ONLINE RECORDING OF WAVELENGTH ABSORPTION SPECTRA IN MEAT

FIELD OF THE INVENTION

The invention relates to invasive, optical measurements of meat products and provides both a method and an apparatus for performing such measurements. In particular, the measurements are used to perform classification or grading of the meat product as well as to determine qualitative and/or quantitative parameters of substances in the meat product.

BACKGROUND OF THE INVENTION

Optical methods for characterizing foodstuff products, such as meat products, can be divided into photometric methods detecting light at a single wavelength or using wavelength-independent detection, and spectroscopic methods recording wavelength spectra.

Photometric methods

It is common to these methods that they only obtain an intensity value and no spectral information. A typical objective of these methods is therefore to estimate a geometrical parameter that could otherwise be determined by visual inspection, such as the position of the fat/lean meat interface or the ratio of fat/lean meat. For this reason, they are often correlated with a measurement of insertion depth.

Some of these methods may apply broad-spectred light, but detects transmitted/reflected light in a process that does not capture the spectral information, i.e. wavelength- independent detection.

The first probes for performing invasive Near-Infrared Reflection (NIR) measurement in meat products, especially carcasses, date back to the late 1970's. In US 4,078,313, Hennessy discloses a measuring device with an invasive probe having a light source and a light sensitive diode therein, so that light reflected from the surroundings of the probe can be detected by the light sensitive diode. A depth gauge determines the depth of the probe in, for example, animal fat and in particular the point at which the probe passes from fat to lean meat can be determined.

A number of later references disclose similar NIR products, among them GB 2 179 443, WO 80/01205, US 4,352,245, US 4,270,274, US 4,825,711, EP 668 999, WO 92/21025 and US 6,859,282. All of these apply NIR measurement using a depth gauge and an

invasive probe for emitting light and for receiving reflected light, in-order to determine the fat/lean meat interface or ratio in the meat product.

Another example is JP 02016434, which discloses a device for determining quality of foodstuff products such as fruits or vegetables. Here, two optical fibre tips (41a and 51a) are inserted to a position determined by the stop 42. The intensity of the transmitted/reflected light is recorded by a photomultiplier tube 61.

Spectroscopic methods

Spectroscopic data contains a lot of information about the chemical composition of the measured portion. The described methods record spectra of a single portion of the meat or homogenized samples.

EP 402 877 and WO 99/19727 (Slagteriernes Forskningsinstitut) disclose invasive methods for NIR determining pigment and drip loss, respectively. According to the methods, a probe is inserted to a fixed position where a reflection spectrum is recorded. The recorded spectrum can be compared with reference spectra to yield a concentration of a selected substance or a quality of the meat. These methods, relying on reflection spectra, have never been used in slaughterhouses due to lack of consistency and precision in the determined values.

US 6,563,580 (also published as WO 00/02043) describes a method for determining tenderness of meat by combining data such as race, age weight, pH, and colour with an optical spectrum recorded by an invasive probe. The optical spectrum may be a reflection spectrum or a transmission spectrum.

When determining parameters such as water holding capacity or contents of specific substances by optical methods, it is customary to perform Near-Infrared Transmission (NIT) spectroscopy on a sample extracted from the carcass. Extracted samples are sent to a central laboratory where they are liquefied whereafter NIT spectra are recorded using highly specialized equipment. WO 01/23868 describes NIT spectra recorded on extracted, liquefied samples.

With the present state of invasive probes, only NIR measurements are performed online and at the site of the carcass. NIR can give estimates of e.g. fat and muscle thickness at selected locations and back fat thickness, where the only requirement is for the probe to be able to determine the depth of transition between various layers. It is a disadvantage that a detailed analysis of e.g. content of specific substances in the meat product is not possible with the presently available invasive probes.

It is of interest to obtain a detailed analysis of meat products, relating to both qualitative and/or quantitative parameters and to content of specific substances. Knowing specifics of the chemical composition of the different layers in meat cuts provides key optimizing factors in the production chain of an abattoir. The quantitative and/or qualitative online description of the different layers can provide input for production sorting algorithms that will heighten the overall yield of meat cut productions.

In order to obtain detailed knowledge of qualitative and/or quantitative parameters of or content of specific substances in the meat product, samples need to be extracted and transported to centralized laboratories where NIT spectra can be recorded on liquefied samples. It is a disadvantage that the extraction of a sample destroys at least part of a carcass. It is another disadvantage that the samples are removed from the conveyor line to be analysed at another location. It is a further disadvantage that samples are liquefied before measurement, as this sample preparation complicates the process and information related to specific layers or regions is thereby lost.

It is another disadvantage that the result from a centralized laboratory only is available so late in the process that is not possible to sort products using the determined parameters, e.g. when the meat product has been dispatched from the plant.

Thus, there is a need for a method and apparatus for online recording of detailed spectroscopic data for different interior portions of a meat product.

SUMMARY OF THE INVENTION

It is an object of the present invention to provide a method and an apparatus for invasive NIT spectroscopy based determination of parameters or substances in meat products, where the NIT spectra are recorded in-situ/online, i.e. on unprepared samples remaining inside the meat products.

It is another object of the present invention to provide a method and an apparatus for recording of wavelength absorption spectra in different portions of a meat product.

It is still another object of the present invention to provide a method and an apparatus for online grading or classification of meat products based on wavelength absorption spectra.

Wavelength absorption spectra may be recorded by NIT spectroscopy. In the present context, a meat product may be any cut or whole carcass from any animal such as mammals, birds or fish, in particular pigs/hogs, cattle or sheep.

To meet the above objects, the invention provides an apparatus comprising a spectrometer and a probe for recording invasive NIT spectra of an interior portion of a meat product without removing the interior portion from the product, and a data processing unit for determining a parameter or a content of the substance in the interior portion by modelling recorded NIT spectra with NIT spectra recorded on meat portions with known parameters or contents of the substance.

In a preferred embodiment, the invention provides an apparatus for in-situ recordings of wavelength absorption spectra of meat products, the apparatus comprising • a cutting probe comprising a point for penetrating a surface of the meat product, light guiding means and light collecting means positioned in an optical path of light to be transmitted from the light guiding means, the probe being shaped so that insertion of the probe into a meat product will position an interior portion of the meat product in said optical path between the light guiding means and the light collecting means and so that changing an insertion depth of the probe into the meat product will change the interior portion in said optical path,

• a light source for providing infrared light of multiple wavelengths to the light guiding means,

• a detecting unit for receiving light collected by the light collecting means and for detecting and recording wavelength absorption spectra of the received light at different consecutive positions of the probe,

• means for determining, during insertion or extraction of the probe, a plurality of consecutive insertion depths, or changes in insertion depth, of the cutting probe in the product, and • means configured to correlate consecutive determined insertion depths or changes in insertion depth with corresponding consecutive recorded wavelength absorption spectra recorded by the detecting unit.

When receiving light from the light source, the light guiding means transmits the light from the probe at a position and in a direction so that the light propagates along the optical path intersecting the light collecting means. The light collecting means is formed to collect at least light propagating along the optical path and guide it towards the detecting unit. The light collecting means is preferably positioned opposite to the light guiding means in the probe. Both the light guiding and collecting means may comprise few or many components such as lenses, collimators, mirrors, optical fibres, prisms, windows, gratings, holograms, splitters, couplers, etc. The absorption spectrum are recorded by recording the intensity of the transmitted light at different wavelengths; in the present context,

absorption spectra and transmission spectra are therefore based on the same recorded intensities and may be used interchangeably. •

As specified, the invention records consecutive spectra of different portions of the substance during insertion or extraction. Therefore, the apparatus must be able to determine a plurality of consecutive positions during each insertion or extraction, and not simply one position for each insertion. As a result, a series of consecutive spectra from consecutive positions are recorded, which provide a representative set of spectra for analysis of the product, regardless of whether the object of the analysis is to obtain average values or to identify variations to be analyses separately. In a preferred implementation, as many as 100 sets of correlated spectra and positions are recorded per centimetre during extraction of the probe, resulting in up to 1,000 data sets for typical insertion depths in carcasses. The preferred sampling rate (number of sets recorded per centimetre) depends on the nature of the meat product and the object of the analysis. For inhomogeneous products whose composition varies over a scale of less than one centimetre, a large sampling rate is required in order to obtain statistically representative data for each sub-segment in the product. In a preferred embodiment, the number of sets of correlated spectra and positions recorded are larger than 5/cm, larger than 10/cm or larger than 25/cm, or preferably larger than 50/cm.

One possible analysis would be to determine the meat product composition at different positions (depths) in the product, or average over portions of similar type (e.g. fat, lean meat etc.). By determining the insertion depth, or a similar indication of position, of the cutting probe and correlating the insertion depth with a recorded wavelength absorption spectrum, each recorded light intensity value may be related to two variables; a wavelength (or frequency) and an insertion depth. The recorded spectra may be presented in a three-dimensional coordinate system with the absorption or transmission shown as a function of wavelength and insertion depth. Meat products are formed by tissue layers such as skin, fat and muscle layers having a given thickness, and the recording of several spectra within each layer can be used to segment the recorded spectra corresponding to each layer. The repeated measurements of absorption spectra within the segment thus provide a good statistical basis for prediction models for a particular layer or segment.

Determining an insertion depth preferably means determining a distance from an exterior surface of the meat product to the optical path between the light guiding means and the light collecting means. The determined number representing the insertion depth may be a number or a sequence being indicative of a depth. In one embodiment, the means record a change in an external parameter in comparison to a clock signal whereby a position can be calculated for a given time indication. In another embodiment, determining an insertion

depth may include determining a distance from an exterior fixed point of reference to the optical path. The fact that each wavelength absorption spectrum is associated with a specific portion via the insertion depth is a novel and unique feature of the present invention.

The light source preferably has an emission spectrum ranging from 800 nm to 3,000 nm, 1,000 nm - 2,350 nm or at least a spectrum having a substantially uniform intensity in the intervals 1,000-1,150 nm, 1,650-1,750 nm, 1,900-1,980 nm, and 2,250-2,350 nm. These intervals are relevant when determining parameters related to the protein, water and fat content in the product. Consequently, the detecting unit can detect and record light in the same range and intervals.

In order to determine qualitative and/or quantitative parameters of the meat product, the apparatus may further comprise a data processing unit for receiving recorded wavelength absorption spectra and being configured to determine such parameters of the meat product, the data processing unit comprising a memory for holding a database of wavelength absorption spectra previously recorded on meat portions with known parameters, and means for comparing a recorded wavelength absorption spectrum with wavelength absorption spectra of the database to identify at least approximate parameters of the corresponding interior portion of the meat product. Similar means may be provided to determine a content of a given substance.

The data processing unit is preferably configured to merge or sort the spectra into segments according to each parameter to be determined, so that the means for comparing identifies parameters of individual segments. For instance, to determine fatty acid composition in the fat tissue, all absorbance spectra from the fat tissue are merged into one segment and all spectra from lean meat are merged into another segment. In another example, a segment may be a depth interval or a layer in the meat product.

By the aid of the internal database, which has previously been calibrated against laboratory data, the actual qualitative and/or quantitative parameter can be predicted using e.g. a PLS or parafac model. The data processing unit preferably contains one database for each qualitative and/or quantitative parameter to be determined.

By configuring the data processing unit, qualitative and/or quantitative parameters as different as

- fatty acid composition expressed as the relative amounts of the monitored fatty acids (e.g. saturated fatty acids, mono unsaturated fatty acids and poly unsaturated fatty acids),

- fatty acid content (expressed as weight or volume percent) of one or more specific fat acids as selected by the user or the accumulated content of a group of fatty acids as selected by the user, such as all of the known fatty acids,

- moisture content, - water holding capacity,

- intermuscular fat,

- intramuscular fat (i.e. fat between muscle fibres or marbling),

- colour,

- chemical residues from medicine and growth promoters, - the lean meat percentage,

- fat and muscle thickness at selected locations,

- back fat thickness,

- the commercial value of each product,

- weight of saleable meat on specific commercial cuts - etc. may be determined for portions, segments or layers of the product or for the entire product. The closest match may be performed using only a given wavelength range or ranges of the spectrum, which range(s) is (are) characteristic for the sought-after parameter. Thereby, the closest match for a recorded spectrum need not be a single spectrum from the database, a recorded spectrum may have different closest spectra depending on the parameter in question.

Previously, it has only been possible to determine the above parameters from liquefied samples in a laboratory, as the invasive probes of the prior art are not capable of recording NIT spectra and determining a parameter or a substance content. The invention presents a breakthrough within the field of slaughtering technology in that it may determine such parameters from in-situ recording where no sample need to be removed from the product. That the absorption spectrum is recorded in-situ means that the measurement of the absorption spectrum is recorded on meat portions residing inside the meat product during the recording. Hence, no sample of the meat product is extracted and the entire process is carried out at the site of the meat product, e.g. conveyor lines at slaughtering plants.

Traditionally, systems for determining qualitative and/or quantitative parameters for products on a conveyor line can be characterised as: - inline/online meaning that the measurements is performed on or in the product (i.e. in-situ, without removing a sample) and that the parameter is available immediately or shortly after the measurement;

- at-line meaning that a sample is extracted from the product and analysed at or near the product line, the parameter is available when the sampled product has moved further down the product line; or

- offline meaning that a sample is extracted from the product and analysed at another location, e.g. a centralized laboratory. The parameter is first available at a later time.

Previously, the grading has been offline and based on absorption spectra recorded in laboratories using samples from a limited number of products at the plant. The resulting grading is typically not associated with a specific meat product, but rather with a batch of products e.g. from a specific livestock. In a preferred embodiment, the apparatus is interfaced with a data-merge system and configured to perform online absorption spectra- based grading or classification of meat products. Here, the grading is available immediately after extraction of the probe. The online grading allows for immediate grading of the meat product by sorting of products or by marking product using e.g. tag encoding, stamping or associating the grading with the meat product in the plant automation system.

Further, the apparatus may be configured to perform grading of the meat product based on the absorption spectra recorded in-situ. Grading (or classification) is used to determine settling price to suppliers, sorting in product categories etc. and may be based on any parameter indicating a relevant product quality, such as any one or more of the previously mentioned.

As described previously, prior art NIT spectroscopy on meat products has been carried out using liquefied or homogenised samples extracted from the meat products. Thereby, the result did not give any indication of the monitored substances' position or distribution in the meat product - in fat layers, muscles, gristle, tendons etc. In the apparatus of the present invention, the processor may be configured to provide the qualitative and/or quantitative parameters as a function of insertion depth. This is advantageous as it provides very valuable information of e.g. the distribution of substances in the meat product. Some substances have a tendency to accumulate in certain types of tissue, which makes it difficult to determine quantitative values from liquefied samples. Also, the invention provides for a much better determination of the meat quality as the parameters can be determined for specific portions of the meat, and not as an average of an entire sample.

In order to determine transmission or absorption spectra, the apparatus may need to record reference spectra at regular intervals. Also, calibration spectra may be recorded for checking or adjusting the accuracy of the apparatus by comparison with a standard

instrument. For these purposes, the detecting unit of the apparatus may be configured to record a spectrum of infrared light received from the light collecting means at user determined intervals when the means for determining an insertion depth indicates that there is no interior portion of a meat product in said optical path between the light guiding means and the light collecting means. Such recorded spectrum may be used for reference or calibration purposes.

In a preferred design, a first surface part of the cutting probe is an optical window through which light from the light guiding means will exit the probe, and an opposite, second surface part of the probe is an optical window through which light from the light guiding means will enter the probe to be received by the light receiving means.

It is considered favorable if insertion of the probe does not detract from the value of the meat product by destroying parts of the product or rendering it unusable in other ways. Cutting in the product cannot be avoided, but it is an object to perform a clean and harmless cut. Therefore, the insertion points of the probe are equipped with sharp knives and the windows preferably have a close fit to the surrounding surface parts so as to form a smooth surface.

To determine quantitative parameters from the recorded spectra, it is preferable that there is an at least substantially constant distance between the windows. For this reason, the cutting probe is preferably formed in a material composition and dimensioned to ensure a constant distance between the light guiding means and the light collecting means during insertion in the meat product.

The dimensioning refers to the thickness of the material used, the length of the probe and the specific shape of the probe. In a preferred embodiment, the cutting probe comprises a first knife holding the light guiding means and a second knife holding the light collecting means, the first and second knifes extending in parallel from a common base and each having a point for penetrating the surface of the meat product and a first cutting edge starting at the point. In the case of two parallel extending knifes, the dimensioning refers to the circumference, shape and length of each knife. Typically, the probe is formed in a metal alloy such as stainless steel and the knifes are bars with interior lumens for holding optical fibres. In one embodiment, the probe has cross-bars between the knifes to ensure that the knifes are not deflected during insertion. In these and similar cases, the shape and size of cross-bars and thickness of the lumen walls are also part of the relevant dimensioning.

To determine quantitative parameters from the recorded spectra, in is further preferable that the optical path in the meat product is of constant length. For this reason, it is an object that the meat product, upon insertion of the probe, abuts the windows closely and without air pockets between the windows and the product. This may be facilitated by designing the probe with smooth surfaces without prominent concave features.

To determine quantitative parameters from the recorded spectra, it is also preferable that the meat product, upon insertion of the probe, abuts the probe in an at least substantially uncompressed and un-stretched state.

When a cutting edge cuts through a material, material on each side of the cut will be pushed away by the sides of the edge and thereby be slightly compressed. As mentioned in the above, the measured meat portion should preferably be uncompressed since such compression affects the measurements. Also, the compression gives a lateral force on the knife which may bend or deflect the knife leading to a change in the distance between the knifes. For this reason, it is preferred that the cutting edges of the first and second knifes can be projected onto the same straight line, meaning that they, when seen from the direction of the points, will extent along the line intersecting the two points. Thereby, the meat portion measured upon will remain uncompressed since it lies in line with the cutting edges. Also, the transverse forces on the knifes are perpendicular to the direction between the knifes and therefore less likely to cause significant changes in the distance between the knifes.

In another preferred embodiment, the cutting edges will, when seen from the direction of the points, have an angle in the interval 0-40°, such as in the interval 10-20°.

The invention further provides a method for recording position dependent wavelength absorption spectra in meat products, the method comprising the steps of: » providing a cutting probe comprising a point for penetrating a surface of the meat product, light guiding means, and light collecting means positioned in an optical path of light transmitted by the light guiding means,

■ inserting the probe into a meat product to position an interior portion of the product in said optical path between the light guiding means and the light collecting means,

■ recording a series of consecutive spectra at consecutive positions in the product by performing the following steps a multitude of times in a consecutive manner: determining an insertion depth of the probe into the product;

- recording a wavelength absorption spectra of light transmitted from the light guiding means through an interior portion of the product currently in the optical path to the light collecting means;

- changing an insertion depth of the probe in the product and thereby changing the interior portion of the product in said optical path.

The penetration of skin and tissue often requires the application of some force. In order to avoid undue compression of the meat portions during recording, the step of changing an insertion depth preferably reduces the insertion depth so that spectra are recorded during extraction of the probe from the meat product.

The considerations and preferred elements presented in relation to the apparatus and probe in the previous sections are also, where applicable, valid for the method described here and vice versa.

The method preferably further comprises comparing recorded wavelength absorption spectra with wavelength absorption spectra from an internal database, which has previously been calibrated against laboratory data relating to parameters of interest. Thereby, corresponding parameters of the portion(s) used in the recording can be determined, e.g. by chemometrical methods such as multivariate analysis.

In order to provide a statistical foundation, the method may also involve pre-processing of recorded spectra to sort the spectra into segments which are relevant to the parameters to be determined. Spectra in each segment may then be merged, or mean values may be formed, so that spectra in a segment are used to represent repeated recordings in the corresponding segments. Preferably, each segment corresponds to a tissue type or a depth interval. As a result, qualitative and/or quantitative parameters may be determined for individual segments instead of for individual portions. This provides a determination of more precise and/or representative parameters for characterising different parts or tissue types of the meat product. The qualitative and/or quantitative parameters from individual segments may then be used to determine an overall grading of the meat product.

The basic idea of the invention is to record position-specific NIT spectra in order to be able to characterize inner portions of solid or firm meat products without removing a sample from the product. It is an advantage that the invention can be applied to inhomogeneous products as well, since the position correlated to each spectrum allows for extraction of both average values for larger regions as well as specific values characteristic for smaller individual portions. Also, it is the principle of the invention that the recording and

processing of data are performed online and in-situ, thereby allowing for an online grading of the products.

It is another advantage that the invention allows for all products in a fabrication line to be characterized, instead of only performing random checks. Since no sample is removed from the product, the recording of the absorption spectra has not damaged the product or its value. Thereby, all products in a fabrication line can be measured upon without any decrease in value.

These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter.

BRIEF DESCRIPTION OF THE FIGURES

Figure 1 is a schematic drawing of a probe according to an embodiment of the invention.

Figure 2 is a cross-sectional view of a knife according to an embodiment of the invention.

Figure 3 is a schematic drawing of a probe according to an embodiment of the invention.

Figure 4 is a cross-sectional view of an apparatus according to an embodiment of the invention.

Figure 5 is a three-dimensional graph showing wavelength absorption spectra recorded at multiple depths.

Figure 6 is a graph showing absorption as a function of depth for a specific wavelength.

Figure 7 is a chart illustrating the data processing in the data processing unit.

Figure 8 is a graph illustrating chemometrical determination of volumetric concentration of fish oil in an olive oil based on NIT measurements.

The figures are not to scale.

DETAILED DESCRIPTION OF THE INVENTION

In this section, the apparatus according to the invention, and the use of the apparatus and the data processing according to the method of the invention, will be described in detail.

In the following, a preferred embodiment of the invention will be described in relation to Figures 1-3. In Figure 1, an invasive probe 1 of an apparatus 2 for in-situ recordings of wavelength absorption spectra is shown. The probe has two knifes 4 and 5 expending in parallel from casing 3, each knife having a window 8 and 9 through which light can be transmitted or received. Each knife also has a tip 10 with a point 13 and a cutting edge 12. The tip can be removed so that the tip and cutting edge can be taken off for sharpening or replacement. The sides of the knifes facing each other are straight and parallel so that a distance 16 between the points 13 is equal to a distance 17 between the windows 8 and 9. Also shown in Figure 1 are the means 14 and 15 for determining insertion depth, which will be described in larger detail later.

When introducing probe 1 into a meat product, one knife may be deflected in relation to the other if it strikes a hard part inside the product. This can lead to a misalignment of windows 8 and 9 whereby less transmitted light will be collected and the recording will be misleading. In order to reduce the risk of deflecting a knife, a cross-bar 18 with cutting edge 19 can be mounted between knifes 4 and 5.

Figure 2 is a cross-sectional view showing the inside of knife 5 (or 4) consisting of a hollow metal rod 20 holding light collecting (or light guiding) means comprising transparent window 9 (8); a mirror 23 formed by a coated surface of metal or glass rod 22; collimator 26 and optical fibre 28 (or 29).

An alternative probe 31 is shown for apparatus 30 in Figure 3. Here, the probe is in one piece with the form of a half-pipe, and having one point 12 and two cutting edges 13. Probe 31 has oppositely positioned windows 8 and 9 through which light is transmitted and collected. An inner diameter 34 of the half-pipe determines the distance between windows 8 and 9 and thereby the optical path of light in a meat portion. The light guiding and light collecting means of probe 30 can be formed in relation to windows 8 and 9 by optical components similar to those shown for probe 1 in Figure 2.

Since probe 31 is formed in one piece, the risk of misaligning windows 8 and 9 by deflecting a knife is removed. In preferred embodiments, both probes 1 and 30 are designed to be waterproof and easy to sterilize and have smooth outer surfaces.

Apparatus 2 applies mechanical means for determining insertion depth, parts of which are shown in Figure 1. Here, ring 14 abuts the surface of the meat product upon insertion of probe 1, and rods 15 are pushed into the apparatus 2. By determining the movement of rods 15, the apparatus can determine an insertion depth of the probe. Any type of commercially available linear position sensor may be used for this purpose.

The apparatus 30 shown in Figure 3 applies optical or sonic means for determining insertion depth. Here, an emitter 36 can emit an optical or a sonic signal which is reflected by the surface of the meat product. The reflected signal is received by receiver 37, and a distance and/or a change in distance can be calculated. Products for optical, sonic or other 5 wireless distance measuring, e.g. infrared distance measuring sensors, are commercially available.

Figure 4 show further details of the apparatus 2 of Figure 1. In the following, the optical path 43 of light in the apparatus 2 will be illustrated using references to details in the knife

10 shown in Figure 2. The optical path in probe 30 of Figure 3 is similar. The apparatus 2 in Figure 4 holds a light source 44 for providing infrared light of multiple wavelengths to the light guiding means in knife 4. The light source can be a Tungsten lamp emitting a broad thermal spectrum including the wavelength range 1,000 - 3,000 nm. Broadband optical fibre 29 receives and transmits light 43 from the light source to collimator 26 which forms

15 directional light 43 deflected by mirror 23 and transmitted by window 8.

Light 43 transmitted by matter interspacing windows 8 and 9 will be collected by the light collecting means in knife 5. Light transmitted by window 9 and incident on mirror 23 will be received by collimator 26 which couple the light 43 to broadband optical fibre 28. Fibre 20 28 transmits light to a detecting unit 45 for detecting and recording wavelength absorption spectra. The detecting unit may be a spectroscope recording light intensity as a function of wavelength (or frequency), e.g. SM 301 PbS spectrometer from Spectral Products Inc., Putnam, Ct-06260, USA.

25 In a preferred embodiment, the means for determining insertion depth comprise step motor 46 operating on the rods 15 and ring 14 which acts as a support for abutting an exterior portion of the meat product. The probe is first inserted into the meat product, either manually or by a machine, whereafter the step motor performs a stepwise extracting by pushing ring 14 away from the housing 3. This provides a very precise

30 extraction of the probe, ensuring that spectra can be recorded at regular intervals. Some slack in the extraction may be experienced due to elasticity in the meat, but this may be accounted for by disqualifying the first one or more centimetres (depending on the scale of the meat product).

35 In an alternative embodiment, the probe is inserted and extracted either manually or by a machine which is not connected to the means for determining insertion depth. In this case, box 46 represents a mechanical sensor for detecting the movement of rods 15, which

provides a signal to a depth calculation unit 47 for determining the depth or a change in depth.

The spectra from spectroscope 45 and depths from depth calculation unit 47 are collected and correlated by means for correlating data in a data processing unit 48. The data processing unit includes a processor and memory 50 holding software to be executed by the processor and databases with previously recorded spectra. In another embodiment, the data processing unit 48 and memory 50 are located external to the casing 3. In this embodiment, the data processing unit 48 may e.g. be a personal computer connected to the casing 3 by a cable 49.

, The means for correlating insertion depths with recorded spectra depend on the embodiment of the means for determining insertion depth. In the preferred embodiment applying a step motor, the correlating means, typically implemented as software or an ASIC, provide trigger pulses to both the spectrometer and the step motor, to ensure recording of a wavelength absorption spectrum for every extraction step of the cutting probe. Numbering of the spectra and knowledge of the step size then provide the correlation between depth and spectra. In the alternative embodiment without step motor, the correlating means can e.g. be based on time stamps on recorded spectra and determined depths.

The data processing in a preferred embodiment of the data processing unit 48 will be described in the following.

Figure 5 is a three-dimensional graph 55 showing absorption spectra covering wavelength ranges from 800 nm to 3,000 nm taken at different depths in, and therefore for different tissues of, a meat product. This is a presentation of the data received by the data processing unit 48 from the spectroscope 45 and the means 47 for determining insertion depth.

The data shown in graph 55 can be pre-processed before comparison with spectra from databases in the memory 50 of the data processing unit. The pre-processing of data consists of sorting the spectra into segments so that a batch of spectra in a segment can be used to extract relevant parameters. The pre-processing and sorting of spectra into segments is illustrated in Figure 7.

One example of this segmentation is described in relation to Figure 6. Here, a graph 60 illustrates absorption A (arbitrary units) as a function of depth in millimeters for fixed

wavelength λ. This corresponds to a slice parallel to the depth axis in the graph 55 of Figure 5. Graph 60 can be divided into segments Si-S ₄ according to the absorption at wavelength λ _n . In the given example, segments correspond to different types of tissue in a pig carcass: Si: skin tissue

S ₂ : fat tissue

S ₃ : meat tissue

S ₄ : air, behind ribs

By analyzing graph 60, the data processing unit 48 can determine insertion depth intervals corresponding to these segments. The final insertion depth intervals may be determined using average values from a number of graphs similar to graph 60 but for different wavelengths. Also, some insertion depth intervals may be better determined using some wavelength ranges. E.g. the start and end of the meat tissue segment S ₃ may be determined using wavelengths in the interval 1,650-1,750 nm which is characteristic for protein. Similarly, the start and end of the fat tissue segment S ₂ can be determined using wavelengths in the interval 2,250-2,350 nm which is characteristic for fat. As a consequence, adjacent segments need not have common boundaries as shown in graph 60. Preferably, spectra recorded in the transition regions between segments are not used as they contribute with noise in the analysis.

In another example, segments may be defined using the absorption level in a predefined wavelength range. If for example the carcass of graph 60 had more than one fat layer, all spectra recorded in fat layers could be identified by their absorption level in the interval 2,250-2,350 nm. Thereby, all spectra recorded in fat layers could be grouped in one segment regardless of their position in depth.

The processing of spectra in segments may be used to extract a large variety of data. In a simple example, the segments are used to determine values for fat and meat thicknesses and percentages. Since the depth for each spectrum is known, each spectrum corresponds to a portion having a given thickness. Hence, using the number of spectra in a segment, the thickness of the corresponding layer can be determined.

In another example, spectra in one segment can be merged into one average spectrum to yield one parameter value for each segment, this is illustrated in Figure 7. Practically, the absorption at a number of specific wavelengths is read from all the spectra in a segment, and average absorption values at these wavelengths are determined. The average absorption values can then be compared to value at the same wavelength in spectra in a database of previously recorded spectra which has been calibrated against laboratory data,

so that a parameter can be determined. In a preferred embodiment, a parafac model is used on the spectra of a segment to determine the concentration of a given substance (e.g. monosaturated fatty acids) in the meat portions used to record these spectra. The database and the parafac model can be specific to the segments and thereby depend on the type of parameter which is sought for using the segment.

The following example illustrates quantitative determination (w/w) of the fatty acid - Linoleic acid - in the fat tissue segment S2. Firstly, an average spectrum for segment S2 is formed by adding absorbance values for each wavelength from the multitude of spectra in the segment (here layer).

I ₁ = (4' + 4 ² + ...+ A;' )IN

Wherein A" _n is the absorbance at wavelength m in the /7'th spectrum of the fat tissue segment S2, and N is the number of spectra in the segment. The average spectrum in segment S2 is then formed by the series of average absorbance values A ₁ , A ₁ ,..., A ₁ .

The content Y of Linoleic acid in (w/w) in the fat tissue segment S2 may then be calculated from

Y = Jc ₀ + Jc ₀ - A, + k ₂ - A ₂ + ... + +Jc _n • !„, , where k ₀ , k _1; k ₂ , ..., k _n are constants previously determined by a regression model, like partial least square (PLS) and/or principal component analysis (PCA), between NIT spectra in the fatty tissue S2 and laboratory analysis data of Linoleic acid. The regression model is based upon a given number of samples e.g. 100 samples.

The parameter values calculated for the different segments as well as values calculated from different processing of the spectra can be used as input to a model for predicting an overall grading or classification of the meat product. A multivariate model can be used if a linear relationship between the parameter values exists. In cases where the relationship between the absorbance spectra and the chemical analysis data is not linear, the regression models partial least square (PLS) and principal component analysis (PCA) can be combined with neural networks in order to compensate for the non-linear relationship.

The gradings obtained using the method and apparatus according to the invention are more precise and characterizing for the meat products than gradings obtained with prior art apparatus. The grading according to the invention is based on much more detailed knowledge of the composition of the meat product, and can take into account both

concentrations of specific substances (e.g. certain fatty acids) in certain tissue types as well as overall properties (e.g. water holding capacity).

An example the ability of NIT measurements for characterizing fatty acid composition is 5 described in the following in relation to Figure 8. The example measures volumetric concentration of fish oil in an olive oil mixture. The two types of oils are used since they each contain very different concentrations of different types of fatty acids.

Samples are prepared with known concentrations of fish oil in olive oil, 0%, 10%, 30%, 10 50%, 70%, 90% and 100% are used. NIT is performed on each sample in the area 1,100 - 2,200 nm, where several recordings are made on each sample to check for inconsistencies in the measurements. The spectral data are corrected against dark and full-scale references. A PLS model is used to predict the volumetric concentration, using a subset of the measured spectral wavelengths. 15

A new set of samples is prepared, this time 0%, 20%, 40%, 50%, 60% and 80% and 100% concentrations. Spectrographic data from NIT measurements on these samples are fed to the above model, and measured concentrations are predicted. The results are presented in the graph of Figure 8, and expresses a R=O.99 with a RMSEP of 3.2 VoI % on 20 the samples unknown to the model.

This example demonstrates the strength of NIT measurements combined with chemometrical analysis, and shows that a detailed knowledge of the fatty acid composition at each position in the meat product can be obtained using the consecutive NIT spectra 25 according to the present invention.