Apparatus for inspecting meat

Abstract

An apparatus 100 for inspecting meat 102 comprises a bioimpedance sensor 103 to measure a frequency-dependent electrical impedance response of the meat and to determine a value of an impedance propert

Description

TECHNICAL FIELD This invention relates to an apparatus for inspecting meat and methods of operating the same. BACKGROUND In the meat production industry, defects in meat associated with post-slaughter changes within the meat, including biochemical and physical processes such as metabolic processes, protein denaturation, and crystallisation, can result in undesirable characteristics, such as unacceptable appearance or structural disintegration, that can lead to significant economic losses for meat producers. Meat quality issues are particularly problematic for producers of processed (e.g., cooked or cured) meat products, where expensive processing steps may be performed on meat with defects. If not detected and sorted, such poor quality meat can result in an end product that cannot be sold. To reduce the risk of losses, monitoring of meat defects may be performed prior to curing or cooking, allowing meat containing defects or abnormalities to be identified at an early stage. Such meat may then be discarded, directed to other product streams or priced down accordingly. However, traditional, instrument-based quality assessments can be unreliable, while visual quality grading is often inconsistent and, due to subjective bias, can be highly variable among producers. An improved approach is therefore desired. The present invention therefore seeks to provide a novel apparatus and methods for identifying the presence of defects in meat, e.g., related to structural disintegration. SUMMARY OF THE INVENTION From a first aspect, the invention provides an apparatus for inspecting meat, the apparatus comprising: a bioimpedance sensor configured to measure a frequency-dependent electrical impedance response of the meat and to determine a value of an impedance property of the meat from the measured frequency-dependent electrical impedance response; an optical sensor configured to receive light from the meat and to determine a value of an optical property of the meat from the light; and a processing system configured to process the value of the impedance property in combination with the value of the optical property, optionally in combination with respective values of one or more further properties of the meat determined by the apparatus, to calculate a defect score representative of a degree of one or more defects in the meat. From a second aspect, the invention provides a method of inspecting meat, the method comprising: determining a value of an impedance property of the meat from a measured frequency-dependent electrical impedance response of the meat; determining a value of an optical property of the meat from light received from the meat; and processing the value of the impedance property and the value of the optical property, optionally in combination with respective values of one or more further properties of the meat, to calculate a defect score representative of a degree of one or more defects in the meat. From a third aspect, the invention provides computer software (and a non-transitory storage medium carrying the same) comprising instructions which, when executed by a processing system, cause the processing system to: receive or calculate a value of an impedance property of the meat determined from a measured frequency-dependent electrical impedance response of the meat; receive or calculate a value of an optical property of the meat determined from light received from the meat; and process the value of the impedance property and the value of the optical property, optionally in combination with respective values of one or more further properties of the meat, to calculate a defect score representative of a degree of one or more defects in the meat. Thus, it will be seen that, in accordance with embodiments of the invention, both an impedance property and an optical property are determined for the meat, and are processed in combination to determine a defect score (i.e. a numerical value) representative of defects in the meat. In some embodiments, these defects may be undesirable post-slaughter changes within the meat. The defect score may, in some examples, be indicative of a degree of a structural defect in the meat. In some embodiments, the defect score is representative of a degree of one or more defects in the meat that are manifest in post-mortem muscle fibres. In some embodiments, the defect score is representative of a degree of one or more defects in the meat associated with one or more post-slaughter changes within the meat. The post-slaughter change may be a biochemical or physical process, such as a metabolic process, protein denaturation, or crystallisation. Determining a defect score by mathematically combining bioimpedance and optical parameters has been found to give an unexpectedly accurate indication of meat defects, which has also been found to be substantially more accurate than considering a bioimpedance property or an optical property alone. Integrating these different sensors and a processing system in the same apparatus allows an objective measurement of a degree of defects in meat to be conveniently determined, without the need for a human expert. It may thus allow for a more accurate estimation of defects to be made than is possible using conventional methods. The apparatus may be used to determine whether defects are present within meat, and to what degree. The one or more defects may include any one or more of: a pale, soft, exudative (PSE) defect, a PSE-like defect, freeze damage, or an anomaly during post-slaughter muscle-meat conversion (e.g., changed rigor strength or dynamics). The meat may be pork, beef, chicken, deer, fish, or another meat. In some embodiments, the apparatus may comprise a pH sensor configured to measure a pH value of the meat. In such embodiments, the processing system may be configured to process the impedance-property value, the optical-property value, and the pH value, in combination, to determine the defect score. By including a value representative of pH, in addition to values representative of optical and bioimpedance properties, when determining the defect score, the accuracy of the defect score (e.g. how well the defect score represents the actual degree of defects in the meat) can be further improved. In some embodiments, the apparatus may additionally comprise a respective sensor for measuring any one or more of: salinity, fat content, protein content, muscle structure, or muscle destructuring, of the meat. It may determine a value of salinity, fat content, protein content, muscle structure, or muscle destructuring, of the meat, and process this value, in combination with at least the impedanceproperty value and the optical-property value, to calculate the defect score. The bioimpedance sensor may comprise a set of any number of electrodes—e.g. a two-probe electrode or tetrapolar electrodes. In some embodiments, the electrode comprises tetrapolar electrodes comprising two signal-generating electrodes and two signal-receiving electrodes. Such an electrode set has been found to support calculating an impedance property that can accurately reflect the state of the meat, i.e. the degree of defects present, which in turn may allow for a particularly accurate defect score to be determined by the processing system. In some embodiments, the bioimpedance sensor is configured to perform bioimpedance spectroscopy to determine the value of the impedance property. In some such embodiments, bioimpedance spectroscopy is performed across a frequency range of at least 10 kHz to 1 MHz. It may be performed across at least the p-dispersion spectral band. In some embodiments in which bioimpedance spectroscopy is performed, determining the value of the impedance property may comprise modelling a set of frequency-dependent impedance measurements according to a predetermined model, e.g. the Cole function. The model may be used to calculate a difference (optionally normalised) between a first electrical impedance model parameter calculated for a first frequency and a second electrical impedance model parameter calculated for a second frequency higher than the first frequency. The impedance property may be Py. The optical sensor may be any sensor configured to sense light reflected from or emitted by the surface of the meat. The optical sensor may be configured to determine an intensity of light received from the meat. The optical property may be intensity. The optical sensor may alternatively or additionally be configured to perform spectral imaging, multispectral imaging, or hyperspectral imaging. The optical sensor may be configured to measure the light (e.g. reflected light) and to generate a spectral profile from the light from which the optical property may be determined. The optical sensor may be configured to determine the optical property from the measured spectral profile. The spectral profile may comprise intensity and wavelength data. In some embodiments, the optical property may be the intensity of light received from the meat, which may be determined across the visible spectrum (e.g. a lightness value), or across only a predetermined portion of the visible spectrum (e.g. a redness value or a yellowness value). In some embodiments, the optical sensor may comprise a colorimeter. In some such embodiments, the optical property may be a colour property. The colour property may be any colour value, but in some embodiments it is a redness value or a yellowness value. In some embodiments, the value of the optical property is a lightness value and respective values of the one or more further properties comprise a redness value and a yellowness value, determined by the optical sensor. In some embodiments, processing the impedance-property value and the optical-property value to determine a defect score may comprise calculating the defect score by evaluating a predetermined function that takes the impedance-property value, the optical-property value, and optionally values of one or more further properties of the meat, as arguments (i.e. as variable inputs). In some embodiments, the one or more further properties may comprise a measured pH of the meat. In other embodiments, processing the impedance-property value and the optical-property value to determine the defect score may comprise using a look-up table, e.g. instead of evaluating a function. The look-up table may be representative of a predetermined function as description herein, or any other function. In some embodiments, processing the impedance-property value and the optical-property value to determine the defect score may comprise inputting the values into a trained model, which may comprise an artificial neural network model. The trained model may implement a predetermined function as described herein, or any other function. In some embodiments, the predetermined function may comprise terms in which one or more of the property values is raised to a power. In some particularly advantageous embodiments, the greatest power may be two, i.e. the predetermined function may comprise one or more quadratic terms. In some embodiments, the predetermined function may be a quadratic function. This functional form has been found to produce surprisingly useful defect scores—i.e. it has been found that, when second order terms, but no higher order terms (e.g. cubic or quartic terms), are included in the predetermined function, the defect score determined by the processing system is particularly accurate. The function may have predetermined coefficients. It may have one or more coefficients that have been determined by regression analysis. The quadratic function may include a square term of the impedance-property value (i.e. with non-zero coefficient) and a square term of the optical-property value. It may have one or more cross terms, i.e. products of the arguments of the predetermined function. This may include one or more crossterms of any measured properties of the meat. It may include a term that is the product of the impedance-property value and the optical-property value. In some embodiments, it comprises a cross-term of a measured pH value and the optical-property value and / or a cross-term of the optical-property value and the impedanceproperty value. However, other embodiments may use a different function, which may include higher-order terms. In some embodiments, the defect score is a value in the interval zero to three. In some embodiments, the apparatus may comprise a temperature sensor. In some such embodiments, measurements of electrical impedance and / or pH and / or light may be performed while the meat is held at a constant temperature as determined using the temperature sensor. In some embodiments, the impedance-property parameter and / or pH value may be compensated based on a temperature value measured using the temperature sensor. The processing system may comprise a memory for storing the defect score and / or an output (e.g. a display or network connection) for outputting the defect score. It may be further configured to output measured bioimpedance and / or optical and / or pH data. It may comprise a processor and a memory storing software for implementing a method and method steps as disclosed herein. The bioimpedance sensor and optical sensor may each be a respective sensor apparatus. They may comprise analogue and / or digital circuitry for processing one or more sensed signals to determine the property values. The processing may be separate from the processing system or at least partly integrated with the processing system. The apparatus may comprise a housing. The processing system, bioimpedance sensor, and optical sensor may each be partly or wholly contained within the housing. However, this is not essential and in some embodiments the apparatus may be distributed. Features of any aspect or embodiment described herein may, wherever appropriate, be applied to any other aspect or embodiment described herein. Where reference is made to different embodiments or sets of embodiments, it should be understood that these are not necessarily distinct but may overlap. BRIEF DESCRIPTION OF THE DRAWINGS Certain preferred embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which: FIG. 1 is a schematic illustration of an apparatus for inspecting meat embodying the invention; FIG. 2 is a plot of example electrical impedance spectra, in ohms against frequency in Hertz (Hz), that might be obtained using such apparatus; FIG. 3 is a flowchart illustrating a method for inspecting meat embodying the invention; and FIG. 4 is a table showing how tissue destruction (DES) scoring was determined in an experimental validation of the method. DETAILED DESCRIPTION Figure 1 shows an apparatus 100, embodying the invention, for inspecting meat 102 for defects , e.g. defects associated with a post-slaughter change within the meat such as a metabolic process, protein denaturation or temperature-induced crystallisation. The meat 102 may be any cut from an animal carcass, or a whole animal carcass. It may be from a cow, sheep, deer, fish, or other animal. The apparatus 100 comprises a bioimpedance sensor 103, an optical sensor 105, and a pH sensor 107. In addition to the sensors 103, 105 and 107, the apparatus 100 includes a processing system 109. The processing system 109 comprises a processor 110 and a memory 111 storing software that can be executed by the processor 110 to allow the processing system 109 to perform method steps as disclosed herein. The processing system 109 is configured to receive analogue or digital signals from each of the bioimpedance sensor 103, the optical sensor 105, and the pH sensor 107. It is configured to process these signals to assess the meat 102 for defects. In some embodiments, the sensors 103, 105, and 107 (optionally excluding electrodes) and processing system 109 are all contained in a common housing 101. However, this is not essential, and the sensors 103, 105, and 107 may be distinct units, whose outputs are provided to the processing system 109 over wired or wireless links or by manual data entry. The apparatus 100 may have a wired or wireless interface 112 (e.g. USB, HDMI, Bluetooth, or WiFi) by which the processing system 109 can output information to an external device 113, which may be a computer monitor, a laptop computer, a server, a mobile phone, etc. In particular, the apparatus 100 may output meat inspection results (e.g. a defect score) for display to a user on the external device 113. Although the apparatus 100 is shown as having three sensors 103, 105, and 107, this is not required in all embodiments, and some alternative embodiments lack the pH sensor 107. Furthermore, the apparatus 100 may optionally include one or more further sensors, such as a temperature sensor, in some embodiments. The apparatus 100 is configured, in use, to measure a frequency-dependent electrical impedance response of the meat 102 using the bioimpedance sensor 103. The bioimpedance sensor 103 (or the processing system 109) then determines a value of an impedance property (also referred to herein as an impedance para meter) from the measured electrical impedance response. The apparatus 100 is also configured to determine a value of an optical property of the meat (also referred to herein as an optical parameter’) from light reflected from the meat 102 using the optical sensor 105. The impedance parameter and the optical parameter are then processed in combination by the processing system 109, optionally in combination with one or more further parameters, such as a pH value determined using the pH sensor 107, to determine a defect score (i.e. a value) representative of a degree of defects in the meat 102. This defect score may be stored in the memory 111 and / or further processed by the apparatus 100 and / or output to an external device 113. The property values may be transmitted to the processing system 109 over one or more wired or wireless links from the respective sensors 103, 105, and 107, or one or more of them may be calculated within the processing system 109 itself, e.g. based on an analogue or digital signal received by the processing system 109 from the respective sensor 103, 105, and 107. The bioimpedance sensor 103 comprises a set of tetrapolar electrodes 114 and an impedance analyser. However, other embodiments may use a different number of electrodes 114, such as two electrodes. In some embodiments, it may be a Zurich Instruments MFIA impedance analyser. The electrodes 114 include two signalgenerating electrodes and two receiving electrodes. In use, the bioimpedance sensor 103 is contacted with the meat 102. In embodiments, the electrodes may be needle-like and penetrate the meat, while in other embodiments they may make surface contact without penetration. An alternating current (AC) generator applies electrical excitation signals (e.g. successive single-frequency sinusoidal or other alternating signals within a range of 1 Hz to 1 MHz) at different excitation frequencies over time to the meat 102, through the signal-generating electrodes of the bioimpedance sensor 103. The excitation could be a time series of different alternating-current frequencies or a frequency sweep. Alternatively, the bioimpedance sensor 103 may comprise a signal generator for applying a composite signal containing a spectrum or set of frequency components simultaneously. The bioimpedance sensor 103 includes a sensor that registers voltage and current responses, from which the impedance (Z=R+jX, where R is resistance, X is capacitance, and j= sqrt(-1), representing the imaginary unit, representing the phase relationship between voltage and current) of the meat 102 is calculated at each of the excitation frequencies using the two receiving electrodes. In particular, the bioimpedance sensor 103 measures impedance values, Z, over a specific spectral band that covers a range of frequencies, between several Hz and tens of MHz (e.g. 10 kHz to 10 MHz), in which the frequency response of the meat 102 is sensitive to the cellular and tissue integrity of the meat 102 and to the ability of the meat 102 to bind water. Figure 2 illustrates a hypothetical plot 200 of impedance spectra 201, 202, 203, and 204 that might be measured using the bioimpedance sensor 103 for different meat samples. The impedance spectra 201 and 202 correspond to samples of meat 102 that are fresh, of normal quality, and largely free of defects, while the impedance spectra 203 and 204 correspond to samples of meat 102 that contain significant defects, such as “PSE-like” defects. As can be seen from Figure 2, the impedance spectra are different between meat containing defects (spectra 203 and 204) and meat that is absent from defects (spectra 201 and 202). Returning to Figure 1, the impedance values, Z, determined using the bioimpedance sensor 103 from the meat 102 across different frequencies, to, are modelled by an analyser in the sensor 103 using the Cole function: Ro-Roz = Rm + ° 1 + where: R~ and Ro are model parameters representing resistance at high and low frequencies, respectively; - j=sqrt(-1); t is a characteristic time constant of the system, corresponding to a specific angular frequency 1 / r = 2nfc, where fc is a characteristic frequency at which the absolute value of the imaginary part of impedance is largest; its value may be pre-set, e.g. having been determined using regression analysis; and a is a distribution parameter; it may be pre-set, e.g. having been determined using regression analysis, or determined graphically by plotting the data in the complex plane (e.g. as a Cole plot or Nyquist diagram). The bioimpedance sensor 103 processes the measured impedance values to determine values for R~ and Ro in accordance with this equation, and generates an impedance parameter for output to the processing system 109. In some embodiments, the impedance parameter is the so-called “Py” value, defined as the normalized difference of Ro and R»: Py = ^—X100 ''o This normalized difference describes the response drop within a specific spectral band (e.g. the p-dispersion), that is sensitive to the cellular integrity of the meat 102. Specifically, damage to meat tissue, e.g. resulting from freezing, or “PSE-like” defects can lead to a reduction in the Py value when compared to fresh, undamaged meat lacking such defects. As such, the Py value can represent a degree of defects in the meat 102. However, the inventors have realised that the utility of Py as an indicator of defects in meat 102 can be significantly and surprisingly enhanced by combining it with an optical parameter, and optionally also a pH parameter, determined from the same meat 102 to determine a defect score that depends upon both parameters. The optical sensor 105 shown in Figure 1 is a colorimeter (e.g. a Konica Minolta Chroma Meter CR-400) configured to measure a lightness value L*, a redness value a*, and a yellowness value b* from light reflected from the meat 102. The apparatus 100 may optionally include an illumination source (e.g. a LED lamp) for illuminating the meat 102. One or more of these values is then output from the optical sensor 105 to the processing system 109 as a respective optical parameter. While a colorimeter is used in this embodiment, it will be appreciated that other embodiments may use a different optical sensor, such as a light meter, an RGB camera sensor, a multispectral imager, or a hyperspectral imager. The optical sensor is preferably calibrated to determine an optical parameter (e.g. lightness) that is consistent across different instances of the apparatus 100, thereby facilitating a more objective assessment of the meat 102. The optional pH sensor 107 comprises a pHT combination electrode 117 (e.g. a BlueLine pHT electrode) that is inserted into the meat 102 for measuring pH, and a pH analyser configured to determine a pH value based on electrical signals from the pHT combination electrode. The pH value is then output to the processing system 109 where it may be used as a further parameter for determining the defect score. Having received each of the impedance parameter (Py), the optical parameter (e.g. lightness), the pH parameter, and further parameters (e.g. redness and yellowness) the processing system 109 determines a defect score, being a score representative of a degree of one or more defects in the meat 102, e.g. arising from changes that are manifest in post-mortem muscle fibres, such as a post-slaughter change within the meat. In some embodiments, the apparatus 100 may additionally comprise sensors for measuring one or more of: salinity, fat content, protein content, muscle structure, or muscle destructuring, of the meat. It may derive one or more further parameters therefrom, for processing in combination with the impedance-property value and the optical-property value to calculate the defect score. In some embodiments, the processing system 109 evaluates a predetermined quadratic function (i.e. with predetermined coefficients) that takes the impedance parameter, the optical parameters and the pH value as arguments, the output of which is the defect score representative of a degree of defects in the meat. In some embodiments, the defect score is calculated as: Score = -82.0 + 17.3 (pH) + 2.33 (L*) + 0.854 (a*) + 0.211(6*) + 0.834 (Py) + 0.0054 (L*2) - 0.035 (a*2) - 0.54 (pH x L*) + 0.1165 (pH x Py) + 0.0034 (L* x Py). This provides a value, spanning at least the interval zero to three, representative of a degree of defects in the meat, where a value of zero indicates meat that is fee of defects, while a value of three indicates meat with a high degree of defects, e.g. in which the meat is severely de-structured compared to normal quality meat. In experimental testing on meat samples (described in more detail below), this quadratic function has been found to have a strong correlation with defect scores determined by human subjects through visual inspection. However, unlike a human, instances of the present apparatus 100 can be used to produce consistent results across a wide range of locations and over time, and can be more cost effective. Nevertheless, other embodiments may calculate the defect score using different functions or model, which are not necessarily purely quadratic. These may include one or more artificial neural networks, e.g. trained on labelled training data. Coefficients or parameters for the function or model may be determined using training data comprising impedance-property values, optical-property values, and defect scores determined by human experts—e.g. using regression analysis. Some embodiments may use a look-up table representing the function, instead of evaluating a function directly. The defect score can be output for display to a user (e.g. as a numerical value or a “traffic-light” indicator or in any other way) and / or stored for further analysis. The apparatus 100 can thus be used to provide an objective analysis of meat quality through the combination of impedance data and spectral data, and optionally further data, with improved prediction accuracy in comparison to conventional approaches. In some embodiments, the processing system 109 also receives a temperature signal from a temperature sensor that measures a temperature of the meat 102, and uses the temperature signal to apply temperature-compensation to the impedance parameter. This may further improve accuracy. Although the sensors 103, 105, and 107 have been described as performing signal processing operations internally, it should be understood that, in other embodiments, some or all of this signal processing may instead be performed by the processing system 109. Any of the processing operations described herein as being performed by the processing system 109 may, in other embodiments, be performed remotely, e.g. by the external device 113. Figure 3 is a flowchart illustrating the main steps of a method for inspecting meat for defects embodying the invention. In step 301, an impedance property of the meat is derived from a measured electrical impedance spectrum of the meat. In step 302, an optical property of the meat is determined from light reflected from the meat. In step 303, the impedance property and the optical property are processed in combination, optionally in combination with one or more further measured parameters, to determine a defect score representative of a degree of defects in the meat. The processing may be in accordance with any of the steps disclosed above. It may use the quadratic formula disclosed above. Experimental Validation The approach, disclosed herein, of combining both optical and bioimpedance properties to calculate a defect score for meat, was validated against human assessment of meat for defects. Pork (Sus scrofa domesticus) ham muscle samples from the semimembranosus muscle (SM), and the adductor muscle (AD) were collected in October and November 2021 from a total of 136 animals. Pigs were slaughtered at one facility in Norway on five different days. Each day, we collected 25 to 36 samples at the cutting line, approximately 24 hours post-slaughter. To obtain samples that may represent a wide quality range, we opted for a heterogenous sample population, which included different producers and, hence, breeds as well as different farming types (free-range and conventional farming). Norwegian pigs typically have genetics that combine two or more of the following breeds: Landrace, Duroc, Z-line, or Hampshire. The ham cuts were tested using subjective visual scoring and objective, instrumental methods. The latter included bioimpedance spectroscopy, pH-testing, and colour-testing, using an apparatus embodying the present invention. Measurements were made in the chilling room of the slaughterhouse. Colour (CIELAB / L*a*b* with L*: lightness, a*: redness, and b*: yellowness) was measured using a Konica Minolta Chroma Meter CR-400 (Konica Minolta Sensing INC, Japan) with illuminant D65, a 2o standard observer, and a diameter of 0.8 cm measurement area. To allow for blooming, meat colour was determined after the surface has been exposed to air for at least one hour. The ultimate pH (pHu) was measured using a WTW pH 3110 (WTW, Germany), equipped with a BlueLine 21 pHT electrode. The pH-meterwas calibrated before measurements using fresh buffer solutions, with pH of 4.0 and 7.0. For each ham cut, pH and colour values were recorded at two different anatomical locations, i.e., in the central region of each muscle (SM, AD). The two muscles were included as there are known musclespecific differences, e.g., in pH and colour. Bioimpedance was measured using a Zurich Instruments MFIA impedance analyser (Zurich Instruments AG, Switzerland). A tetrapolar electrode setup with two signalgenerating and two receiving electrodes was used. Spectra were recorded for a frequency range from 10 Hz to 1 MHz, with 40 distinct frequency points and an applied voltage of 300 mV rms. Electrode pins were made of stainless-steel, had a diameter of 2 mm, a length of 12 mm with 18 mm spacing between the middle (voltage) pick-up electrodes. Shielded cables of 1 m length were used to connect the impedance analyser and the electrode socket. Bioimpedance was measured at the two anatomical locations noted above. Two readings were recorded for each location of the electrode. The electrode was cleaned after every measurement. For visual evaluation, we adapted a ranking system for features related to tissue destruction (DES) and colour established by I Fl P in “Prediction level of meat quality criteria on ‘PSE-like zones’ defect of pork’s ham”, Vautier et al., 54th ICoMST, 2008. A first step was to expose the inner part of the SM and AD muscles that were in close contact with the femur bone (Os femoris), where most of the structural defects are typically spotted. Then a cut with a knife was done inside the muscle tissue for internal defect detection. Lastly, the degree of structural defect severity and Japanese colour values were subjectively judged by two evaluators separately. During the study, the first 25 samples (day 1) were used for training and ‘normalizing’ the visual DES scoring among the two observers. In the training, both observers jointly evaluated the samples. For the remaining 111 samples, visual DES scores were given separately by the two observers, to also assess potential subjective, between-observer differences. The observers gave DES scores according to the feature sets detailed in Figure 4. Scoring was based on evaluating structural disintegration and visual colour. Average scores were grouped into different meat defect ranks as follows: • 0 to 0.5 = DESO: with no defects detected, • 1 to 1.5 = DES1: mild, • 2 to 2.5 = DES2: moderate, • 3 = DES3: severe We tested how visual DES scores compared with Py and the other instrumentbased measurements (pHu and CIE L*a*b* colour values) in the AD muscle. Correlation strength varied markedly among the parameters with the strongest correlation found for Py against DES score (r = -0.46, P = 0.00), followed by pHu (r = -0.44, P = 0.00), with moderate correlations for the optical properties such as colour value a* (r = 0.21, P = 0.01). Correlations were only moderate for direct comparisons between the instrument-based tests (Py, pHu, and CIE L*a*b*), indicating that these variables cover different features of quality defect. The inventors have realised that prediction of visual DES scoring can be improved by stepwise regression modelling and including more parameters than just Py to predict visual DES scoring results. In particular, they have realised that a linear multiple regression model (N = 136) of quadratic form can provide a markedly higher correlation (r = 0.71) than was found for correlations of visual DES scores with individual instrument-based parameters only. The prediction error of this model, using the Score equation set out above, was 0.76, i.e., the average deviation of the model prediction from the actual scoring data is less than 1 DES score point. Lastly, we performed a stepwise multiple regression analysis of DES scores with Py and the sensor-based tests also for the dataset excluding training data, i.e., where the two observers gave a joint score (N = 111, see above). We found that the resulting model gives a prediction error of 0.80, i.e., close to what we found for the model of the full N = 136 dataset. Hence, prediction errors for both models were comparable, i.e., independent of using the full dataset with joint scores and average DES scores or the subset with only average scores. Lastly, prediction errors for the 5 multivariate models to predict visual DES scores were also comparable to the error calculated to assess how good subjective evaluation of one observer can predict the other observer’s scoring, i.e., a measure of subjective bias. This experimental testing shows that combining optical and bioimpedance and pH 10 properties gives a more useful defect score than any of these measures independently. It also demonstrates that a quadratic model can give particularly good results, although embodiments are not limited to this. It will be appreciated by those skilled in the art that the invention has been 15 illustrated by describing one or more specific embodiments thereof, but is not limited to these embodiments; many variations and modifications are possible, within the scope of the accompanying claims.

Claims

1. An apparatus for inspecting meat, the apparatus comprising:a bioimpedance sensor configured to measure a frequency-dependent electrical impedance response of the meat and to determine a value of an impedance property of the meat from the measured frequency-dependent electrical impedance response;an optical sensor configured to receive light from the meat and to determine a value of an optical property of the meat from the light; anda processing system configured to process the value of the impedance property in combination with the value of the optical property, optionally in combination with respective values of one or more further properties of the meat determined by the apparatus, to calculate a defect score representative of a degree of one or more defects in the meat.

2. The apparatus of claim 1, further comprising a pH sensor configured to determine a pH value of the meat, and wherein the processing system is configured to process the value of the impedance property, the value of the optical property, and the pH value, in combination, to determine the defect score.

3. The apparatus of claim 1 or 2, wherein the bioimpedance sensor is configured to perform bioimpedance spectroscopy to determine the impedance parameter.

4. The apparatus of any preceding claim, wherein the optical sensor is configured to perform multispectral or hyperspectral imaging of the meat to measure a spectral profile from the light, and wherein the value of the optical property is determined from the measured spectral profile.

5. The apparatus of any preceding claim, wherein the optical sensor is configured to determine a lightness value, and wherein the value of the optical property is the lightness value.

6. The apparatus of any of claims 1 to 4, wherein the optical sensor comprises a colorimeter configured to determine a colour value, and wherein the value of the optical property is the colour value.

7. The apparatus of any of claims 1 to 5, wherein the optical sensor comprises a colorimeter configured to determine a lightness value, a redness value, and a yellowness value, and wherein the value of the optical property is the lightness value, and the values of the one or more further properties comprise the redness value and the yellowness value.

8. The apparatus of any preceding claim, wherein processing the value of the impedance property and the value of the optical property to determine the defect score comprises calculating the score by evaluating a predetermined function that takes the value of the impedance property, the value of the optical property, and optionally respective values of the one or more further properties of the meat determined by the apparatus, as arguments.

9. The apparatus of claim 8, wherein the predetermined function is a quadratic function.

10. The apparatus of claim 8 or 9, wherein the predetermined functioncomprises a cross term between the value of the impedance property and the value of the optical property.

11. The apparatus of claim 10, wherein the predetermined function further comprises a cross-term of a measured pH value of the meat and the value of the optical property.

12. The apparatus of any of claims 1 to 7, wherein processing the value of the impedance property and the value of the optical property to determine the defect score comprises using a look-up table.

13. The apparatus of any preceding claim, further comprising a temperature sensor, and configured to compensate the value of the impedance property fortemperature, based on a temperature value measured using the temperature sensor.

14. The apparatus of any preceding claim, wherein the defect score is representative of a degree of one or more defects in the meat that are manifested in post-mortem muscle fibres.

15. The apparatus of any preceding claim, wherein the defect score is representative of a degree of one or more defects in the meat associated with one or more post-slaughter changes within the meat16. The apparatus of claim 15, wherein the one or more post-slaughter changes comprises protein denaturation.

17. The apparatus of claim 15 or 16, wherein the one or more post-slaughter changes comprises crystallisation.

18. A method of inspecting meat, the method comprising:determining a value of an impedance property of the meat from a measured frequency-dependent electrical impedance response of the meat;determining a value of an optical property of the meat from light from the meat; andprocessing the value of the impedance property and the value of the optical property, optionally in combination with respective values of one or more further properties of the meat, to calculate a defect score representative of a degree of one or more defects in the meat associated with post-slaughter changes within the meat.

19. The method of claim 18, wherein the meat has at least one of: a pale, soft, exudative (PSE) defect; a PSE-like defect; freeze damage; or an anomaly during post-slaughter muscle-meat conversion.

20. Computer software comprising instructions which, when executed by a processing system, cause the processing system to:receive or calculate a value of an impedance property of the meat determined from a measured frequency-dependent electrical impedance response of the meat;receive or calculate a value of an optical property of the meat determined5 from light received from the meat; andprocess the value of the impedance property and the value of the optical property, optionally in combination with respective values of one or more further properties of the meat, to calculate a defect score representative of a degree of one or more defects in the meat associated with post-slaughter changes within the10 meat.

Images