FROTH BUBBLES

FIELD OF THE INVENTION

The present invention relates to froth flotation and in particular, to measurement of reflectivity of flotation froth bubbles. The invention is particularly useful in mineral processing, but its application is not limited to this field.

DEFINITIONS

“Colour” refers not to the human perception of light intensities in red, green and blue wavelength bands, but in general, relative intensity components as measurable by an apparatus in two or more wavelength bands.

“Saturated” means that the relative intensity of one or more of the wavelength bands in a light measurement is out of the range of the instrument measuring the light.

“Image” means a collection of light measurements at more than one spatial position, including measurement positions that are mapped or represented in a data cube, rectangular area, circular area, straight line or curved line.

“Pixel” means the measurement result from one of the spatial positions represented by an image, containing one or more colours.

“Measurement point” means one or more pixels measured from a localized spatial position significantly smaller than a total image.

“Multispectral image” refers to an image composed of two or more wavelength bands, which could include red-green-blue colour images and/or hyper-spectral image data cubes.

“Multispectral camera” or “Multispectral light sensor” refers to any camera or spectrometer system that can produce a multispectral image, whether by use of sensors sensitive in specific wavelength bands, use of optical filters, varying of illumination wavelength intensities, or use of different illumination sources with differing wavelength bands.

BACKGROUND

Froth flotation is a widely used method used to separate hydrophobic materials from hydrophilic. This is used in the mining, recycling and waste-water treatment industries. In the mining industry it is used for separating and concentrating the valuable components of an ore from waste, with the objective of producing a grade of mineral concentrate suitable for feeding to pyro-metallurgical or hydro-metallurgical processes. It is an efficient and thus and attractive method used to concentrate valuable minerals in low-grade ore resources.

The flotation process requires mixing with air and the addition of chemicals to render target materials hydrophobic. These hydrophobic materials then attach to air bubbles, which rise to the froth layer on the top of the slurry, where the concentrate materials are collected. The control and modelling of flotation processes are challenging due to their multi-variable nature. Flotation processes are influenced by many factors, such as reagent doses of chemicals, air flow, feed material grade, particle size, etc.

In the mining industry it is widely known that the froth characteristics such as the human perception of colour is closely related to process variables such as mineral grade (concentration). In the past, the control of flotation processes depended heavily on human operators’ various experience by viewing the visual appearance of the froth.

As a soft measurement in mineral flotation process, cameras with image analysis based on several image processing techniques can provide quantitative measures of froth characteristics, objective description of froth appearance and nonintrusive real time monitoring and process control with little maintenance.

More recent developments in digital image processing and processing hardware provided opportunities to gain further understanding of the industrial flotation process. Typical image processing sensors include red-green-blue (“RGB”) colour cameras, or multispectral / hyper-spectral cameras that produce several images (one for each wavelength range) or spectrophotometers, that produce spectra from measurement points located on the froth surface.

However, use of multispectral images to monitor froth colour to infer and control mineral grade or concentration is currently known to be hampered by some practical challenges, including:

1 . The sensitivity required to measure small changes in average colour due to: a) small colour differences between some minerals; and b) low presence of some crucial minerals, for example gangue minerals in cleaner flotation cells;

2. A resulting requirement of accurate spectral calibration to compensate white balance for system changes over time; and

3. The occurrence of bright spots of specular reflection on bubbles.

The lack of reliable, accurate online sensors to provide objective machine measurement of flotation froth colour poses difficulties and consequently, many operational changes are still made by operators based on subjective visual appearance of the froth together with other measurements and their experience.

To quantify mineral grade, the absolute intensity (brightness) of each wavelength band is of less importance than the relative intensity between the wavelength bands - i.e. , the “colour”, as defined above. Even though measured brightness is dependent on mineral concentration, it is also directly dependent on the measurement system sensitivity and the highly variant flux density at each measurement point, which is a function of the illumination source’s spectral flux at the specific time, its (usually) uneven distribution over the Image as well as the variable angle between the incoming light rays and the local bubble surface gradient at each measurement point.

By contrast, measured colour is only dependent on mineral grade, the measurement system sensitivity and the illumination source’s colour. Therefore, calibrating the measurement system’s white balance to the illumination source’s colour is sufficient though crucial to infer mineral grade. RELATED ART

In prior art, spectral information of the top surface of flotation froth bubbles is measured with multispectral cameras operating in the ultraviolet, visible, and near-infrared ranges. Where different materials present in the froth shows different reflection spectra, multispectral images can potentially quantify the amount of each material present in the froth and therefore the grade of material that is being produced. In many cases, off-the-shelf three-channel RGB colour cameras are used, and the human perception of “froth colour” is used to describe the spectral reflectivity of the froth. Ideally the more channels used, the more information would be available.

In order to ignore highly variable brightness and focus on froth colour as perceived by humans, several standard conversions of the RGB space are used in prior art, including “colour vector angle”, “HSV”, “HSL”, “Hunter Lab” and “CIE L*a*b*”.

The normal changes in grade of material being produced often only causes small and subtle differences in spectral reflectivity, it has been found that repeatability and accuracy of measurements are greatly improved if multispectral cameras are regularly calibrated against white or grey reference targets (sometimes referred to as “calibration objects”) that are positioned as close as possible to the froth surface. (E.g. US US9376627 proposes determining the ratio of each spectral sample measurement to that of an illuminated 99% reflectance panel taken with the same geometry.)

In spectroscopy, it is well known that illumination sources, thin layers of dirt on exposed optical surfaces, optical components and sensors can change over time due to ageing, temperature, humidity and vibration and therefore regular instrument calibration is performed as good practice.

Therefore, the regular measurement of correctly positioned, clean grey or white reference targets to correct white balance has been found to be crucial for measurement accuracy and repeatability. Such physical reference targets have however been found not to be practical in production environments, as they should be positioned close to the bubbling froth but need to be kept perfectly clean to provide a pure grey or white reflectivity reference. Practical problems that have not been addressed in prior art, include the effect of dirt on exposed optical surfaces, stray light, and sensor dark signals, which lead to offsets in colour signals and therefore potentially false information. A true black reference would solve the problem, but it would be just as impractical to position a clean black reference close to the froth surface, as a clean white reference.

A further problem encountered when measuring spectral information is specular reflection of the illumination source from the tops of the bubbles (and in some cases the valleys between bubbles), causing glaring bright spots and (and in some cases streaks) in the images. Attempts to reduce the glare include using diffuse illumination, illumination from a low angle or a combination of a polarizing filter before the camera and positioning the camera and illumination at the air-water Brewster’s angle of ± 53 degrees. (E.g. US6727990 proposed applying illumination of the froth surface within a low range of angle.) However, such bright spots are also valuable in image processing as starting points to segment images to determine bubble size and therefore by necessity, are seldom eliminated in practice.

Other solutions proposed in the prior art do not seek to reduce the glare that result in bright spots in images, but instead propose identifying and cancelling unwanted spectra, or removing bright spots by replacing them by propagating the surrounding information using techniques akin to image inpainting.

Due to the small spectral changes, the impracticality of physically positioning calibration objects, and complication of measurements caused by bright reflections, multispectral imaging of flotation froth is currently of limited value for providing production grade information for automatic or manual process control.

The present invention seeks to provide accurate measurement of colour of flotation froth bubbles, that overcomes the shortfalls of the prior art described herein above, at least in part. SUMMARY OF THE INVENTION

According to a first aspect of the invention there is provided a method of determining the colour of flotation froth, the method comprising:

Illuminating the flotation froth with an illumination source; capturing a digital image of the flotation froth in a multispectral light sensor; and using specular reflection intensity of the illumination source from gas-liquid interfaces on flotation froth bubbles in the flotation froth as white reference to calibrate different wavelength bands (i.e. correct the white balance) of the multispectral light sensor.

The method may include the steps of: identifying bright areas in the digital image; identifying bright measurement points amongst the bright areas where none of the wavelength bands are saturated; and calibrating the different wavelength bands of the multispectral light sensor, using the colour of the bright measurement points as reference colour.

The method may include calculating an average colour of the bright measurement points and using the average colour as the reference colour.

Instead, the method may include fitting a vector through the bright measurement points in a multi-dimensional space where each dimension maps the bright measurement points’ intensities in a specific wavelength band, the direction of the resulting vector representing the colour of the illumination source to serve as the reference colour.

The illumination source may be a hard-edged illumination source and the method may include: identifying a sharp local illumination image on the digital image, that mirrors the hard-edged illumination source; identifying the sharp local illumination image as a gas-liquid Fresnel reflection; and preferably identifying the bright measurement points within the gas-liquid Fresnel reflection.

According to another aspect of the invention there is provided a method of determining the colour of flotation froth, the method comprising:

Illuminating the flotation froth with an illumination source; capturing a digital image of the flotation froth in a multispectral light sensor; identifying non-bright areas in the digital image; identifying non-bright measurement points amongst the non-bright areas; fitting a vector through the non-bright measurement points in a multi-dimensional space where each dimension maps the non-bright measurement points’ intensities in a specific wavelength band, the direction of the resulting vector representing the relative spectral intensities of the flotation froth.

The method may further include correcting the relative spectral intensities of the flotation froth by using a white reference as described herein above.

The method may include the steps of: causing the light of the illumination source to be polarized, for instance by using a polarized light source or passing light from the illumination source through a first polarizing filter before it illuminates the flotation froth; and splitting light reflected from the flotation froth in a polarizing beam splitter that is aligned with the polarization of the illumination, into crossed polarized light and parallel polarized light.

The cross polarized light and the parallel polarized light may be directed to separate multispectral light sensors.

The method may include determining a white reference from an image produced from the parallel polarized light and may include or using the white reference to correct the white balance of an image produced from the crossed polarized light. Instead, the white reference may be used to correct the white balance of the image produced from the parallel polarized light.

Instead, the method may include the steps of: causing the light of the illumination source to be polarized, for instance by using a polarized light source or passing light from the illumination source through a first polarizing filter before it illuminates the flotation froth; and receiving light reflected from the flotation froth in a polarization camera, including internal polarization filters over some pixels that are aligned parallel to the polarization of the illumination and other polarization filters over some pixels that are aligned orthogonal to the polarization of the illumination, to limit the light to crossed polarized light falling on some pixels and parallel polarized light falling on some other pixels.

According to a further aspect of the invention there is provided a method of determining the colour of flotation froth, the method comprising: causing the light of the illumination source to be polarized, for instance by using a polarized light source or passing light from an illumination source through a sending polarizing filter; illuminating the flotation froth with the polarized light; passing light reflected from the flotation froth through a receiving polarizing filter; and capturing a digital image of the flotation froth in a light sensor, from the light passed through the receiving polarizing filter; wherein the polarization of the illumination light and the receiving polarizing filter have crossed (also described as orthogonal) orientations.

The method may include circular polarizing filters for the sending and receiving polarizing filters, both of the same polarization, either both right hand polarized or alternatively both left hand polarized, to produce circularly polarized illumination and minimize bright specular reflections. The two circular polarizing filters may be one filter extending over both the illumination source and light sensor.

The method may include: capturing a first digital image of the flotation froth, without light from the illumination source; capturing a second digital image of the flotation froth with light from the illumination source; and subtracting the first digital image from the second digital image, to eliminate the effect of uncontrolled background illumination and isolate the lighting effect of the illumination source The method may include: capturing a plurality of digital images of the flotation froth in a multispectral light sensor, with a different narrow-band illumination source switched on when each of the digital images are captured, wherein a wavelength band of each of the illumination sources is narrower than a sensitivity wavelength range of the multispectral light sensor.

According to a further aspect of the invention there is provided apparatus for determining the colour of flotation froth, the apparatus comprising: an illumination source that is configured to illuminate the flotation froth; a multispectral light sensor that is configured to capture a digital image of the flotation froth; and a processor that is configured to process the digital image according to the methods of the invention described herein above.

The apparatus may include a first polarizing filter disposed between the illumination source and the flotation froth, and a polarizing beam splitter that is aligned with the polarizing filter and is configured to split light reflected from the flotation froth into crossed polarized light and parallel polarized light.

The apparatus may include two of the multispectral light sensors that are disposed to receive the crossed polarized light and parallel polarized light, respectively.

The apparatus may include a first polarizing filter disposed between the illumination source and the flotation froth, and the multispectral light sensor may be a polarization camera including internal polarization filters that are aligned parallel to the first polarizing filter and other polarization filters that are aligned orthogonal to the first polarization filter, to split the light into crossed polarized light and parallel polarized light.

The apparatus may include a first polarizing filter disposed between the illumination source and the flotation froth, and a second polarizing filter that is disposed between the flotation froth and the multispectral light sensor, the first polarizing filter and the second polarizing filter having crossed (also described as orthogonal) orientations. The apparatus may include a plurality of the illumination sources, each of the illumination sources being configured to provide a different narrow-band illumination, a wavelength band of each of the illumination sources being narrower than a sensitivity wavelength range of the multispectral light sensor.

BRIEF DESCRIPTION OF THE DRAWINGS

For a better understanding of the present invention, and to show how it may be carried into effect, the invention will now be described by way of non-limiting example, with reference to the accompanying drawings in which:

Figure 1 shows a diagrammatic representation of illumination and optical measurement of an object, in general that forms part of the prior art and that is used in some embodiments of the present invention;

Figure 2 shows a test image of chalcopyrite froth;

Figure 3 shows a histogram of colour intensities of the image of Figure 2;

Figure 4 shows a scatterplot of blue versus red intensities for each pixel in Figure 2.

Figure 5 shows a diagrammatic representation of a second embodiment of apparatus according to the present invention;

Figure 6 shows a variation on the embodiment of the present invention of Figure 5; and

Figure 7 shows an illumination source according to the present invention, containing multiple illumination sub-sources with differing wavelength bands.

DETAILED DESCRIPTION OF THE INVENTION

When measuring the colour of flotation froth bubbles, for each measured colour component intensity at each measurement point on the froth surface, the measured intensity is a function of the illumination flux density and direction, material surface orientation angles towards both the illumination and sensor directions, material composition, sensing optics, dirt/fouling, sensor response and signal processing. The illumination flux density is a sum of controlled and uncontrolled (background) illumination. This is illustrated in Figure 1 which shows an illumination source 1 proving illumination 2 to the top of flotation froth 3, background illumination 4 and the reflected light 5 that is measured by a light sensor or camera 6. What is not shown in Figure 1 , but is implicit in all apparatus described and illustrated herein, is a processor that is configured to process digital image data captured by the light sensor or camera 6.

The measurable light 5 arriving at the sensor 6 from each measurement point on the froth 3 is a sum of different sources: a) Fresnel reflections of the gas-liquid interfaces on the outside and inside of the bubble membranes; b) Diffuse reflection and fluorescence from solid particles attached to the froth 3; c) Metallic lustre from solid particles, e.g. from chalcopyrite particles in the froth 3; d) Light returned from inside the bubble where a window on the bubble with little solid material is present, which is usually darker than other areas; and e) Stray light due to optical imperfections and dirt on exposed optical surfaces.

Sources (a) and (c) are due to specular reflections, preserving polarization to a large extent, source (a) preserves the illumination colour, and source (b) scrambles polarization. The aim is to measure the colour of sources (b) and (c). Source (c) can correspond to valuable materials like metal sulphides and source (b) can correspond to impurities like organic carbon, talc and pyroxenes. Source (a) is both a hindrance to determine true material colour and a help in other image processing steps such as bubble size segmentation. Finally, source (e) can lead to false colour information if it is not corrected for. A practical problem is that source (e) is commonly caused by scattering of light from optical surfaces covered with dust from the same solids that are sought to be measured, causing varying false background signals superimposed on the useful signals.

While bright spots due to specular reflection has been treated in the prior art as a problem for colour measurement of flotation froth, the present invention uses such specular reflections to calculate an accurate white calibration reference for use in white balance correction, as well as methods and apparatus to separate the specular reflections from the colour information that it is hiding, with the result that colour measurement according to the present invention is considerably more accurate, repeatable and useful/valuable.

In most cases, white balance of multispectral images must be corrected to compensate for system changes affecting the relative signal strengths in each colour band. Such changes can be changes in illumination, fouling of optical surfaces, optical components, light sensors, analogue electronics or other causes of change in final digital readings not related to the actual changes in colour of mineral solids attached to the bubbles.

The term “useful specular reflectance measurement points” is used herein to refer to areas with maximum brightness in an image, excluding measurement points where one or more of the colour components are saturated, because saturation loses crucial information.

Figure 2 shows a test image of chalcopyrite froth, which has a slightly yellow metallic lustre, while Figure 3 shows a histogram of intensities of this image, showing a large peak around intensity 64 on the horizontal axis and a smaller peak (encircled) around intensity 224. The large peak corresponds to the average bulk intensity in the image. The second smaller peak corresponds with the small areas of brightest specular reflections in the test image of Fig 2.

The measured colour intensities are shown in Figures 3 and 4 as relative intensities on a scale of 0 to 255 - as is common practice in the art.

Figure 4 shows a scatterplot of blue versus red intensities for each pixel in Fig 2. Methods according to the present invention analyse the set of spectral intensities of each pixel as a point in a n-dimensional space, where n is the number of spectral bands and each of the point’s n coordinate values corresponds to the measured intensity in spectral band n for the pixel. For the present example, n=2 is used for two chosen spectral bands namely blue and red. The choice of two dimensions is for practical reasons, because two dimensions can be illustrated clearly in a document. However, even though only two dimensions are used in the present explanatory example, the principles are valid for more dimensions as well, because the vector principles as shown here for two dimensions can be expanded equally for more dimensions according to known linear algebra.

In all cases, the lengths of these n-dimensional vectors are closely related to intensity and therefore less important than the n-dimensional directions of these vectors. In the case of colour analysis based on human vision, the lengths are closely related to and commonly denoted as “Value” or “Lightness” in “HSV”, “HSL”, and “Lab” transformations.

By contrast, the direction of vectors represent colour, is most important and can be denoted inter alia as vector angles, relative magnitudes of vector components, or coordinates of a unit vector. In the case of colour analysis based on human vision, the direction is closely related to and commonly denoted as “hue and saturation” or “a* & b*” in “HSV”, “HSL”, and “Lab” transformations. For present purposes, reference will be made herein to vector direction, relative intensities or colour.

In Figure 4, vector 41 shows the Red-Blue pixel values of the darkest parts of the froth in the image. Vector 42 shows the pixel values for the brightest parts of the froth where no specular reflection is visible. Vector 43 shows the pixel values of the brightest parts of specular reflection on the froth.

According to the prior art, the pixel values depicted by the cloud of points between the endpoints of vectors 41 and 42 are useful for colour analysis, but the pixel values depicted by the cloud of points between the endpoints of vectors 42 and 43 should be disregarded for colour analysis, because they contain bright specular reflections.

By contrast, the present invention does not ignore the points depicted by vector 43 but use this vector as a simple, yet useful first approximation for illumination colour. According to the present invention, the brightest areas are identified using methods similar to current practice, such as local thresholding, but instead of ignoring the brightest areas, bright measurement points are identified amongst the bright areas where none of the wavelength bands are saturated, i.e. where the intensities of all the wavelength bands are within the measurable range (i.e. less than 255 as shown in figure 4). The average of the bright measurement points (vector 43) is then used as a white reference to correct the white balance of the image. This calculated value (vector 43) is a simple, yet useful representation of the illumination colour measured according to the sensitivity of the light sensor.

The present invention thus uses the specular reflection intensity of an illumination source from the gas-liquid interfaces on flotation froth bubbles as white reference to calibrate the different wavelength bands of a multispectral light sensor, also known as “correcting white balance”, without the use of a clean, white reference calibration object or target.

The specular air-water Fresnel reflections tend to give small mirror-like images of the illumination sources due to the water’s gradually changing angle, while specular solidparticle reflections tend to produce larger, poorly defined images of the illumination sources due to more random orientations of solid particle surfaces.

An illumination source with hard edges can be used to cause a sharply defined illumination local image on the image of the bubbles due to gas-liquid Fresnel reflection that can then be used to differentiate more clearly between gas-liquid Fresnel reflections and other reflections (such as the poorly defined images from solidparticle reflections), for example by using image process techniques such as local thresholding.

As a further step, the white reference determination is refined by using only the Fresnel reflections of the air-water interfaces as they preserve the illumination colour of the illumination source. By fitting a line through the cloud of points that represent pixel values of the brightest areas as determined above, a vector 44 is calculated that is a better representation than vector 43 above, of the illumination intensity and colour measured according to the sensitivity of the light sensor. This is because the vector 44 is independent of diffuse and metallic lustre reflections of solid particles on the bubbles and independent of offsets due to stray light from optics imperfections and scattered light from fouling on exposed optics surfaces. It should be noted that a general offset is apparent among the lower intensity values in Figure 4 that tends towards higher red values and lower blue values. This could be due to several reasons, including due to dust on exposed optics that absorbs more blue light than other wavelengths.

A vector (vector 44) is thus fitted through the points representing the measurement points of the highest values in the multi-dimensional space where each dimension maps measurement points’ intensities in a specific wavelength band and where the resulting vector (vector 44) is calculated as the white reference with its direction as the relative white reference.

As a further step, the present invention does not use the pixel values depicted by the cloud of points (the non-bright points), between the endpoints of vectors 41 and 42 directly for determining froth colour, as is done in the prior art, but instead fits a line through the pixel values as shown by vector 45 in Figure 4. The direction of vector 45 produces a statistically sound, more accurate measure of froth colour, which is independent of varying offsets in colour signals that are caused by dirt on exposed optical surfaces, stray light and sensor dark levels. If it were practical to insert a physical black reference target object into the image, it could have provided a similar function as this method to correct for the varying offsets in colour signals that are caused by dirt on exposed optical surfaces, stray light and sensor dark levels. However, since the use of a correctly positioned clean black reference is not practical, the use of this method of fitting a vector 45 through the pixel values is an improvement to the prior art. The white balance of the image is then corrected by using the white reference as determined in the steps described above, to provide a froth colour measurement that is considerably more accurate and repeatable than in prior art.

A vector is fitted through the non-bright measurement points in the multi-dimensional space where each dimension maps measurement points’ intensities in a specific wavelength band and where the resulting vector is calculated as the spectral intensities for a multispectral light sensor measuring flotation froth bubbles and where the vector direction is the relative spectral intensities (defined as colour). This method minimizes the offset effect of dirty optics, stray light and non-perfect sensor black level correction, without the use of a clean, black reference calibration object or target. The calculated froth colour (relative spectral intensity) is corrected for white balance by using the white reference of the previous step.

Referring to Figure 5, as a further step, a linear polarizer 7 is placed before the controlled illumination source 1 and a polarizing beam splitter 8 that is aligned with the linear polarizer 7, is used to produce crossed-polarized light to one camera 9 and parallel-polarized light to a second camera 10.

This serves several purposes. First, the image produced by camera 9 contains minimal specular reflection, which removes both the Fresnel reflections of the air-water interfaces and metallic lustre of metal sulphides, providing an image where sensitive measurements of diffuse reflecting materials can be performed, for instance naturally hydrophobic gangue minerals like organic carbon, talc and pyroxenes that may contaminate the final product and needs to be effectively controlled during froth flotation. Second, the image produced by camera 10 contains enhanced specular reflection, making the determination of the specular white reference more efficient. Third, the combination provides an extra degree of information about the froth, that is useful for other image analysis like bubble size segmentation and differences in material content on bubble surfaces versus the material in valleys between bubbles.

The present invention thus uses a first linear polarizing filter 7 between the illumination source 1 and flotation froth 3, plus a polarizing beam splitter 8 between the froth and two multispectral light sensors (cameras) 9,10 where the polarizing beam splitter is aligned with the first polarizing filter.

A white reference is determined from the image received in the camera 10 receiving the parallel polarized light. The white reference obtained is used to correct the white balance of either the image produced by the camera 9 receiving the crossed polarized light or of the image produced by the camera 10 receiving the parallel polarized light.

The suppression of specular reflection is used to enhance the sensitive measurement of diffuse reflecting materials. More specific in the case of mineral processing, the suppression of metallic lustre from sulphide minerals is used to enhance the sensitive measurement of other materials, including but not limited to naturally hydrophobic gangue minerals such as organic carbon, talc and pyroxenes.

The combination of two filters and cameras 8, 9 and 10, can also be replaced by a single polarizing camera. A first linear polarizing filter can be used between the illumination source and froth, plus a polarization camera where some of its multiple internal pixel-sized polarization filters are aligned parallel to the first filter and some are aligned orthogonal to the first filter.

A simplification of the apparatus of Figure 5 is shown in Figure 6, where the combination of beam splitter 8 and two cameras 9 and 10, can be replaced by a single polarizing filter 11 and camera 12. In this case the polarizing filters 7 and 11 are chosen and oriented to remove specular reflections (of both the Fresnel reflections of the airwater interfaces and metallic lustre of metal sulphides). Either both are linear polarizing and orthogonally positioned, or alternatively both are circularly polarized. In the latter case both are of the same polarization, namely both left-hand polarized or both right-hand polarized as viewed from the froth surface towards the filters. In this latter case, the filters 7 and 11 can be one filter that extends over the light path of both the illumination source 1 and camera 12.

This provides an image where sensitive measurements of diffuse reflecting materials can be performed for the case where white balance correction is not required.

The present invention thus uses a first polarizing filter between the illumination source and flotation froth bubbles, and a second one between the froth bubbles and monochrome or multispectral light sensor.

Linear polarizing filters are used with their orientation orthogonal (also known as crossed) to each other, to minimize bright specular reflections. Alternatively, circular polarizing filters are used, both of the same polarization, either both right hand polarized or alternatively both left hand polarized, to produce circularly polarized illumination and minimize bright specular reflections. The two circular polarizing filters can be one filter extending over both the illumination source and light sensor. It is further possible to correct for uncontrolled background illumination that may be too diffuse to cause enough useful specular reflections for its colour to be characterised. This is done by capturing a first image with the controlled illumination and a second image without the controlled illumination. The two images are subtracted from each other, resulting in an image that only contains the controlled illumination.

At least two measurements are made; one with the illumination source off or at a dim setting and one measurement with the illumination source on or at a brighter setting.

The darker Image is subtracted from the brighter Image, in order to eliminate the effect of uncontrolled background illumination and therefore to isolate the lighting effect of the controlled illumination source.

Referring to Figure 7, in another embodiment of the present invention, an illumination source 1 contains multiple illumination sub-sources 13 to provide illumination 2, where each illumination sub-source produces light over a band of wavelengths that overlaps with a part of at least one of the image sensor’s bands of sensitivity but does not cover the sensor’s complete band of sensitivity. The light sensor takes a set of more than one exposures in succession, while at least one of the illumination sources is switched off or dimmed for at least one of the total set of exposures. The light sensor can be monochrome with only one wavelength band of sensitivity, or it can be a multispectral light sensor.

In the preferred embodiment a set of 14 high brightness, narrow-band light-emitting diodes are used as sub-sources 13, covering the range from 360 nm to 1050 nm and taking 16 images with a monochrome camera set to an integration time of 20 ms per exposure, switching one LED wavelength on for each exposure and leaving all of the LEDs off for the first and last exposures, in order to produce an average background illumination reflection level that can be subtracted.

The set of 14 exposures (after background subtraction) are then processed to remove the minimal movement between images that happens with flotation froth in the fraction of a second during which the set of exposures were captured, for instance by using a block matching algorithm that searches for maximum correlation between images, or the images can be left as-is, depending on the accuracy required. The image can then be processed as a 14-colour multispectral image, providing better accuracy for colour detection than the prior art use of simple 3-colour RGB cameras and providing a lower cost solution than the use of expensive hyper-spectral cameras or 2-dimensional spectrometers.

At least two measurements are made; with a different narrow-band illumination subsource switched on during each measurement and where the wavelength bands of each illumination sub-source are narrower than the sensitivity wavelength range of the sensor, in order to produce a multispectral image.

All the calculations of this invention as described above produce better results when an image sensor system with a linear output response against input light is used; and a sensor with a linear output response to light input is preferred.

The combination of illumination source parameters, spatial positioning, filter specifications and sensor settings are chosen and adjusted to minimize (if not eliminate) saturation for any of the measurement points and any of the wavelength bands, keeping all signals higher the minimum and lower than the maximum possible/available values.