GB2420403A

GB2420403A - Analysing a food sample by the use of light scattering

Info

Publication number: GB2420403A
Application number: GB0425753A
Authority: GB
Inventors: Flemming Moller
Original assignee: Danisco AS; Danisco US Inc
Current assignee: DuPont Nutrition Biosciences ApS; Danisco US Inc
Priority date: 2004-11-23
Filing date: 2004-11-23
Publication date: 2006-05-24
Also published as: GB0425753D0

Abstract

The method includes: Directing a beam 101 from a source such as a laser 10 at the food sample 60; obtaining a digital image of light scattered 103 from the food sample; and performing statistical analysis on the image to extract parameters that are useable to assess the sample. The light beam may comprise visible radiation with a cross-section diameter in the range of 0.5 -10 mm. The beam may be incident to the sample at angle ranging from 30 to 60 degrees. The scattered light 103 may be to within 10 degrees to normal to the surface of the sample. The image of the scattered light (200, fig. 2) may have a first and second portion (202, 201 respectively in fig. 2). The image of the second portion (201, fig.2) comprises the light scattered from within the sample, whereas the first portion (202) comprises the image of scattered light from the surface of the sample. The statistical analysis may include a textural analysis using a gray level co-occurrence matrix. Multiple measurements may be performed to follow the food sample though a dynamic change, heating process, fermenting, or identifying a gradual degradation. Beam illumination, and scattered light reception may be performed via an optical fibre that is to be inserted into the food sample for doing the analysis.

Description

FOOD SAMPLE ANALYSIS USING LIGHT SCATTERING

Field of the Invention

The present invention relates to the analysis of food samples, and in particular to the use of scattered light for performing such an analysis.

Background of the Invention

The use of laser light as a diagnostic or analytical tool is known in various areas of technology. One particular application of laser light utilises scattering to determine particle size. This application is based on the principle that small particles tend to scatter light more than large particles. A review of scattering measurements can be found in "Particle Size Distribution - Determination by Laser-Light Scattering" by Harry G Brittain, in Pharmaceutical Technology, Volume 27/10, 2003, pages 102-114. The general aim of scattering measurements is to ascertain the intensity of scattered light as a function of the scattering angle. This angle is measured between the incident light and the scattered light. The scattering is normally assumed to be rotationally symmetric about the axis of the incident light.

Where the diameter of the particles being measured is small compared to the wavelength of light, for example less than 10 per cent, particle size can be estimated on the basis of classical (Rayleigh) scattering theory. Alternatively, if the particle size is significantly greater than the light wavelength, for example four or more times the wavelength, diffraction theory can be used to interpret the scattering results. For intermediate values, a more complex approach based on the Lorenz-Mie theory must be used.

The above-mentioned paper by Brittain illustrates a typical detector for the measurement of scattered light intensities, comprising concentric rings for measuring scattered light intensity at different scattering angles. The centre of the ring is defined by the straight-through direction for the incident light. One commercial system for performing such laser-light particle size analysis is the Malvern Mastersizer 2000 system.

Scattering measurements are harder to interpret if light is scattered off multiple particles, i.e. first from one particle, then onto another particle, and potentially onto further particles, before being detected. In such cases it can be difficult to relate directly the scattered light intensity (as a function of scattering angle) to the particle size.

Consequently, particle size measurements using laser-light scattering are frequently performed on very dilute samples, in order to minimise such multi-particle scattering.

It is also known that the scattered or reflected light from a sample may be altered in frequency compared to the incident radiation. This frequency change is caused by a Doppler shift due to the Brownian motion of the scattering particles. Accordingly, by looking at the frequency broadening of the scattered radiation, it is possible to obtain information about the range of particle motions within the sample. This leads to another method of particle size estimation, known as photon correlation spectroscopy (PCS).

Again, PCS normally requires the use of very dilute solutions.

The use of polarisation in light scattering measurements is described in "A New Integrated Rheo Small Angle Light Scattering (Rheo-SALS) Device", by Lauger, Weigel, Higher and Pfeifer, in Annual Transactions of the Nordic Rheology Society, Volume 12, 2004, pages 137-140. If the incident laser radiation onto a sample is polarised, then the scattered light from the sample may be at least partially depolarised. The above paper shows how a small angle light scattering (SALS) device can be combined with a rheometer in order to provide simultaneous particle size and rheometry measurements. In the described apparatus, a CCD camera is used record the different polarisation patterns of the scattered light, which are dependent upon the relative orientation of the initial polansed light from the laser compared to a subsequent polarised filter placed after the sample. This device is described primarily in context of investigating suspensions and emulsions such as fresh concrete.

The laser-light scattering method can also be used to derive particle concentrations as well as particle sizes, since particle concentration affects the amount of scattered light. One example of this approach is described in "Platelet Dysfunction in Acute Type A Aortic Dissection Evaluated by the Laser-Light Scattering Method" by Tanaka, Kawahito, Adachi and mo, in the Journal of Thoracic and Cardiovascular Surgery, Volume 26/3, September 2003, pages 837-841. In this paper, laser-light scattering is used to quantify changes in the number of platelet aggregates of different sizes in blood.

Another approach to the user of laser-light scattering to determine particle size is described in "Measurement of Droplet Size Distribution of Gasoline Direct Injection Spray by Droplet Generator and Planar Image Technique" by, Park, Cho, Yoon and Mm, in Measurement Science and Technology, Volume 13, 2002, pages 859-864. In this paper, gasoline droplets are illuminated by a laser and a CCD camera is used to detect both scattered light and also light produced by fluorescence of the sample - note that these two different mechanisms are distinguished by using frequency filtering to separate the two components. The radial distribution of the scattered light is determined, and the comparison of the scattered and fluorescent light yields an improved measurement of droplet size.

The use of low angle laser-light scattering in the ceramic industry to investigate particle shape for kaolins is described in "Extraction of Shape Information from Particle Size Measurements' by Pabst, Kunes, Gregorova and Havrda, in British Ceramic Transactions, Volume 100/3, 2001, pages 106-109. In this paper it is observed that two particle size measurement techniques, namely sedimentation and laser-light scattering, are both dependent upon the shape of the particle. Accordingly, a comparison of particle size measurements from these two different techniques can provide information about particle shape.

Another example of measurements using a laser diffraction particle sizer is described in "Effects of Flow Behaviour on the Aggregation of Whey Protein Suspensions, Pure or Mixed with Xanthan", by Walkenstrom, Neilsen, Windhab, and Hermansson, in the Journal of Food Engineering, Volume 42, 1999, pages 42-56. This paper investigates the impact of particle size on flow behaviour. Prior to measurement with the laser diffraction particle sizer, a relatively time-consuming procedure must be followed to dilute the samples.

In summary therefore, laser-light scattering techniques have generally concentrated on the measurement of certain specific physical parameters, especially particle size and size distribution, particle shape, and particle concentration. For these parameters, there is a reasonable understanding of the underlying physics, so that a determination of the relevant parameter(s) can be made by measuring the variation of scattered light intensity with scattering angle. However, such measurements generally have to be performed in carefully controlled conditions, often with diluted samples, and so are difficult to apply in many circumstances.

Summary of the Invention

Accordingly, one aspect of the present invention provides a method of analysing a food sample using light scattering by: directing a light beam at the food sample; obtaining a digital image of light scattered from the food sample; and performing a statistical analysis on the digital image to extract one or more parameters, wherein the one or more extracted parameters may be used assess the sample.

The approach described herein offers a quick, non-invasive, nondestructive, low- cost and highly flexible method for analysing food samples. The approach described herein can be used while the sample is subjected to any desired processing (e.g. heating and so on), and does not contaminate or otherwise alter the food sample being measured.

Accordingly, the approach described herein is well-suited to use in a production (manufacturing) environment, such as for performing quality control on input to andlor output from a manufacturing process, and/or for controlling the operation of the manufacturing process itself.

The approach described herein differs significantly from most existing analysis tools, which generally try to measure directly some specific physical property of interest (e.g. acidity, particle size, and so on), and so normally try to simplify the circumstances of the measurement as much as possible (e.g. by dilution) in order to aid physical interpretation of the measurement. In contrast, the approach described herein concentrates on a statistical analysis of the light scattering measurements rather than a direct physical interpretation, and can therefore accommodate the complexity that arises from in situ investigations.

The light beam may be produced by a laser or other suitable light source, and may comprise visible radiation or some other suitable wavelength (e.g. near infrared). The beam of light may have a cross-sectional diameter of at least 0.5 mm, more preferably a cross-sectional diameter in the range 1-10 mm, and more preferably a cross-sectional diameter in the range 1.53 mm. The diameter of the light beam used in any given measurements may be adjusted depending upon the properties of the particular food sample being investigated. For example, the choice of beam diameter may be influenced by the scale of any surface structure for the food structure (e.g. for a foam), as well as by the opacity of the sample.

The beam of light may have an angle of incidence onto the surface of the sample in the range 30 to 60 degrees, and more preferably an angle of incidence in the range 40 to 50 degrees. The image may be obtained for light scattered no more than 10 degrees from the normal to the surface of said sample. Such a configuration minimises any reflected light in the image.

The image may be obtained from light scattered within a cone of 10 degrees apex or less, more preferably within a cone of 4 degrees apex or less. In other words, the image is obtained at a relatively constant scattering angle within the cone. This is in contrast to many conventional light scattering techniques, which work by measuring scattered light intensity across a range of scattering angles.

The image formed by the scattered light may comprise a first portion and a second portion, where the first portion comprises light scattered from the surface of the sample, and the second portion comprises light scattered from within the sample. The first and second portions therefore investigate different aspects of the sample (surface and volume). In general the second portion is found to extend around the first portion.

In one experimental configuration, the scattered light may extend across at least pixels of the digital image, and more preferably across at least 200 pixels of the digital image. Having the image extend across a large number of pixel provides better resolution within the image, and allows a more accurate statistical analysis to be performed.

The statistical analysis may comprise performing a textural analysis on the image, and this analysis may be performed at multiple image scales. The textural analysis may involve calculating a gray level co-occurrence matrix, or be based on any other appropriate analysis technique, such as auto-regression. The statistical analysis may comprise performing a principal component analysis, wherein the extracted parameters comprise one or more principal components from the principal component analysis.

Such a principal component analysis generally helps to obtain results in a more compact and usable format. The principal component analysis may be performed on statistical image parameters obtained from a gray level cooccurrence matrix for the digital image or from any other textural analysis technique. Alternatively, the statistical analysis may extract parameters directly from a textural or other image analysis technique, without performing a principal component analysis.

The approach described herein may include multiple measurements performed with different experimental parameters. For example, one measurement may be performed with a first wavelength for the light beam, and a second measurement may be performed with a second wavelength for the light beam. Other experimental parameters that may be varied between measurements include the polarisation, intensity and/or angle of incidence of the light beam. In addition, the food sample may also be subject to different conditions between measurements, such as a different temperature. Varying experimental parameters in this manner provides additional data that can be incorporated into the statistical analysis.

The approach described herein may also be used to follow the food sample through a dynamic change, such as heating the sample, fennenting the sample, adding a food ingredient to the sample, aging the sample, stirring the sample, and so on. The quick response time for obtaining results (image capture followed by data processing) allows repeated measurements to follow accurately any change in the sample. In some cases, the measurements may just monitor the change; in other cases the measurements may be used to actually control a process performed on the food sample that produces the change, such as by determining an endpoint of the process (i.e. when the process should stop), and/or a dynamic trajectory of the process (e.g. what temperature to apply to the food sample on an ongoing basis). The process being monitored and/or measured might represent the degradation of a food product, the manufacturing process for the food product, or any other process of interest.

In one implementation, an optical fibre may be used to direct the light beam at the food sample. In some cases, the optical fibre may be inserted into the food sample, thereby providing a light source within the food sample.

For some measurements, multiple light beams may be directed at the food sample, and for each light beam the light scattered from the food sample is imaged. In some cases, the light scattered from the food sample may captured on a single image for all light beams, while in other cases separate images may be obtained. The use of multiple light beams on a single sample can be used for checking homogeneity within a product, and/or for obtaining more accurate measurements on a single product.

Another embodiment of the invention provides apparatus for analysing a food sample using light scattering comprising: a light source for directing a light beam at the food sample; an imaging device for obtaining a digital image of light scattered from the food sample; and a computational device for performing a statistical analysis on the digital image to extract one or more parameters, wherein the one or more extracted parameters may be used assess the sample.

It will be appreciated that this apparatus embodiment of the invention may utilise and benefit from the same particular features as described above in relation to the method embodiment.

As used herein, a food sample may comprise (without limitation) a food ingredient, a food product, a food supplement, or a substance used in the production or preparation of food. The food may be for human or animal consumption - in a preferred aspect, the food is for human consumption. The food sample may (without limitation) be in the form of a solid, a solution or liquid (such as a drink), a foam, a paste, a jelly and so on.

As an example, the food sample may be selected (without limitation) from eggs, egg-based products, including but not limited to mayonnaise, salad dressings, sauces, ice creams, egg powder, modified egg yolk and products made therefrom; baked goods, including breads, cakes, sweet dough products, laminated doughs, liquid batters, muffins, doughnuts, biscuits, crackers and cookies; confectionery, including chocolate, candies, caramels, halawa, gums, including sugar free and sugar sweetened gums, bubble gum, soft bubble gum, chewing gum and puddings; pies, noodles, snack items such as crackers, graham crackers, pretzels, and potato chips, and pasta; frozen products including sorbets, preferably frozen dairy products, including ice cream and ice milk; dairy products, including cheese, butter, milk, coffee cream, whipped cream, custard cream, milk drinks and yoghurts; mousses, whipped vegetable creams, meat products, including processed meat products; edible oils and fats, aerated and non-aerated whipped products, oil-in- water emulsions, water-in-oil emulsions, margarine, shortening and spreads including low fat and very low fat spreads; dressings, mayonnaise, dips, cream based sauces, cream based soups, beverages, spice emulsions, sauces and mayonnaise; or a plant-derived product such as flours, pre- mixes, cocoa butter, coffee whitener, and peanut butter.

A food sample to be investigated may comprise a complete food product or item, such as a whole loaf of bread, or a bulk material, such as dough; alternatively, a food sample may comprise a portion removed from such an item or material for testing. In the former case, the food sample may comprise an item or material at some stage of a food production process, and which is destined for human or animal consumption after the testing.

Brief Description of the Drawings

One or more embodiments of the invention will now be described in detail by way of example only with reference to the following drawings: Figure 1 is a schematic representation of apparatus for performing a laser-light scattering analysis in accordance with one embodiment of the invention; Figure 2 is a schematic representation of an image obtained from the apparatus of Figure 1 in accordance with one embodiment of the invention; Figure 3 is a flowchart depicting a laser-light scattering analysis in accordance with one embodiment of the invention; Figure 4 is a flowchart depicting in more detail the image analysis from the laser- light scattering analysis of Figure 3 in accordance with one embodiment of the invention; Figures 5A is a simulated schematic plot of principal component values for various food samples in accordance with one embodiment of the invention; Figures 5B and 5C are simulated schematic plots showing the variation in principal component values with time for a food sample being processed or treated in accordance with one embodiment of the invention; Figure 5D is a simulated schematic plot of principal component values for various food samples having known properties in accordance with one embodiment of the invention; and Figures 6 and 7 are plots resulting from a principal component analysis performed on data obtained in an experiment on different types of drinking yoghurt.

Detailed Description

Figure 1 illustrates apparatus 5 for performing a laser-light scattering analysis in accordance with one embodiment of the invention. The apparatus includes a laser 10 for producing an output light beam 101. Apparatus 5 further includes a tray 50 which holds a sample 60. This sample generally comprises all or part of a food ingredient, a combination of food ingredients, or a food product. The sample may also represent an intermediary or catalyst used in making a food product. The laser beam 101 is directed at the surface of the sample. The laser beam 101 strikes the sample 60 in region 75 (note that the shading of region 75 in Figure 1 does not imply that region 75 is intrinsically different from the rest of the sample 60). The incident light beam 101 on sample 60 produces a reflected beam 102 and also scattered light from region 75.

Apparatus 5 further includes a camera 30. In one embodiment, camera 30 comprises a digital video camera or a standard digital camera, such as an Olympus Camedia C-5050 or C-3030. A webcam or similar device could also be used for camera 30. Camera 30 is provided with a pixel-based imaging system, such as a charge-coupled device (CCD). Camera 30 may be arranged to obtain colour or black and white images (note that light-beam 101 from laser 10 will generally be monochromatic).

In one particular experiment, an image resolution of 1280x 840 pixels was used, together with an exposure time of about 1/60 second, but it will be appreciated that these could be modified as appropriate. For example, in some cases sample 60 may be subject to some noise or vibration, perhaps as part of a manufacturing process, or potentially as part of the experimental investigation. This noise or vibration may impact the surface of sample 60. In these circumstances, the exposure time for camera 30 could be relatively short (compared to the noise or vibration timescale) to avoid or reduce any blurring of the image. It may also be desirable in such a situation to obtain multiple images (whether in parallel and/or in sequence) to try to average out the impact of any vibrations.

Camera 30 obtains an image of scattered light from region 75, as represented by light path 103. It will be appreciated that region 75 will normally scatter light in a range of directions, not simply as a single beam 103; however, for the illustrated position of camera 30 (approximately overhead region 75), only the portion of scattered light adjacent to path 103 is detected by camera 30, and hence only path 103 is relevant for present purposes.

In one embodiment, laser 10 and camera 30 are attached to a common mount 20.

Mount 20 can be used to hold laser 10 and camera 30 in a fixed configuration with respect to one another. Mount 20 allows individual adjustment of the position and orientation of laser 10 and of camera 30. Having a single mount 20 for these two devices makes it easier to control their relative position and orientation, although separate mounts could be used if so desired.

The sample to be analysed is placed on tray 50, which is supported by an adjustable height control 55. Control 55 can also be manipulated to ensure that tray 50 is in a properly horizontal position. It will be appreciated that the position and orientation of the laser 10 and camera 30 are arranged in combination with the height of the tray 50 so that camera 30 images the region of sample 60 where light beam 101 is incident, thereby allowing the scattered light to be detected by the camera.

Tray 50 may allow sample 60 to be manipulated or controlled in various ways.

For example, tray 50 may allow sample 60 to be rotated, vibrated, or tilted. Tray 50 may also allow sample 60 to be heated or cooled, or maintained at some predetermined temperature (whether hot or cold).

In one particular embodiment, laser 10 represents a diode-based laser pointer, and outputs red light with a wavelength of approximately 650 nm and a power of approximately a few milliwatts. The diameter of beam 101 is approximately a couple of millimetres. The wavelength and cross- sectional size and shape of beam 101 may be altered (either by adjusting laser 10 or replacing it with a different laser), dependent upon the type of sample 60 being investigated.

For example, it may be desirable to ensure that laser beam 101 is not too similar in colour to sample 60, since this may lead to a loss of discrimination in the image from camera 30. In some cases, the wavelength of laser beam 101 may be outside the visible range, e.g. near infrared or ultraviolet, although having a visible laser beam helps with experimental practicalities (since the laser beam can be located and hence positioned by eye). It will be appreciated that camera 30 is selected to match laser beam 101, so that the camera has a sensitivity at the wavelength of laser beam 101 (and hence of scattered beam 103).

The diameter of laser beam 101 can be selected as appropriate, for example so that laser beam 101 is sufficiently large to encompass various surface texture properties of sample 60. Thus if sample 60 comprises a foam with a characteristic bubble size, then the diameter of laser beam 101 may be arranged to significantly exceed this size.

The cross-sectional shape of laser beam 101 can be selected as appropriate, and may not be circular. For example, it may be desirable to use a laser beam with a cross- section in the shape of a line or some other shape (e.g. a U-shape, a V- shape, and so on).

The shape of the laser beam may be selected based on various criteria, such as apparatus design, the form of sample, and so on, as discussed in more detail below.

In the embodiment of Figure 1, the laser 10 is located approximately 40 cm away from the sample 60. This is a convenient distance for allowing access to the various components of apparatus 5. Also in the embodiment of Figure 1, the laser beam has an angle of incidence onto the sample of approximately 40 degrees (as measured from the normal), while the camera 30 is located approximately overhead the sample 60.

Consequently, the scattered light 103 from sample 60 arrives at camera 30 in a direction approximately normal to the surface of sample 60.

The relative location of laser 10 and camera 30 ensures that the main reflected light beam 102 from the sample 60 avoids the camera. The overhead position of the camera 30 also ensures that the surface of the sample 60 is generally perpendicular to the line of sight from the camera, which helps to improve image quality. In addition, the overhead camera position minimises the travel distance through sample 60 (i.e. the optical depth) for scattered radiation emerging from region 75.

The distance from camera 30 to the sample 60 is similar to the distance from the laser to the sample, again for reasons of convenience and ease of access. Note that the distance from the camera to the sample may also be adjusted to control the image scale at the camera (the image scale may also be manipulated by the zoom function, if any, on camera 30).

Figure 2 is a simplified depiction of an image of region 75 as viewed from camera 30. The image includes a spot 200 representing the scattered light from laser beam 101.

The regions of the image outside spot 200 correspond to a normal image of the sample 60 (and are not specifically depicted in Figure 2). The laser spot 200 includes a bright central portion 201 (shown by heavy shading in Figure 2) , and a fainter or diffuse portion 202 (shown by light shading in Figure 2).

It will be appreciated that the laser spot of Figure 2 is schematic only, and does not show the full detail of a measured laser spot. For example, Figure 2 omits any texture or other form of brightness variation for laser spot 200. In addition, it will be appreciated that in practice regions 201 and 202 have fuzzy boundaries (especially outer region 202), rather than the sharp boundaries shown in Figure 2. Thus diffuse region 202 becomes increasingly faint towards its outer edge, so that the limit of region 202 is therefore defined in practice by the boundary at which region 202 can no longer be discriminated from the remainder of the sample. Furthermore, diffuse region 202 can be regarded as extending across central region 201, although in the example of Figure 2 it is only readily visible outside central region 201.

The central portion 201 represents the region where the laser beam 101 is directly incident upon the surface of sample 60. Note that central portion 201 has an elliptical shape due to the angle of incidence being about 40 degrees (assuming that the laser beam 101 has a circular cross-section). The semi-minor axis of central region 201 corresponds approximately to the radius of the light beam 101 (i.e. about one millimetre or so). The intensity of the central region 201 can be regarded as the scattering intensity that would be measured at a scattering angle determined by the position of the camera 30 (i.e. for a scattering angle defined by the angle between light beam 101 and light path 103).

The diffuse region 202 around the outside of central portion 201 represents light scattered multiple times from within the sample 60 (and more particularly from within region 75). Diffuse region 202 therefore corresponds to volume or bulk scattering, rather than to surfacescattering, and hence is probing inside sample 60 (within region 75). It will be appreciated that scattered light from within the bulk of sample 60 also emerges from the sample within central region 201 (i.e. as mentioned above, diffuse region 202 extends across inner region 201, rather than being annular in shape).

In the example shown in Figure 2, the region of diffuse light 202 is significantly bigger than the central region 201, perhaps by a factor of approximately 10, although this is very dependent upon the properties of the particular sample 60 being investigated, such as refractive index, opacity, and so on. In one particular experiment, diffuse light region 202 may extend across perhaps 30-40 mm of sample 60. Note also that the diffuse light region 202 has a broadly circular shape, in contrast to the ellipse of central region 201.

This circular shape arises naturally from multiple scattering within region 75.

For certain other samples, in particular those with very high opacity (especially certain solids), the detectable region of diffuse light 202 may be smaller than (and contained within) the region 201 of surface scattering. It will be appreciated that this represents the region of greatest density for volume scaftering. Accordingly, the diffuse region of volume scattering may range in size from being smaller than to being larger than the region of surface scattering, dependent upon the properties of the sample itself.

The size of the surface scattering region itself is dependent upon the surface properties of the sample, especially texture, as well as the geometric properties of the laser beam cross- section and orientation.

Note that the sample 60 should be arranged to have sufficient depth in tray 50 so that region 75 (where laser beam 101 penetrates, and from where the bulk scattering light emanates) does not extend to the bottom of the tray 50. Otherwise, some of the light scattered back to camera 30 may in fact have been reflected or scattered from the bottom of the tray 50, and hence the resulting image will be measuring the combined properties of tray 50 plus sample 60, rather than just sample 60 alone.

In one embodiment, camera 30 is positioned and configured so that the diffuse region 202 represents approximately one quarter to one half of the full image width available to camera 30. This enables the background of the sample 60 outside the diffuse region 202 to be clearly seen on the camera image, while at the same time providing a significant number of pixels across region 202 for image analysis purposes. In one particular embodiment, region 202 corresponds to approximately 200-300 pixels at the image plane of camera 30. It will be appreciated that increasing the number of pixels generally allows the camera to obtain a better representation of spot 200 and its surroundings.

Although the laser beam 101 has a circular cross-section in the embodiment of Figure 2, other cross-sectional shapes may be employed instead, for example to alter the shapes of central region 201 and diffuse region 202. For example, if laser beam 101 is elliptical in cross- section, then central region 201 can be arranged to appear circular (at an appropriate orientation), where the symmetry may help image analysis (see below).

Alternatively, having a laser cross-section in the form of a line (i.e. a thin rectangle) may help to provide more illumination into the diffuse region 202 outside the central region 201, since laser light incident in the middle of centre region 201 in Figure 2 will generally may make a reduced contribution to diffuse region 202. An 0-shaped (annular) cross-section for laser beam 101 may be useful for samples of high opacity, since as mentioned above, such samples tend to have a small diffuse region 202. In this case, the diffuse (volume scattered) light could then be visible in the centre of the 0, with relatively little contamination from surface-scattered light.

Table I provides a listing of some of the parameters that may be varied for the experimental implementation of Figures 1 and 2, along with some suggestion of the motivation for varying the respective parameters. Note that Table 1 is not exhaustive, and other parameters, such as the distance from the laser to the sample, and from the sample to the camera, may also be varied as appropriate.

Table 1

* angle of incidence of laser beam 101 - to adjust the size and shape of central region 201, as well as the penetration of the laser beam into the sample (i.e. the depth of region 75); * direction angle of detected scattered light 103 - moving the camera 103 from the overhead position can give an increased line of sight depth into the sample without reaching the bottom of the tray 50; * wavelength of laser beam 101 - in view of sample opacity, and general wavelength dependency of scattering; * cross-sectional size of laser beam 101 - in view of sizing of features on the surface of sample (e.g. foam size); * cross-sectional shape of laser beam 101 - in view of geometrical relationship between surface-scattered and volume-scattered light; * colour response of camera 30 - i.e. whether monochrome or colour, in view of variations in colour of laser beam 101 and sample 60; * sample conditions - e.g. temperature, vibrational and/or rotational conditions, since different samples may be impacted in different ways by changes in these conditions.

Figure 3 is a flowchart illustrating sample analysis using laser-light scattering in accordance with one embodiment of the invention. The method commences with directing a laser beam onto a sample (310), and then capturing a digital image of the scattered light from the sample (320). It will be appreciated that these operations can be performed using the apparatus illustrated in Figure 1, and the image captured will generally correspond to the schematic representation of Figure 2.

An image analysis is now performed on the digital image of the scattered light in order to extract multiple statistical parameters from the image (330). These parameters can then be compared with analogous parameters obtained from previous samples (340) (perhaps averaged into some form of template) in order to assess the sample. For example, it may be determined that the current sample belongs to a different batch from previous samples, or has some different properties from previous samples. This may impact desired features of the sample, such as taste, consistency and so on. In a real-time manufacturing environment, the comparison may also indicate when a particular process has either completed (i.e. the sample has reached a desired end-state) or has failed to complete properly, which can then be used for process control. Such a determination may be made by matching the extracted parameters on a regular basis with parameters from earlier samples.

Although the flowchart of Figure 3 has been described in the context of a single experimental configuration, the approach presented herein may easily be extended to performing multiple experiments on a given sample. These multiple experiments may be performed with the same conditions (i.e. replications, to check for consistency). It is also possible to adjust various parameters between the different experiments. For example, one or more of the parameters listed above in Table 1 can be adjusted between different measurements on a given sample, in order to investigate in more detail various aspects of the sample, and to try to provide improved sensitivity and discrimination between samples.

Figure 4 is a flowchart illustrating in more detail the image analysis of operation 330 from Figure 3 in accordance with one embodiment of the invention. In this particular embodiment, the image analysis is performed using a (gray- level) co-occurrence matrix (GLCM), which provides a form of textural image analysis.

If we represent the original image by I and the GLCM by M, where I(j,k) represents the value of matrix I for row j, column k, and M(J,K) represents the value of matrix M for row J, column K, then we can define M as follows: M(J,K)= k E Fgray (I(j,k), I(F0,(j, k))) Eq. 1 The function F0 defines a positional relationship between a first and second pixel, for example F0 (j,k) j, k+1 selects a second pixel immediately to the right of the initial (first) pixel, while F0 (j,k) =j+2, k+2 selects a second pixel diagonally two up and across from the first pixel. The function Fgray selects pixel pairs that have the relevant intensities for that position in the GLCM, so that: Fgray(I(j,k), I(F0(j, k))) 1 if I(j,k)=J and I(F0(j, k))=K, and =0 otherwise In other words, the GLCM selects a set of pixel pairs from the original image, where the two pixels in each pixel pair have a certain spatial relationship, as specified by Different positions in the GLCM then reflect the number of pixels pairs in the set that share a given combination of intensity (i.e. gray level) values.

Note that there is some variation in the art regarding the precise manner of calculation of the GLCM. For example, the intensity values may be binned or aggregated to reduce the size of the GLCM. In addition, rather than defining a single pixel, the function F0 (j,k) may define a set of pixels (e.g. all pixels that are immediate neighbours of pixel (j,k)). In this case, the value of Fgray may be determined for each pixel in the set, and then averaged across the set (prior to the summation of Equation I). Further details about calculating a GLCM can be found in "Statistical and Structural Approaches to Texture" by R Haralick, in Proceedings of the IEEE, Volume 67/5, May 1979, pages 786- 804, and "Classification of Cereal Grains using Machine Vision: III Texture Models", by Maj umdar and Jayas, in Transactions of the American Society of Agricultural Engineers, Volume 43(6), 2000, pages 1681-1687, as well as from the web-site: http://mathworks. com/accesslhelpdesklhelp/toolbox/images/enhanc 1 6.html.

The processing of Figure 4 commences with specifying an offset or separation for the pixel relationship in the GLCM (410). This corresponds to defining the function For example, if F0 (j,k) =j+N, k+N, then N can be considered as determining the offset in the pixel relationship for the GLCM. For low values of N, corresponding to a small offset, the GLCM is measuring fine texture in the image. Conversely, for large values of N, corresponding to a high offset, the GLCM is measuring coarse texture in the image.

The GLCM for the specified offset is now calculated (420), and various image parameters are determined from the GLCM (430). In one particular embodiment, the image parameters calculated correspond to those listed in Table 2 (and are formally defined in equations (4) to (13) of the abovementioned paper by Majumdar and Jayas), which is herein incorporated by reference in its entirety. Note that all of these image parameters are defined in relation to a normalised GLCM. Table 2 also gives a very approximate physical meaning for these parameters in terms of the original image, but it will be appreciated that this meaning is subsidiary to the formal mathematical definition of the parameters.

Table 2

* mean - mean intensity * variance - variance or power in image * uniformity - low uniformity of GLCM implies randomness in original image * entropy - similar to uniformity * maximum probability - regular feature on scale of pixel relationship * correlation - positive correlation can imply lack of variation on scale of pixel relationship * homogeneity - implies lack of variation (homogeneity) on scale of pixel relationship (or periodicity on scale of pixel relationship) * inertia - variation (especially periodicity of twice scale of pixel relationship) * cluster shade - high brightness regions or low brightness regions (not both) on scale of pixel relationship * cluster prominance - high contrast - high brightness regions and/or low brightness regions on scale of pixel relationship In many applications, the GLCM is calculated at just a single offset. However, in the present embodiment, the GLCM is used to calculate image parameters for multiple different offsets. Accordingly, it is determined at operation 440 whether image parameters for all offsets have been calculated yet, and if not, we return to operation 410 to select the next offset for processing.

Once all the image parameters for the desired offsets have been calculated, a statistical analysis is performed to look for interrelationships between the calculated parameters and to extract useful information (450). This processing can include analysing the image parameters as a function of offset (such as illustrated at the abovereferenced web site).

In one embodiment, the statistical analysis of operation 450 is performed via a principal component analysis (PCA) on the image parameters for the different pixel separations. PCA is a well-known statistical technique to simplify complicated data sets, and maps input data from a first set of axes, as defined by the input measurements, to a second set of mutually orthogonal axes, as defined by PCA, where the axes of the second are formed from linear combinations of the axes in the first set. The PCA axes are selected to allow as much as possible of the variation in the data set to be accounted for by as few axes as possible. The most significant new PCA axis in terms of the amount of variability for which it can account is referred to as principal component 1 (PCI), the next most significant PCA axis is referred to as PC2, and so on. In general, PCA allows most (if not all) of the variability in a data set to be represented by fewer parameters or variables (i.e. principal components) than used in the original data set. For example, in an x-y scatter plot, where x and y are generally proportional to one another, the x-y data set might be represented by a single variable (representing a principal component) defined along the line of proportionality.

The output from a PCA is therefore a set of statistical parameters comprising the principal component values for each sample: (PC 1, PC2, ...PCN) Note that N is expected to be less than the number of parameters for each sample in the input data set, since the sequence of principal components can be terminated once all the variability in the data set has been accounted for. In addition, some of the higher order principal components may be discarded if they make only a relatively small (even if formally significant) contribution to the variability.

In the present embodiment, the input to the PCA is a matrix, where each row in the matrix represents the results for one sample and one separation of the GLCM, comprising: (Offset, IP(l), IP(2),...IP(lO)) where offset represents the separation of the GLCM used to obtain the image parameters, and IP(1), IP(2)... correspond to the 10 image parameters mentioned above (mean, variance, uniformity, etc). It has been found that in many cases a relatively small number of principal components (say two or three) can account for a significant percentage of the variation in such a data set (for example, more than 70%, or perhaps more than 80%).

The principal component values therefore allow the variation within the GLCM image parameters to be represented in a more compact and manageable form. This then allows the results from different images and different samples to be more easily compared with one another.

One complexity with using PCA is that it can be difficult to fully interpret the results, given that each principal component represents a linear combination of the variables used to define the input data set. Consequently, even if it is detennined that two samples are separated along a particular principal component, it may not be easy to appreciate exactly what this implies for the relative properties of the two samples. In other words, although using a PCA representation of the data can help to identify significant aspects of the data, it can be difficult to go back from such an identified aspect to some meaningful property (or properties) of the underlying data, given the more complex definitions of the principal components.

It will be noted in the present context however, that the input variables to the PCA are already one stage removed from physical properties of the sample. Thus the pattern of scattered laser light from a food sample, such as shown in Figure 2, is a complex function of the physical (and hence perceptual properties) of the food sample, such as taste, consistency, and so on. Consequently, it is already difficult to go directly back from the image parameters listed in Table 2 (for example) to the properties of the food sample per Se. As a result, the use of PCA does not in practice obscure the relationship between the observed parameters and the physical properties of interest, since it is Is generally just as difficult to determine a relationship between the underlying food properties and the PCA input as it is to determine a relationship between the underlying food properties and the PCA output.

Some implementations of the present invention may exploit the principal components determined at operation 450 directly - i.e. without any immediate attempt to relate them back to specific (known) properties of the underlying food samples. For example, Figure 5A is a schematic diagram showing the results of a principal component analysis in respect of six different samples labelled Si, S2, S3, S4, S5, and S6. It is assumed that these samples can be represented by two main principal components, specified as PCi and PC2. Figure 5A illustrates the position of each sample, as defined by the values of PCi and PC2 for that particular sample. In Figure 5A, it can be seen that samples SI, S2, S3, S5, and S6 cluster together in the PC1-PC2 plot, while sample S4 is separate from these other samples. This indicates that sample S4 is somehow anomalous, for example, sample S4 might be from a different manufacturing batch, or it might have altered in transit and/or storage.

The separate positioning of sample S4 compared to the other samples within Figure 5A can be determined either by visual inspection of the PC1- PC2 plot, or by various known statistical computational techniques for assessing clustering. Such techniques would reveal that sample S4 is an outlier compared to the other samples.

Another possibility is to define a region of the PCI-PC2 plot, shown in Figure 5A by dotted circle 501A, that represents "normal" or acceptable samples. Circle 501A may defined by performing a clustering analysis on the data set shown in Figure 5A, or alternatively circle 501 A might have been defined previously, based on measurements of other samples prior to Si, S2, S3, S4, S5, and S6. In either case, it can be seen from Figure 5A that only S4 of the current set of samples lies outside the zone of normality 501A.

It will be appreciated that the character or nature of the difference between sample S4 and the remaining samples is not immediately known from the PC 1 -PC2 plot of Figure 5A, and may turn out to be advantageous or disadvantageous (or neutral). In other words, without further information, Figure 5A only demonstrates that sample S4 is different from the other samples, but does not provide an information as to whether this difference represents an improvement or a degradation. This reflects the fact that the detected difference is directly related to laser light scattering properties rather than to underlying food properties (the former is assumed to be related to the latter, but in a complex manner that is not fully understood).

Nevertheless, in many circumstances, simply knowing whether or not a sample matches other samples can be useful. For example, in a manufacturing process, the laser light scattering technique may be used to monitor each food item as it is produced. If any food item is found to lie outside the zone of acceptance (such as circle 501A in Figure 5A), this may be used to trigger a further investigation of the anomalous item and/or the current state of the manufacturing apparatus. For example, the food item may be examined by microscope, and such further investigation may help to understand the anomaly in terms of the physical properties of the item found to lie outside the boundaries of acceptance.

Figure 5B schematically illustrates the use of the PC1-PC2 plot to monitor a manufacturing process. In this diagram it is assumed that a chronological sequence of measurements is performed at times Ti, T2, T3, T4, T5, and T6 upon a single food sample that is undergoing some processing, such as fermentation, heating, the addition of another ingredient, and so on. The circular region 501 B depicts an acceptable status for the sample once this process has finished, and is assumed to have been determined by locating previous acceptable samples in the PC1PC2 plot of Figure 5B. Accordingly, the manufacturing control process repeatedly performs laser-light scattering measurements on the food sample, and determines the location of the sample on the PCi- PC2 plot. When the sample falls within region 501B, it is known that the process has successfully completed and can therefore be terminated accordingly, such as by stopping heating, removal of a fermentation enzyme, or stopping the addition of any further ingredient (depending upon the details of the particular process that is being monitored and controlled). Alternatively, if the food sample does not enter region 501B, then this may indicate some anomaly or flaw with the processing, and require further investigation.

One possibility is that the food sample may have to be discarded, or submitted to some additional treatment.

Although Figure 5B depicts an end point region for processing, in other implementations region 501 may represent a trajectory that the sample should follow through the PC1-PC2 plot. This situation is illustrated in Figure SC, in which Ti, T2, T8 again represent a chronological sequence of measurements upon a first food sample, and T*l, T*2, ...T*8 represents the same sequence of measurements upon a second food sample. In this case the monitoring may ensure that the sample starts off in region 501C (at time Ti), progresses through region 501D (after having left region 501C), and finally arrives at region 501E. From Figure SC, it will be seen that the first food sample follows the desired trajectory, while the second sample does not, in that the measurements at T*5, T*6, and T*7 lie outside region 501D (despite arriving at the desired endpoint region 501 E forT*8). This may then lead to rejection or further investigation or treatment of the second sample and/or the manufacturing apparatus.

It will be appreciated that in some implementations the processing or treatment may be terminated as soon as there is a deviation from a desired trajectory. For example, with the second sample in Figure 5C, treatment may be terminated after the measurement at T*4, since it is already known at this stage that the sample will be anomalous.

Alternatively, it may be desirable to complete treatment, such as shown in Figure 5C.

This may be easier from a process point of view (i.e. all samples undergo the same full batch processing). It is also possible that further investigations will show that the food item being processed is acceptable, despite the excursion outside region 501D.

As previously indicated, there is a complex inter-relationship between the principal component values extracted from the laser-light scattering measurements and the physical, chemical and perceptual properties of a food sample, such as acidity, ingredient mix, sweetness, consistency, and so on. In some circumstances, it is possible that a difference detected from PCA is not significant in practical terms for the food product; for example, the difference might arise from small variations in particle size that are found not to be significant in terms of food properties. Alternatively, there may be certain differences in food properties, such as acidity perhaps, that are not necessarily revealed from the measurement procedure shown in Figures 3 and 4.

It may be desirable to perform some form of calibration of a PCA, such as by making measurements on a range of samples having known properties. The results of the PCA for these samples can then be correlated or assessed against the known properties of the samples. Such a calibration is illustrated in Figure 5D, in which sixteen different samples are plotted. The samples are arranged in four different groups, denoted AB, AY, XB, and XY, with four samples (numbered 1 to 4) within each group. The first letter in the group identifier (A or X) can be considered as identifying whether a first property (such as acidity) is high or low, while the second letter in the group identifier (B or Y) can be considered as identifying whether a second property (such as fat component) is high or low.

Figure 5D illustrates a PCI -PC2 plot of the different groups of samples, and shows that they are generally clustered together. Consequently, any further sample could be assigned to a group based on its position in the PC 1 -PC2 plot, AB for region 51 OA, XB for region 51 OE, XY for region 51 OD, and AY for regions SlOB and 51 OC. This would then allow a deduction to be made about the value of the first and second properties for the further samples.

Note that in the example of Figure 5D, the second property (as specified B or Y) is generally correlated with the value of PCI, so that a low value of PC 1 corresponds to a B sample, and a high value of PCi corresponds to a Y sample. The distribution of the first property (as specified A or X) on the PC 1 -PC2 plot is more complex however, since there is a strong dependency on the value of the second property (B or Y).

Note that the samples for group S*AY are split into two regions 51 OB and 51 OC in the PC1-PC2 plot of Figure 5D. This split would generally represent some distinction between the samples, so that samples S*AY1 and S*AY2 would belong to a first subgroup, while samples S*AY3 and S*AY4 would belong to a second subgroup. This divergence may represent some variation in the processing for S*AY samples that has not hitherto been recognised. In other words, all the S*AY samples might previously have been considered homogeneous, whereas from Figure 5D this is clearly not the case (although the difference between the two subgroups may not be significant from a food property perspective - e.g. no discernible taste difference for humans).

It will be appreciated that the PC1-PC2 plots of Figure 5A-5D represent simulated rather than real data, and are provided for illustrative purposes. Accordingly, they all depict two principal components for ease of representation, although PCA may lead to a different number of significant principal components. In addition, the regions of "normality" arising from the PCA, such as region 501A in Figure 5A, have generally been depicted as circular for ease of representation. However, it will be appreciated that a more formal statistical specification for clustering and outlier identification may be adopted in practice.

Although the above analysis has been conducted using a GLCM and PCA, it will be appreciated that many other techniques could be used for analysing the image of laser scattered light, such as shown in Figure 2. Thus rather than using a GLCM, other image analysis techniques such as Fourier decomposition or autoregression could be employed instead - the above-mentioned paper by Haralick discusses some of the known available approaches for textural image analysis. Similarly, in some analyses, it may be desired to use the derived image parameters directly, rather than through a PCA. For example, if it were found that certain of the parameters listed above in Table 2 were highly discriminatory of images to be processed, then these parameters might be monitored directly, without recourse to PCA.

In another embodiment, rather than comparing statistical parameters extracted from an image, it may be possible to perform a direct statistical comparison between images. For example, a cross-correlation might be performed between an image from a food sample underinvestigation, and an existing image from a "normal" food sample. A high correlation between these images might then imply that the food sample under investigation is a good match to the "normal" food sample, and so could be accepted.

This method could be extended by having multiple images from previous food samples, and either performing a cross-correlation against each of these earlier images, or using the multiple images to derive some form of average template for cross-correlation against new samples.

Note that the experimental configuration such as shown in Figure 1 for the laser- light scattering analysis is generally maintained as consistent as possible between measurements. This then ensures that any variations in the size and shape of spot 200 reflect the intrinsic properties of the sample itself, rather than any changes in the experimental configuration. For example, if different samples are being investigated, then by using the same laser and camera positioning and orientation, the spots observed from the different samples should be directly comparable. In certain circumstances however, some variation in experimental configuration may be unavoidable. For example, some samples may be especially large or small, and this may then lead to a variable distance between the camera and the sample. In such circumstances, it may be desirable to normalise the observed spot to a fixed imaging distance, whereby a given spot size always corresponds to the same number of pixels in the output image. In some implementations it may be possible to make such an adjustment prior to image acquisition (e.g. by using a zoom function in camera 30) to compensate for the change in image size, while in other implementations the adjustment may be made on the image data itself, such as by using known pixel re-sampling techniques to adjust the image scale.

Likewise, if the sample surface is tilted - i.e. no longer perpendicular to the line of sight to camera 30, this will cause the spot image to be elongated. The resulting spot image 200 can be transformed using standard image processing techniques to remove this elongation (note that such transformation is generally simpler if the camera is aligned so that any elongation is along either the x or y pixel axis).

Of course, if different configurations are employed for different measurements as part of the experimental protocol, such as assessing a single sample at two or more angles of incidence for laser beam 101, then the changed configuration will itself lead to different spot properties. It is possible to allow for this change by including the configuration parameters in the PCA. In other words, the input variables to the PCA might now comprise offset plus the ten image parameters of Table 2, combined with the angle of incidence used for the measurement concerned.

The experiments so far described used only a single laser beam 101. However, it will be appreciated that in other embodiments, two or more laser beams may be used, each generating its own laser spot 200 of scattered light. In some configurations the different spots may be separately imaged, while in other embodiments camera 30 may be able to record multiple spots within a single image (especially with the increasing availability of cameras with very large numbers of pixels). These multiple spots may be used for assessing the homogeneity of a single sample - i.e. do all the images for the sample look the same. The multiple spots may also be used to provide a more accurate overall measurement for a single sample, by averaging results for the sample across the multiple spots.

The laser-light scattering analysis described herein has many advantages over other forms of analysis, including: flexible: the analysis can be performed on a very wide range of samples, whether liquid or solid, including paste, foam, gel, etc, and for a wide range of sample sizes. The samples do not need to be specially prepared (e.g. diluted or sliced), but rather can be analysed in their normal condition.

non-invasive: the analysis can be performed without having to contact the food sample. This is very important for food production, since it maintains hygiene standards, and avoids any risk of sample contamination. In addition, analysis may be performed even if the sample is maintained in a protective environment, for example by using a window into the environment to allow the ingress of laser beam 101 and the egress of scattered light 103.

non-destructive: the analysis does not destroy or otherwise alter the sample, since the level of light radiation involved is very low (and only needs to be applied for a very short time).

cost: the analysis can be performed using a laser pointer, a digital camera, and a computer, all components that are readily available without undue expenditure.

speed: the time taken to acquire an image is very quick, and the results can be obtained largely in real-time, assuming that the image is automatically downloaded from the camera and analysed straightaway. This fast response time allows the analysis to follow changing sample conditions, for example due to heating, fermentation, and so on, and enables the results to be used for real-time process control.

quantitative: the results of analysing a sample can be represented by just a few numbers (say two or three principal components). This facilitates easy decision-making in respect of a sample (such as "is it acceptable ?") in contrast to more qualitative approaches, such as microscopy.

sensitivity: the analysis has been found to provide a reliable, accurate, and sensitive investigation tool for food samples, and is able to discern relatively small variations between different food samples. In addition, the laser-light scattering analysis is sensitive to multiple physical properties of a sample, and hence obviates the need to make separate measurements of these various properties.

The above advantages make laser-light scattering analysis a valuable tool across a wide range of applications. For example, in an experimental or laboratory setting, the approach described herein can be used as a diagnostic technique. For example, if tests are being performed to determine how long or in what conditions a food sample can be kept without degrading, the laser-light scattering analysis might be employed to differentiate those samples that have degraded from those that have not. The laser-light scattering analysis might also be used to gain a better understanding of dynamic processes, such as fermentation, by monitoring sample changes during such a process.

The laser-light scattering analysis can also be used in a production environment for assessing food sample quality. These quality checks might be performed on food ingredients (i.e. inputs to the production process), final products (whether for shipment to a consumer or to another company for further processing), as well as food samples at any intermediate stages. The quality checks might also be performed on items such as enzymes used in the food production process, even if such items are not directly included in the final food product. In this context, the laserlight scattering analysis can determine whether or not the relevant food samples conform to expected standards.

Note that since the laser-light scattering analysis is non-invasive and non- destructive, the analysed food items do not have to be removed from the production process. Moreover, since the analysis technique is quick and cheap to implement, it can be applied to a substantial portion (potentially all) of the relevant food items. For example, every food product from a manufacturing process might be tested using the laserlight scattering approach. Such testing can replace or supplement other forms of testing that may only be able to be performed on a subset, sometimes very small, of the output products (such as a human taste test).

As well as sample analysis at certain fixed stages within a food production process (such as for inputs and outputs of the process), the laser-light scattering analysis may also be performed on a dynamic basis to monitor the food production process. In some situations the laserlight scattering analysis may represent a passive tool to monitor the process, such as described above with reference to Figure 5C, for example, to verify that the process remains within set parameters. Alternatively, the laser-light scattering analysis may represent a facility to actively control the process, whereby the results of the analysis are used to determine actions that are performed in the production process. For example, as suggested with respect to Figure 5B, some production process such as heating, ingredient adding, etc, may be initiated, modified andlor terminated dependent upon (real-time) results from the laser-light scattering analysis - e.g. the production process may alter the relative addition rates of two ingredients, based on the ongoing results from the laser-light analysis technique. It will be appreciated that in some implementations, the laser-light analysis technique may be used both for active control of a process, and also for monitoring the process to ensure that the sample properties do not go outside predetermined limits.

Table 3 presents some of the potential measurement applications for the laser light scattering analysis technique described herein:

Table 3

protein particle size air bubbles water droplets physical particle shape parameters refractive index concentration rheology dynamic manufacturing flow in pipes (process) measurements I enzymatic kinetics __________________________ fermentation temperature dissolution of hydrochiorides time shear enzymes temperature gas formation bread baking viscosity change food gelatinization of starch fermentations dough, rising oxidation of fat foam stability ice cream ice crystals air cells meat (age and tenderness) bloom of chocolate Juice stability yoghurt base size powder properties ____________________ final product shape flour quality viscosity cosmetics emulsion stability foam, air cell size fats and oils crystal structure _________________________ __________________________ __________________ plastics It will be appreciated that Table 3 is provided by way of illustration only, and the laser-light scattering approach described herein is not limited to these particular applications. In addition, the skilled person will be aware of many further potential experimental configurations. For example, one possibility would be to have a probe such as an optical fibre that enters the sample, and so provides a light source inside the sample.

This would then avoid any surface scattered light. Fibre optics might also be used for detection of the scattered light at many well-defined locations if appropriate.

A further possibility is to use polarised light for the scattering analysis. The polarisation of the incident beam 101 onto sample 60 can be controlled by inserting a polariser into laser beam 101 (N.B. the output from laser 10 may already be polarised, although some lasers have a very short coherence length). Likewise, the polarisation of scattered light travelling along path 103 to camera 30 can be controlled by inserting a polariser into light path 103. Images of polarised scattered light may then be acquired at various orientations of the polariser(s). For example, a polariser in beam 101 perpendicular to a polariser in light path 103 can be used to determine the amount of depolarisation as a result of the scattering. Images obtained using polarised light should be particularly useful for investigating samples containing crystals, fibres, and so on.

is A further possibility is to use a video camera for camera 30 to follow any changes in sample 60. Note that in some cases the laser image may change relatively quickly, for example if one food ingredient is being added to another. Rather than using a video camera, this situation could also be followed by using a still camera to take a sequence of pictures with a relatively short exposure time.

Example: Different formulations of Drinking Yoghurt An investigation was performed of yoghurt bases made from fresh pasteurised milk bought from Arla Foods a.m.b.a (see www.arlafoods.com). The milk was re- pasteurised at 95 C for 10 minutes, cooled to 42 C, inoculated with yoghurt culture Yomix NM 1-20 (available from Danisco AIS), and incubated to pH 4.2 +1-0.1 at 42 C. The curd was cooled down to 10 C and broken by agitation.

A stabilizer was dissolved in water at 80 C and cooled to 40 C. The stabiliser and the remaining ingredients were added to the yoghurt base and dispersed with a mixer from Silverson Machines Ltd (see www.silverson. com) for 5 minutes. The pH was adjusted to 4.0 by using citric acid. Finally, the drinking yoghurt was homogenised at 300 barsl60 C, pasteurised at 90 C/15 seconds, cooled to 10 C, and filled into bottles.

Table 4 provides information about 9 yoghurt drinks that were produced, and which are labelled from 19 to 27.

Table 4

Ingredients 19 20 21 22 23 24 25 26 27 Water (Tap) 0.00 0.00 10.00 0.00 10.00 56.30 55.00 0.00 55.00 Yoghurt base 0.00 89.60 0.00 0.00 89.60 33. 30 0.00 89.20 0.00 (9% MSNF) Yoghurt 1,5% 0.00 0.00 89.10 89.20 0.00 0.00 31.90 0.00 31.90 fat Yoghurt 3,5 % 89.60 0.00 0.00 0.00 0.00 0.00 0.00 0. 00 0.00 fat Cream 38 % fat 0.00 0.00 0.45 0.45 0.00 0.00 2.70 0.00 2.70 Sucrose 10.00 10.00 0.00 10.00 0.00 10.00 10.00 10.00 10.00 Aspartame 0. 00 0.00 0.049 0.00 0.049 0.00 0.00 0.00 0.00 (sweetener) Pectin AMD 0.00 0.40 0.00 0.00 0.40 0.00 0.00 0.00 0.40 GRINDSTED 0.00 0.00 0.40 0.00 0.00 0.40 0.40 0.00 0.00 Pectin AMD GRINDSTED 0.40 0.00 0.00 0.40 0.00 0.00 0.00 0.00 0.00 Pectin AMD Starch 380 ("1" 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 choise) Total 100.0 100.0 99.99 100.1 100.0 100.0 100.0 100.0 100.0 percentage Table 5 provides further information about the composition of the 9 yoghurt drinks that were produced.

Table 5

Calculations 19 20 21 22 23 24 25 26 27 Total Fat 3.14 0.11 1.51 1.51 0. 11 0.04 1.50 0.11 1.50 Total MSNF 7.97 8.06 7.95 7.96 8.06 3.00 2.99 8.03 2.99 TotalDryMatter 21.46 18.51 11.65 21.61 8.56 13.38 15.48 18.88 15.48 TotalSugar 14.18 14.21 4.17 14.17 421 11.62 11.60 1499 11.55 Total Protein 3.05 3.14 3.13 3.13 3.14 1.17 1.18 3.12 1.18 The laser-light scattering evaluation of the drinking yoghurts was performed in a petri dish (diameter of 10 cm, 12 mm depth). Between 5 and 9 laser dot images were obtained for each sample, and the images were processed to produce the grey level co- occurrence matrix described above. The results from the image analysis were then subject to a principal component analysis using the Unscrambler program v8.0, available from Camo Process AS of Norway (see www.camo.com).

All data was centred and weighted with 1/(standard deviation) to produce the PC1vPC2 plot of Figure 6. It can be seen from Figure 6 that there is a good degree of discrimination between the different yoghurt drinks, although drinks 25 and 27 are clustered together, and likewise drinks 20 and 23. In addition, there is clear a tendency for yoghurts with high milk solids non fat (MSNF) or fat containing drinks to have a relatively high PCi score. The drink stabilised with starch (#26) is clearly distinguished from pectin-stabilised drinks (the rest). It is also noted that the standard deviation for samples of the pectin drinks is smaller than the standard deviation for the samples of the starch-containing drink (based on how big a circle is necessary to encompass each point representing a sample of a given drinking yoghurt).

The PCA suggested that PCi through PC4 are statistically significant, and can be used to discriminate between the different drinks. This is shown by the two plots of PC1vPC3 and PC1vPC4 in Figure 7. Note that drinks 25 and 27 are clearly separated in the PC1vPC3 plot, while drinks 20 and 23 are separated in the PC1vPC4 plot.

Accordingly, this experiment demonstrates the ability of the laser-light scattering analysis described herein to discriminate between food samples. In addition, these results show that this approach is very good for the prediction of the sensory properties of food (which depend on factors such as fat content, sugar content, and so on).

In conclusion, although a variety of particular embodiments have been described in detail herein, it will be appreciated that this is by way of illustration only, and the skilled person will be aware of many possible modifications to the techniques described above that will fall within the scope of the invention as defined by the claims and their equivalents.

Claims

Claims 1. A method of analysing a food sample using light scattering by:

directing a light beam at the food sample; obtaining a digital image of light scattered from the food sample; and performing a statistical analysis on said digital image to extract one or more parameters, wherein said one or more extracted parameters may be used to assess the sample.
2. The method of claim 1, wherein said light beam is produced by a laser.
3. The method of claim 1 or 2, wherein said light beam comprises visible radiation.
4. The method of any preceding claim, wherein said light beam has a crosssectional diameter of at least 0.5 mm.
5. The method of claim 4, wherein said light beam has a cross-sectional diameter in the range 1-10mm.
6. The method of claim 5, wherein said light beam has a cross-sectional diameter in the range 1.5-3 mm.
7. The method of any preceding claim, wherein said light beam has an angle of incidence onto the surface of said sample in the range 30 to 60 degrees.
8. The method of claim 7, wherein said light beam has an angle of incidence onto the surface of said sample in the range 40 to 50 degrees.
9. The method of any preceding claim, wherein said image is obtained from light scattered within 10 degrees of the normal to the surface of said sample.
10. The method of any preceding claim, wherein said image is obtained from light scattered within a cone of 10 degrees apex or less.
11. The method of claim 10, wherein said image is obtained from light scattered within a cone of 4 degrees apex or less.
12. The method of any preceding claim, wherein said image of scattered light comprises a first portion and a second portion.
13. The method of claim 12, wherein said second portion surrounds said first portion.
14. The method of claim 12 or 13, wherein said first portion comprises light scattered from the surface of the sample.
15. The method of any of claims 12 to 14, wherein said second portion comprises light scattered from within the sample.
16. The method of any preceding claim, wherein said scattered light extends across at least one quarter of said image.
17. The method of any preceding claim, wherein said scattered light extends across less than three-quarters of said image.
18. The method of any preceding claim, wherein said scattered light extends across at least 50 pixels of the digital image.
19. The method of claim 18, wherein said scattered light extends across at least 200 pixels of the digital image.
20. The method of any preceding claim, wherein said statistical analysis comprises performing a textural analysis on the image.
21. The method of claim 20, wherein said textural analysis is performed at multiple image scales.
22. The method of claim 20 or 21, wherein said textural analysis is performed using a gray level co-occurrence matrix.
23. The method of any preceding claim, wherein said statistical analysis comprises performing a principal component analysis.
24. The method of claim 23, wherein said extracted parameters comprise one or more principal components extracted from said principal component analysis.
25. The method of claim 23 or 24, wherein said principal component analysis is performed on image parameters obtained using a gray level cooccurrence matrix.
26. The method of any preceding claim, further comprising performing multiple measurements.
27. The method of claim 26, wherein said multiple measurements are performed at successive times to follow the food sample through a dynamic change.
28. The method of claim 27, wherein said dynamic change comprises heating the sample.
29. The method of claim 27, wherein said dynamic change comprises fermenting the sample.
30. The method of claim 27, wherein said dynamic change comprises adding a food ingredient to the sample.
31. The method of any of claims 27 to 30, wherein said multiple measurements are used to control a process performed on the food sample.
32. The method of claim 31, wherein said multiple measurements control an endpoint of the dynamic change.
33. The method of claim 31 or 32, wherein said multiple measurements control a trajectory of the dynamic change.
34. The method of any of claims 27 to 33, wherein said multiple measurements are performed as part of a manufacturing process for a food product.
35. The method of any of claims 27 to 33, wherein said dynamic change occurs as part of the degradation of a food product.
36. The method of claim 26, wherein said multiple measurements are performed with one or more different experimental parameters for different measurements.
37. The method of claim 36, wherein said one or more different experimental parameters comprise the wavelength of said light beam.
38. The method of claim 36, wherein said one or more different experimental parameters comprise the angle of incidence of said light beam.
39. The method of claim 36, wherein said one or more different experimental parameters comprise the polarisation of said light beam.
40. The method of any preceding claim, further comprising using at least one optical fibre to direct the light beam at the food sample.
41. The method of claim 40, wherein said optical fibre is inserted into the food sample.
42. The method of any preceding claim, further comprising directing multiple light beams at the food sample, and imaging light scattered from the food sample for each light beam.
43. The method of claim 42, wherein the light scattered from the food sample for each light beam is captured on a single image for all light beams.
44. The method of any preceding claim, wherein said method is performed in situ in a production environment.
45. The use of a method of any preceding claim for investigating one or more physical, chemical or sensory parameters of a food sample.
46. The use of a method as claimed in any of claims 1 to 44 for performing quality control on a food sample.
47. The use of a method as claimed in any of claims 1 to 44 for manufacturing a food product.
48. A food product manufactured using the method of any of claims ito 44.
49. Apparatus for analysing a food sample using light scattering comprising: a light source for directing a light beam at the food sample; an imaging device for obtaining a digital image of light scattered from the food sample; and a computational device for performing a statistical analysis on said digital image to extract one or more parameters, wherein said one or more extracted parameters may be used to assess the sample.
50. The apparatus of claim 49, wherein said light beam is produced by a laser.
51. The apparatus of claim 49 or 50, wherein said light beam comprises visible radiation.
52. The apparatus of any of claims 49 to 51, wherein said light beam has a cross- sectional diameter of at least 0.5 mm.
53. The apparatus of claim 52, wherein said light beam has a crosssectional diameter in the range 1-10 mm.
54. The apparatus of claim 53, wherein said light beam has a crosssectional diameter in the range 1.5-3 mm.
55. The apparatus of any of claims 49 to 54, wherein said light beam has an angle of incidence onto the surface of said sample in the range 30 to 60 degrees.
56. The apparatus of claim 55, wherein said light beam has an angle of incidence onto the surface of said sample in the range 40 to 50 degrees.
57. The apparatus of any of claims 49 to 56, wherein said image is obtained from light scattered within 10 degrees of the normal to the surface of said sample.
58. The apparatus of any of claims 49 to 57, wherein said image is obtained from light scattered within a cone of 10 degrees apex or less.
59. The apparatus of claim 58, wherein said image is obtained from light scattered within a cone of 4 degrees apex or less.
60. The apparatus of any of claims 49 to 59, wherein said scattered light extends across at least 50 pixels of the digital image.
61. The apparatus of claim 60, wherein said scattered light extends across at least 200 pixels of the digital image.
62. The apparatus of any of claims 49 to 61, further comprising at least one optical fibre to direct the light beam at the food sample.
63. The apparatus of claim 62, wherein said optical fibre is to be inserted into the food sample.
64. The apparatus of any of claims 49 to 63, wherein multiple light beams are directed in use at the food sample, and for each light beam a digital image is obtained from light scattered from the food sample.
65. The apparatus of claim 64, wherein the light scattered from the food sample for each light beam is captured on a single image for all light beams.
66. A method for analysing a food sample substantially as described herein with reference to the accompanying drawings.
67. Apparatus for analysing a food sample substantially as described herein with reference to the accompanying drawings.