"Food Monitoring System"
FIELD OF THE INVENTION
This invention relates to the cooking of food products, and in particular to the monitoring and control of the cooking process.
BACKGROUND OF THE INVENTION
By visually inspecting food it is possible to grade the quality of food, which adds value and altows producers to recover more of the end product. For example, in the case of pre-cooked chickens, the golden coloured one will sell much quicker than the paler one, even if both are safely cooked to a core temperature specified by safety standards. In the case of cooked meat this generally darkens with time after the cooking process has been completed. Therefore, it is possible to ensure a longer shelf life for the product by cooking it to the correct colour in the first instance and not allowing it to get too dark. This increases profits and reduces wastage for the customer e.g. the supermarket. Several consumer surveys have shown that consumers attach more importance to high quality produce than on low prices and thus there is a need for a system which quantitatively can detennine the important physical parameters of quality.
In European Patent Specification No. 0232802 there is disclosed apparatus for monitoring cooking of a food product. A probe is inserted into the food product. Light is transmitted between spaced-apart rods of the probe, the amount of light transmitted through the food material giving an indication of the cooked state of the food product. When the change in transparency of the food material measured over a selected time interval tends to zero this gives an indication that the food product is cooked. In WO 91/11136 there is disclosed apparatus and a method for reflecting light from a food product and comparing ratios of two reflected wavelengths to give an indication of the cooked state of the meat product.
SUMMARY OF THE INVENTION
Essentially the invention provides a food monitoring system in particular for use during a food product cooking process, said system using the sensed cotour of a food product to determine the cooking of the food product with a view to optimising the cooking process.
Conveniently the system of the invention can be used for determining the cooked state of a food product by sensing the cotour of the food product and using this sensed colour to control the cooking process.
The internal cotour and/or the external colour of the food product may be determined and used to give an indication of the cooked state of the product. The system may further provide for the sensing of the temperature of the food product.
In a particularly preferred embodiment the system includes:
means for generating electromagnetic radiation,
means for directing said electromagnetic radiation at a food product,
means for collecting reflected electromagnetic radiation from the food product, and
means for analysing said reflected radiation for determining the state of the food product,
characterised in that said electromagnetic radiation is in the visible light range and means is provided for generating a spectral pattern of the reflected light and means is provided for analysing the spectral pattern of the reflected light to determine the state of the food product.
In a preferred embodiment the system further includes a temperature sensing means operable for determining the temperature of the food product.
In another embodiment the means for directing the light at the food product and collecting reflected light from the food product includes juxtaposed optical fibres mounted on an optical probe, including at least one light emitting optical fibre connected between a light source and a window of the probe and at least one light collecting optical fibre connected between the window of the probe and the light analysing means.
In another embodiment two optical fibres are provided in the probe.
In a further embodiment a light emitting optical fibre is provided and a plurality of light collecting optical fibres are provided. Conveniently the light emitting optical fibre is surrounded by said plurality of light collecting optical fibres.
In another embodiment the temperature sensing means includes a temperature sensing probe mounted on the optical probe.
In another embodiment means is provided for generating a spectral pattern of the reflected light and means is provided for comparing said spectral pattern with a library of known spectral patterns associated with different cooking states of the food product to determine the state of the food product.
In a further embodiment the system includes an artificial neural network which is operable for analysis of the spectral patterns generated to determine the cooked state of the food product.
In a further embodiment the system includes means for controlling operation of a food product cooking process in response to the sensed state of the food product.
In another embodiment the means for generating electromagnetic radiation is a white light source.
In another embodiment the temperature sensing means includes:
a pulsed laser diode light source,
a temperature sensing probe,
a first optical fibre connected between the pulsed laser diode light source and the temperature sensing probe,
a second optical fibre connected between the temperature sensing probe and a fluorescence decay time processor,
said temperature sensing probe comprising a body within which is housed an optical fibre,
the optical fibre having an inner end and an outer end,
a temperature sensitive crystal being mounted at the inner end of the optical fibre,
the outer end of the probe optical fibre being connected to the first and second optical fibres.
The temperature sensitive crystal mounted within the temperature probe alters its optical properties in the presence of a temperature change so that the temperature may be measured in this way. The temperature is determined by measuring the fluorescence decay time constant of the received signal foltowing a pulse by the input laser diode.
Conveniently the light detector may be provided by a spectrometer. A spectrum generated by the collected light is compared with a database of stored spectral signatures associated with different stages of cooking of the particular food product being tested for determining the cooked condition of the food product.
In an alternative arrangement a CCD detector may be used instead of the spectrometer.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention will be more cleariy understood by the foltowing description of some embodiments thereof, given by way of example only, with reference to the accompanying drawings, in which:
Fig. 1 is a schematic illustration of a food monitoring system according to the invention;
Fig. 2(a) is a detail sectional elevational view of a colour sensor probe used in the system of the invention;
Fig. 2(b) is a detail sectional elevational view of an optical temperature probe of the system;
Fig. 2(c) is a cross sectional view of another colour sensor probe;
Figs. 3 to 6 are illustrations of different spectral signatures associated with colour changes in selected food products as they cook;
Fig. 7 is a schematic illustration of a neural network structure for the system;
Fig. 8 is a training graph; and
Fig. 9 is a typical temperature sensor output whilst monitoring a product which has been cooked in an oven and then cools gradually,
Fig. 10 is a plan view of a cooking system incorporating the monitoring system according to the invention;
Fig. 11 is a detail perspective view showing a sensor probe mounting carriage of the system; and
Fig. 12 is a detail perspective view of the sensor probe mounting carriage.
DETAILED DESCRIPTION OF THE INVENTION
Referring to the drawings, and initially to Figs. 1 and 2 thereof, there is illustrated a food monitoring system according to the invention indicated generally by the reference numeral 1. The system 1 has a tungsten halogen white light source 2 with means 3 for directing light from the source 2 at a food product sample 4. Said means 3 includes a light emitting optical fibre formed by a first optical fibre 5 which terminates in an internal colour sensor probe 6, the first optical fibre 5 delivering white light to the food sample 4. A light collecting optical fibre is formed by a second optical fibre 7 which extends between the colour sensor probe 6 and a spectrometer 8, delivering reflected light from the food sample 4 to the spectrometer 8. The spectrometer 8 is connected to a PC 10 by a communications link 9.
Temperature measuring means includes a pulsed infrared laser diode light source 11 with means, indicated generally by the reference numeral 12, for directing light from the source 11 at the food product sample 4. Said means 12 includes a third optical fibre 13 which terminates at a temperature sensor probe 14, the third optical fibre 13 delivering the pulsed laser diode light to the temperature probe 14. A fourth optical fibre 15 extends between the temperature sensor probe 14 and an optical detector and fluorescence decay time processor 16, delivering reflected light from the temperature sensor probe 14 to the fluorescence decay time processor 16. The fluorescence decay time processor 16 is connected to the PC 10 by a communications link 17.
Referring in particular to Fig. 2(a) portion of the colour sensor probe 6 is shown in more detail. The colour sensor probe 6 has a stainless steel body 18 within which outer ends of the first and second optical fibres 5, 7 are mounted. These first and second optical fibres 5, 7 have a diameter in the order of 1 mm. Each of the first and second optical fibres 5, 7 is encased in a plastic sheath 19 and 20 respectively. The first and second optical fibres 5, 7 are mounted spaced-apart within the stainless steel body 18 and terminate at a window 28 at an outer end of the probe body 18. The first optical
fibre 5 transmits white light from the white light source 2 to the food sample 4 and the second optical fibre 7 returns reflected light from the food sample 4 back to the spectrometer 8.
Referring in particular to Fig. 2(b) the optical temperature sensor probe 14 is shown in more detail. The temperature sensor probe 14 has a ceramic or stainless steel body 21 within which a single silica optical fibre 22 is mounted. A temperature sensitive crystal 23 is mounted at an inner end of the optical fibre 22. An outer end of the optical fibre 22 is coupled to the third and fourth optical fibres 13, 15 externally to the temperature sensor probe 14 using an optical coupler 25 (Fig. 1). The optical temperature sensor probe 14 and the internal colour sensor probe 6 may be mounted in the same housing for measurement purposes if so desired.
The optical temperature sensor probe 14 comprises a single optical fibre 22 inside a stainless steel protective sheath with the temperature sensitive crystal 23 mounted at its end and fused to it for a good optical coupling. This too is protected by a stainless steel sheath or body 21 and therefore no direct contact is made between the food sample 4 and the optical part of the temperature probe 14. The optical signal propagating in this probe 14 is coupled into two smaller diameter optical fibres 13 and 15 which transmit the optical signals to and from the light source 11 and the fluorescence decay time processor 16 respectively. The infrared laser diode pulse is transmitted via the fibres 13 and 22 to the crystal element 23 at the probe end of the sensor 14 and this excites a fluorescence in the crystal element 23. When the laser light pulse ceases (the end of the pulse) although this can be considered to be instantaneous, the resulting fluorescence signal returns via fibre 15 actually decays more gradually to zero. The length (in time) of this decay signal is related to the temperature of the food sample 4 in which it is embedded. The fluorescence decay time processor 16 accurately measures this decay time and converts it to a temperature.
Referring now to Fig. 2(c) an external optical colour sensor probe is shown indicated generally by the reference numeral 30. This has a stainless steel outer casing 31 within which are mounted a number of optical fibres comprising a central light emitting optical fibre 32 which is encircled by an associated ring of light collecting optical fibres
33. The optical fibres 32, 33 are all encased in plastic sheaths 35, 36 respectively. The light emitting optical fibre 32 is connected to the white light source 2 and the light collecting optical fibres 33 are connected to the spectrometer 8. This type of optical fibre probe 30 is useful for determining the extemal colour of the food product 4. Light is delivered from the white light source 2 through the central optical fibre 32 and light reflected from the surface of the food product 4 is collected through the outer optical fibres 33 for delivery to the spectrometer 8 for freatment and analysis as described later. This probe 30 would normally be located close to the surface of the food product 4, typically within 3 mm of the surface of the food product 4 without touching said surface of the food product 4.
Referring to Figs. 10 to 12 inclusive the food monitoring system 1 is shown incorporated in a cooking system 50. The cooking system 50 has an oven 51 through which a steel mesh food transport conveyor 52 travels to deliver food products such as burger patties and the like through the oven 51 for cooking the food products. Downstream of the oven 51 the conveyor 52 may optionally deliver cooked food products through a chiller 53 prior to packing the food products. Any suitable oven may be provided such as that described in Irish Patent No. 74240 for example.
A sensor assembly indicated generally by the reference numeral 55 is mounted above the conveyor 52 downstream of the oven 51. While the sensor assembly 55 is shown in the open in Fig. 10, this is for illustration purposes and in practice the sensor assembly 55 may be housed within an outer casing (not shown).
The sensor probes 6, 14, 30 are mounted on an X Y Z actuator having a transverse slide mechanism 56 for movement in the transverse or X direction, a longitudinal slide 57 for movement in the longitudinal or Y direction and a pair of vertical slides 58, 59 for movement in the vertical or Z direction. The vertical slides 58, 59 are slidable vertically on a mounting frame 60. This mounting frame 60 is in tum slidable along an arm 61 of the longitudinal slide. An inner end 62 of the longitudinal slide arm 61 is in tum mounted on a carriage 63 for sliding transversely on a transverse slide ami 64 of the transverse slide 56.
The extemal probe 30 is mounted on one of the vertical sliders 59 and the internal
probe 6 and temperature probe 14 are mounted on the other vertical slider 58. The internal probe 6 and temperature probe 14 are movable downwardly through a split sponge 66 mounted in a holder 67 on a support arm 68 on the mount 60. A reservoir 70 of sterilising fluid is connected by feed pipe 71 with the sponge 66 to supply sterilising fluid to the sponge 66. The probes 6, 14 move through the sponge 66 downwardly into a food product on the conveyor 52 and are cleaned as they are retracted up through the sponge 66 upon withdrawal from the food product. A stripping arm 74 mounted on the support 60 extends below the holder 67 for the sponge 66 and has a window 75 at its outer end through which the probes 6, 14 extend into the food products.
Laser or ultrasonic sensors detect the presence of a food and direct the actuator to position the probes 30, 6, 14 over the meat product on the conveyor. The arm 61 is moved along the arm 64 to align the external probe 30 with the food product on the conveyor 52. The vertical slide 59 is then operated to lower the probe 30 to within 2 or 3mm of the surface of the food product. The support 60 is moved along the arm 61 to keep pace with the conveyor 52 so there is no relative movement between the probe 30 and the food product on the conveyor 52. The probe 30 can then take a reading of reflected light from the surface of the food product as previously described. Next the internal probe 6 and temperature probe 14 are positioned over the food product and these probes 6, 14 are advanced into the food product to take the internal readings. Again, the support 60 keeps pace with the food product on the conveyor 52 so there is no longitudinal movement of the food product relative to the probes 6, 14. When the readings of internal light reflection and temperature have been taken the probes 6, 14 are retracted upwardly, being cleaned by the sponge 66. The apparatus 55 is then repositioned for the next reading.
The sensor assembly 55 is connected by communications link 80 with a controller 81 for controlling operation of the oven 51. Thus the oven 51 can be cook controlled in response to the sensed state of the food product for optimum cooking of the food product.
The visible spectrum measured by spectrometer 8 is made up of wavelengths that can be perceived by the human eye and range from 400nm (violet) 700nm (red). By
shining a white light source on an object it absorbs some wavelengths and emits others, giving it its cotour. The tungsten halogen white light source 2, used to obtain the colours of the different cooking stages of a meat food product 4, ranges from 400nm to 2μm.
Cooking meat consists of the reduction of deoxymyoglobin and oxymyoglobin to metmyogtobin and sulfymyoglobin. There are three visible bands at 485nm, 560nm and 635nm for metmyoglobin, oxymyoglobin and sulfymyoglobin and during cooking there is an increase in intensity at bands 485nm and 635nm and a decrease of intensity at 560nm. These are the wavelengths that colour changes occur in meat, i.e. they are in the visible region of the electromagnetic.
Spectra of several samples of meat (internal and extemal) at various stages of cooking are obtained. The spectra of the food samples are recorded on the spectrometer 8, acquired using LabVIEW software via a National Instruments PCI data acquisition card and stored on the computer 10 for subsequent analysis using the Stuttgart Neural Networks SNNS software. The system includes one or both of the colour sensor probes 6, 30 with an Oceanoptics S2000 spectrometer 8 coupled to a Pentium PC 10. An option exists for controlling the spectrometer 8 through GPIB (General Purpose Interface Bus), but for this application it is sufficient to operate it in normal mode i.e. "continuously monitoring." The system configuration is shown in Fig. 1.
Tests were carried out on various types of foods including patti-burgers, minced beef burgers, processed rib-steaks, sausages and mushrooms. The food products were cooked in a large-scale industrial oven, on a conveyor belt.
The spectra of Figs. 3 to 5 show the clear variation in the received light intensity for the cases of minced beef, patti-burgers and sausages respectively at the different stages of cooking. They correspond to variations in appearance from too light (undercooked) correct (cooked), and too dark (overcooked). This means that the neural network should have no difficulty in categorising each colour. The majority of the differences occur between 40nm and 750nm, so this is the range that is used for the classification.
A neural network can be designed, based on the spectral graphs for the food products which successfully distinguishes between the cotour being too light, too dark or correct.
The results shown in the graphs of Fig. 3 to Fig. 5 all demonstrate the changes that occur in meat, however it is possible to apply the same theory to other foodstuffs, e.g. vegetables. Fig. 6 shows the difference between raw and steamed mushrooms, and it is clear from Fig. 6 that difference does exist.
The spectrum of each food sample represents the distribution of received light across a wide range of wavelengths in the visible and near infrared ranges. These signals are complex and the patterns may be masked by interfering parameters such as localised light absorption due to fatty acids. It is therefore necessary to apply advanced signal processing and pattern recognition techniques to categorise each sample and isolate interfering parameters. The neural networks facilitates pattern recognition. This multilayer network is capable of classification even in the face of linear inseparability of the input parameters.
The software simulator used to implement the neural network is SNNS (Stuttgart Neural Networks Software), developed at the Institute for Parallel and Distributed High Performance Systems (IPVR) at the University of Stuttgart All pre-processing of the input signals to the network is performed by LabVIEW™ data acquisition and analysis software.
Initially the spectrum of each sample is normalised to its maximum value, this being assigned a value of 1 and all other values lie in the range between 0 and 1. This pre processing of the data is required by the Neural Networksince the activation function used by the processing units of the network is the Sigmoid Function (Equation 1). This is a monotonic and continuous with outputs in the range 0 to 1. It is also non-linear which enables the network to solve problems that are linearly inseparable.
1
/(*) = l + e -X (1)
The input layer receives the pre-processed input data and the ANN distributes the output o each of these nodes or processing units to the next layer of nodes i.e. the hidden layer. There can be more than one hidden layer, although for this investigation only one was used. Together with the output layer, this is a processing unit which receives data from the previous layer. The input layer is defined mathematically in Equation 2
/ net = Yjwijoi (2) ι=0
Where / refers to the input layer and j the hidden layer. w<, is the weight ( a numerical factor) of the link between the input and hidden layers. It is called the activation since it gives information to the processing unit about how active the connection are. The output (Equation 3) of the processing unit in the hidden layer depends upon the activation function, in this case the sigmoid function (equation 1). This is the forward propagation of the Network
Oj = f net) (3)
The network chosen is a feed forward with one hidden layer, made of 3 units. The learning function used is Backpropagation. During learning the output of the output layer is compared with the desired pattern response. The error § (Equation 5) i.e. the difference between the output and the desired pattern response fyof a unit j, is used together with the output of the preceding unit oy to calculate the necessary changes to the weight of the link between units /and./, i.e. w, (Equation 4). Since there are no desired pattern responses for the outputs of the hidden layers, &, is computed using the < s of the following layer, which will have already been calculated (Equation 6). In this way the <?s are propagated backwards, thus the expression Backpropagation.
Awy it + ^ ηjO; (4)
δ ~ f net J ){tj - Oj ) if j is an output unit (5)
j ~ J (net)2 ι °kwjk if j is a hidden unit (6)
Where
η learning factor (a constant for a given network) δ error (difference between the real output and the teaching output) of unity tj desired pattern response of unity
Of output of preceding unit /
/ index of a predecessor to the current unit with link w,, from /' to y j index of current unit k index of a successor to the current unit j with link wjk fromy to k
The spectra observed using the Ocean Optics Spectrometer each occupy a wavelength range of 200 nm to 875 nm. Rather than using all of this information to represent the input to the neural network, it is more benefit to carry out feature extraction. This is due to the Curse of Dimensionality which states that including too many input values can actually lead to a reduction in the performance of the classification system. Therefore, looking st the various spectra, it is clear that most differences occur in the region between 400 nm and 750 nm (i.e. the visible and part of the near infra-red spectrum) and it is therefore this range, extracted from the overall range that is used in the training. This reduces the number of inputs to 1062 input units instead of 2048. However, 1062 is still a large number and the input data are highly correlated, therefore further pre-processing could be carried out while still maintaining the same spectral shape. It has been found that by taking every 11th input, instead of all of them. The shape is very well maintained but the number of inputs is reduced to 96 indicated generally by reference numeral 40 in Fig. 7. After the samples have been conditioned, pattern files are generated, again using LabVIEW ™ in the foπnat that is compatible for use with SNNS.
The number of outputs 41 depends on the amount of conditions being tested e.g. too light, correct colour and too dark. For example a neural network has been designed for the classification of patti burgers according to their colour. The samples are named according to their core temperature at each colour. In this case there are burgers cooked to 82 °C (too light), 91 °C (correct colour) and 95 °C (too dark), therefore there are three outputs 41 for the network of this investigation. The network topology is shown in Fig. 7.
Determining the amount of processing units in the hidden layer 42 is largely achieved by trial and error, but it is however desirable to have as few hidden layers and nodes within the layers as possible as this reduces the size of the network. It also means that the Network will train in the minimum number of cycles (provided it does train). The best training graph achieved (Fig. 8) corresponds to a network with 8 units in the hidden layer. Fig. 8 shows the sum squared error per number of outputs as a function of training. Since three conditions were being tested, there were three output nodes. The desired pattern response were as follows: too light = 100 (binary representation), correct colour =010, too dark = 001. The parameters chosen for this network were learning rate a = 0.5, maximum tolerated error δmm = 0.1. These parameters are required for the SNNS simulator.
When the network was tested with different samples than those used in training, it clearly recognised each colour from the acquire optical spectrum.
The optical temperature signal obtained manually at the end of the oven is shown in graphical form in Fig. 9. Fig. 9 shows the temperature recorded against time (probe inserted into chicken following 50 seconds of elapsed time) as a sample of chicken at approximately 65° C emerges from the end of an oven. In the time period just prior to emerging from the end of the oven (between 50 seconds and 100 seconds of elapsed time), the temperature is relatively stable at about 65° C. Immediately upon exiting the oven it then cools rapidly to about 55° C where it becomes more stable and cools more slowly (100 seconds to 150 seconds of elapsed time). The probe was withdrawn from the chicken at an elapsed time of about 150 seconds and the temperature reading rapidly returns to room temperature.
The results have shown that it is possible to use optical fibre sensors to determine if a product is cooked to the optimum colour. It gives rise to the possibility of intelligent ovens that are controlled to cook, bake or roast to a certain cotour once the core temperature, as recommended in the regulations of the food safety authorities, is reached. The system of colour measurement according to the invention ensures that the colour appearance, internally and externally will be consistent for the customer and the temperature measurement ensures that the food product is safely cooked.
The invention is not limited to the embodiments hereinbefore described which may be varied in both construction and detail within the scope of the appended claims.