasonic system relative to the carcass: instead of moving the transducer in step with the display device to generate an image, as in B-mode ultrasonics, the whole carcass is moved past an array of 16 ultrasound transducers mounted in a U shape to match the profile of the dorsal region of the carcass. With 200 sets of measurements made along the carcass, by pulsing every 0.1 ms then switching to sampling at 5 MHz, this creates a three-dimensional image of A scans. The Autofom may process 900 carcasses per hour and is located immediately after the scalding tank when carcasses have maximum flexibility for being pulled through the transducer U frame. As well as predicting lean yield, the Autofom also provides information on the distribution of meat within the carcass which, when coupled with individual carcass identification encoded into the gambrel, allows carcasses to be sorted for maximum utilization in meat cutting. Other ultrasonic fat-depth systems also are available, such as the CSB Ultra-Meater.


As we saw earlier, fat-depth detection has always been a struggle between two competing technologies - optoelectrical versus ultrasonics. Ultrasonics seems to be winning again, offering a major advantage microbiologically in not being based on invasive probe technology. However, invasive probes are simpler, less expensive and decentralised, and it is possible to obtain fairly reliable information on meat quality while the fat-depth measurement is being made. So, keep your eyes open for further developments in this field!