AU2005305581A1 - An apparatus for and method of sorting objects using reflectance spectroscopy - Google Patents

Description

WO 2006/054154 PCT/IB2005/003443 AN APPARATUS FOR AND METHOD OF SORTING OBJECTS USING REFLECTANCE SPECTROSCOPY BACKGROUND OF THE INVENTION THIS invention relates to an apparatus for and method of sorting granular materials or other objects of a range of sizes from 300 micron to 2 mm using reflectance spectroscopy in the visible (VIS) to the near infrared (NIR) spectral range, and in particular between the 400 nm and 1000 nm spectral range. In particular, the present invention provides two general embodiments, involving conventional reflectance spectroscopy and hyperspectral imaging techniques. Hyperspectral imaging, sometimes referred to as imaging spectrometry, differs from conventional imaging techniques in that it covers many narrowly defined spectral channels, whereas conventional imaging techniques look at several broadly defined spectral regions. Thus, the broad idea behind hyperspectral imaging is to obtain a continuous spectrum of electromagnetic radiation reflected from the surface/particles being imaged. In order to determine what the reflectance represents, the reflected spectral data obtained by the hyperspectral sensor is analysed using a classifier that has been trained on spectral data from known particles. Thus, a primary goal of using spectral data is to discriminate, classify, identify as well as quantify materials being investigated. Turning now to the primary application of the present invention, in the process of diamond exploration, soil and stream samples are treated in WO 2006/054154 PCT/IB2005/003443 -2 order to identify and extract minerals that indicate the presence of a kimberlite source, these minerals being termed kimberlitic indicator minerals. The process involves various density separation methods, yielding a heavy mineral concentrate. The size distribution of these concentrates typically lies in the range from 300 to 2000 microns. Concentrates are usually sieved into narrower size ranges, before further treatment. These are then handsorted under microscope, by highly skilled and trained laboratory staff, in order to extract the kimberlitic indicator minerals. Samples usually yield heavy mineral concentrates of the order of a hundred grams, which can consist of in excess of a million grains. The handsorting process is arduous, time consuming, and requires a large complement of appropriately skilled staff. The current invention relates to the automated sorting of heavy mineral concentrates, for specific indicator minerals including, but not limited to, garnet, chrome diopside, and olivine, yielding a reduced quantity of material that needs to be handsorted. SUMMARY OF THE INVENTION According to a first aspect of the invention there is provided a hyperspectral imaging apparatus for identifying target particles within a batch of particles, the apparatus comprising: a tray for supporting the batch of particles; leveling means for leveling the batch of particles into substantially a monolayer; a hyperspectral scanning system for scanning the batch of particles, to produce a hyperspectral image of the batch of particles; WO 2006/054154 PCT/IB2005/003443 -3 a classifier for determining the pixel coordinates of target particles in the hyperspectral image; converter means for converting the pixel coordinates to world coordinates of the target particles on the tray; and target particle extraction means for picking the target particles based on the calculated world coordinates and for transferring the picked target particles to a storage arrangement. Significantly, the size range of the particles is -2.0+0.3 mm. Preferably, the target particles are kimberlitic indicator minerals. Typically, the tray comprises a tapered rim and a ridge provided at the bottom of the tapered rim. Conveniently, the tray is part of a batch sample preparation assembly, which further includes a vibrating stage assembly comprising a spring plate and a linear electromagnetic vibrator, the electromagnetic vibrator being used to produce the required vibration in the plate so as to yield the monolayer of batch particles. The hyperspectral scanning system comprises a spectrograph and a camera, the spectrograph being arranged to produce a line element in the field of view of the camera, the spectrograph and camera being arranged to move linearly along the length of the tray so as to produce a hyperspectral image of the batch of particles. Conveniently, the hyperspectral image is stored as a three-dimensional array, comprising the two spatial dimensions of the tray and one spectral dimension.

WO 2006/054154 PCT/IB2005/003443 -4 Preferably, the apparatus is primarily aimed at the visible (VIS) to the near infrared (NIR) spectral range, with the classifier derived from a database in which reflectance spectra for target and non-target particles are stored. Conveniently, the tray comprises a plurality of calibration points that form part of a spatial calibration system, for ensuring that the world coordinates determined by the converter means coincide with the coordinate system used by the target particle extraction means. According to a second aspect of the invention there is provided a reflectance spectroscopy apparatus for identifying target particles, the apparatus comprising: a moving conveyor belt for carrying particles including target particles, the conveyor belt defining a plurality of grooves. a feed presentation sub-system comprising at least one vibratory feeder that is arranged to define a monolayer of particles and for feeding the monolayer of particles to the moving conveyor belt; an optical detection sub-system located operatively above the conveyor belt for illuminating the particles and storing the reflected spectra; a classifier for determining the location and mineral type of the particles within the grooves of the conveyor belt; and target particle extraction means for picking the target particles based on the classifier's determined location of target particles. Significantly, the size range of the particles is -2.0+0.3 mm. Preferably, the target particles are kimberlitic indicator minerals.

WO 2006/054154 PCT/IB2005/003443 -5 Preferably, the feed presentation sub-system comprises two vibratory feeders, each vibratory feeder comprising a tray for carrying the particles. A first feeder extends between a hopper and a second feeder, with the tray of the second feeder defining a plurality of grooves that are arranged to line up with the plurality of grooves in the conveyor belt. Conveniently, the feeder trays are machined from aluminum, and are sand-blasted. According to a third aspect of the invention there is provided a hyperspectral imaging method of identifying target particles within a batch of particles, the method comprising: leveling a batch of particles into substantially a monolayer; hyperspectrally scanning the batch of particles to produce a hyperspectral image of the batch of particles; determining the pixel coordinates of target particles in the hyperspectral image; converting the pixel coordinates to world coordinates of the target particles on the tray; extracting the target particles based on the calculated world coordinates; and transferring the picked target particles to a storage arrangement. Typically, the method further includes the step of vibrating the batch of particles to produce the monolayer of batch particles. According to a fourth aspect of the invention there is provided a reflectance spectroscopic method of identifying target particles comprises: WO 2006/054154 PCT/IB2005/003443 -6 providing a moving conveyor belt for carrying particles including target particles, the conveyor belt defining a plurality of grooves. providing a feed presentation sub-system comprising at least one vibratory feeder that is arranged to define a monolayer of particles and for feeding the monolayer of particles to the moving conveyor belt; providng an optical detection sub-system located operatively above the conveyor belt; illuminating the particles on the conveyor belt; storing the reflected spectra; determining the location and mineral type of the particles within the grooves of the conveyor belt; and extracting the target particles based on the determined location of target particles. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 shows a schematic view of a hyperspectral imaging apparatus according to a first, preferred embodiment of the present invention in which hyperspectral imaging is done in a batch process; Figure 2 shows a detailed side view of a batch preparation arrangement used in the apparatus shown in Figure 1; WO 2006/054154 PCT/IB2005/003443 -7 Figure 3 shows a cross-sectional side view of a tray used in the batch preparation arrangement shown in Figure 2; Figure 4 shows a detailed view of a particle picking nozzle using the apparatus shown in Figure 1; Figure 5 shows a top view of the batch preparation arrangement shown in Figure 2, illustrating the spatial calibration technique used in the present invention; Figure 6A shows a highly schematic side view of an apparatus according to a second embodiment of the present invention in which conventional spectroscopy is done in a continuous process; Figure 6B shows a schematic top view of the apparatus shown in Figure 6A; and Figure 6C shows a cross-sectional side view of a conveyor belt used in the apparatus shown in Figure 6A. DESCRIPTION OF PREFERRED EMBODIMENTS In broad terms, the present invention discloses an apparatus and method that uses visible reflectance spectral classification of individual grains in an online, automated process, for subsequent sorting. In particular, the invention uses the visible reflectance spectra of the particles being imaged in order to determine whether they are target kimberlitic or non target grains, and extracts target grains to produce a concentrate. This requires the acquisition of a pure spectrum, originating only from the surface of the single grain without interference from other grains or a background surface, from each mineral grain in the heavy mineral concentrate. A spectral classification algorithm, which allows for the WO 2006/054154 PCT/IB2005/003443 -8 identification of the specific grains of interest here, has been developed, and is described in detail later on in the specification. To this end the following criteria / guidelines were used in the design of the apparatus of the present invention: 1. No loss of mineral grains, i.e. no spillage. 2. The grains were not to be damaged. 3. The size range of the grains is -2.0+0.3 mm, divided into sub-ranges varying by at most a factor two in diameter. 4. The goal for the feedrate was 30 grains per second i.e. 1 x 105 grains per hour. 5. The goal for the discrimination of the kimberlitic grains was >85 % correctly report to the concentrate, and <10% of non-target grains report incorrectly to the concentrate. 1. First embodiment In the first, preferred embodiment shown in Figures 1 to 5, a hyperspectral imaging apparatus 10 is shown, in which the material is treated in separate batches of approximately 5 grams each, each batch being placed on a tray 12. The material is first presented to a push broom hyperspectral scanning system 14, in the form of a dense, stationary, monolayer. The spectra from all particles in the hyperspectral image are then sent to a classifier. This returns the pixel coordinates of target particles in the image. The pixel coordinates are then converted to world coordinates of the particles on the tray 12. A robotic picking system 16 is then directed to the target coordinates, to pneumatically pick the target particles up and place them into appropriate concentrate bins 18. In use, the material in any particular batch is first sieved to size fractions such that the upper diameter is at most a factor two times greater than the lower particle diameter. A batch of material is placed on a flat tray 12 WO 2006/054154 PCT/IB2005/003443 -9 atop a batch or sample preparation assembly 20. Referring to Figure 3, the tray 12 has a tapered rim 22, to keep particles from moving to the edge of the tray 12. A small ridge 24, approximately 300 micron deep, is provided at the bottom of the taper 22, for preventing particles from being pushed up the taper 22 in the preparation process. The surface 26 of the tray 12 has a uniform white, grey, or black, diffuse reflective coating. The area of the surface 26 of the tray 12 required depends on the amount of material being prepared. For 5 grams of material a surface area of approximately 100 cm 2 is required to avoid masking of particles. Referring to Figure 2, the batch preparation arrangement 20 includes a vibrating stage assembly 28 comprising a spring steel plate 30 and a linear electromagnetic vibrator 32. The electromagnetic vibrator 32 is used to produce the required vibration in the steel plate 30. The spring steel plate 30 is supported on a leveling mechanism comprising four bolts 34A, 34B, 34C and 34D, which in turn are fitted to a base plate 36 and secured in position by means of nuts 38A to 38H. A similar arrangement is used to mount secure the bolts 34C and 34D in position. The vibrating stage assembly 28 is arranged to distribute the particles into a uniform monolayer by applying a low amplitude (approximately 0.5 mm), high frequency (50 Hz) vibration to the tray 12. The vibrator 32 is switched off after a fixed time. With an accurately machined and leveled tray 12, a uniform monolayer is achieved within a fixed time of approximately 30 seconds. The hyperspectral scanning system 14 is a commercially available system. The current embodiment uses a dispersive prism-grating-prism spectrograph 40, produced commercially by Specim Ltd, Finland, coupled to a Dalsa IM30P digital CCD camera 42. A hyperspectral image of the particles on the tray 12 is acquired by scanning in a push-broom arrangement. The spectrograph 40 and camera 42 are mounted on a high precision linear scanning module 44, which is arranged to move WO 2006/054154 PCT/IB2005/003443 -10 the camera in the direction indicated by arrow 46. Illumination is by means of a tungsten halogen light source, with dual fibre-optic light lines 48 and 50, that illuminate the line element 52 in the field of view of the spectrograph 40. The spectrograph 40 has a 14300 x 13.1 micron slit. Each camera frame therefore represents the signal from an object segment of width w and length /, as shown in Figure 5. The optics disperse the signal in the direction perpendicular to the slit. Each frame on the CCD therefore contains spectra, dispersed along one dimension of the CCD (referred to as the spectral dimension of the CCD), originating from points in the line scan at corresponding pixel positions along the other dimension of the CCD (referred to as the spatial dimension of the CCD). The spatial resolution is determined by the slit dimensions and the magnification factor chosen. The magnification factor is chosen so that the spectrograph field of view I is slightly greater than the width of the particle tray 12, as shown in Figure 5. The CCD size in the spatial dimension is chosen to match the spectrograph image size in the spatial dimension. This optimises the spatial resolution in the direction perpendicular to the scan direction. For a 10 cm wide tray, this leads to a resolution of 100 microns per pixel, using the 1 megapixel Dalsa 1M30P camera 42. Spatial resolution in the scan direction w is determined by the width of the slit and magnification factor. For the situation described above, the scan direction resolution is 100 microns. This only holds true if the scan rate and frame rate are such that the camera is moved by not more than 100 microns per frame. In the current application a frame-rate of 30 fps and scan rate of 3 mm/s, is used. This allows for a 30 ms camera integration time. The spectrograph gives a spectral resolution of 2.8 nn (full width half maximum) in the 400 to 1000 nm range. In the present system, the WO 2006/054154 PCT/IB2005/003443 -11 spectrum is dispersed over part (approximately 500 pixels) of the CCD in the spectral dimension. The hyperspectral image is stored in memory of a control PC as a three dimensional array, often referred to as a datacube. This represents the two spatial dimensions and one spectral dimension. The processing of the datacube is described in the next section. The signal processing algorithm consists of the following steps: 1. Identify calibration point and sample region in image: Calibration points are identified automatically using a template matching technique. The spatial calibration system of the present invention will be described in more detail further below. Similarly, the borders of the sample region are determined automatically by identifying the tapered tray edge in the image, which is painted in a contrasting colour. 2. Determine central pixel of particles in images With the sample region identified, the datacube is truncated so that only this region of the image is processed further. A grey scale image is extracted from the datacube, by averaging over a spectral band, typically 600 to 650 nm. This gives a good contrast to the tray surface reflectance. The image is then convoluted with a matched filter, with circular geometry, and diameter corresponding to the centre of the size fraction in the batch. Pixels corresponding to the centres of particles are then identified by finding local maxima, above a certain threshold, in the convoluted image. The spectra at each of these central pixels are then sent on to the pre-processing and classification algorithm. The method is robust in situations where the particles are densely packed, and may in some cases be touching. By choosing the threshold WO 2006/054154 PCT/IB2005/003443 -12 conservatively, it is ensured that at least one spectrum per particle in the image is sent on for further processing. 3. Spectral pre-processing Pre-processing involves the following steps: 3.1. Spectrometer response and dark current correction: I = (l - IB)(S - IB), where Im - measured spectrum IB - dark current spectrum, which is used to correct for the dark current at all pixels across the line scan Is - spectrometer response (signal obtained from white reflector, e.g. Spectralon I - corrected spectrum 3.2. Spectral binning: Binning to a lower spectral dimension may or may not be carried out at this stage. This depends on the dimension of the classifier being used, which will be described in further detail below. 4. Spectral classification The classifier is based on a large training database, acquired using the hyperspectral imaging system described above. Representative mineral grains were obtained and categorised into separate target and non-target categories by expert mineral sorters. Reflectance spectra were obtained in the VIS to the near infrared (NIR) spectral range, with a sampling interval of approximately 2 nm. The database consists of reflectance spectra for over seven thousand representative target and non-target mineral grains.

WO 2006/054154 PCT/IB2005/003443 -13 The spectral classification involves extracting discriminating features in the spectra, and using these features to distinguish target grains from non-target grains. Public domain algorithms have been used to extract optimal features, based on the reflectance database described above. Specifically, the current embodiment for kimberlitic mineral identification employs a feature space derived from the reflectance spectra by a non-linear Fisher map technique, as set out by Loog M., Duin R.P.W., and Haeb-Umbach R. in Multiclass Linear Dimension Reduction by Weighted Pairwise Fisher Criteria, IEEE Trans. Pattern Analysis and Machine Intelligence, 23(7), 2001,762. Typically, spectra are sampled at 256 bands, leading to an initial feature space of 256 dimensions. The non-linear Fisher map reduces the dimension of the feature space to n - 1, where n is the number of classes to be classified. Classification of the spectra can then be accomplished using a number of potential classifiers, applied in this derived feature space. In the current embodiment for kimberlitic mineral sorting, both linear discriminant and nearest neighbour methods have been implemented. If a spectrum is classified as a target spectrum, its associated pixel coordinates are added to the list of target coordinates. 5. Transform target pixel coordinates to world coordinates The world coordinates (x,y), relative to a calibration point (0,0)word (which will be explained in further detail below), are obtained from the pixel coordinates (ij), measured relative to the image calibration point (0,O)image, by the following transformation: WO 2006/054154 PCT/IB2005/003443 -14 (x,y) = (iv/F, jR), where v - camera velocity (microns/s) F - frame rate (fps) R - spatial resolution perpendicular to scan direction (microns/pixel) The world coordinates (x,y) of target particles are then passed on to the extraction system. 6. Target extraction system Target extraction is by means of a robotic pneumatic picking system 16. In this application a Bosch-Rexroth SR4 SCARA robot, with rho 4.0 controller, is used. This system 16 can place a nozzle 54 at an (x,y) position on the plane of the tray surface 26 to an accuracy of 100 micron, and a z (i.e. height above tray 12) accuracy of 100 micron. A schematic of the nozzle 54 is given in Figure 4, and comprises a stainless steel body 56 terminating in a 0.5 mm diameter circular aperture 58, covered by a 200 micron aperture steel gauze 60. The nozzle 54 is used to extract particles between 300 and 2000 micron in a pick-and-place procedure. Particles are placed into one of any number of concentrate bins 18, situated near to the sample preparation assembly 20. The pneumatics of the system are designed to allow for the application of a reverse pressure, to ensure particles are placed into the bins 18. The bins 18 are at least a few centimetres deep, to avoid grains being blown out of the bin 18 by the reverse air blast. The robot cycle time is rated at approximately 300 microseconds, allowing for up to 3 pick-and-place movements per second.

WO 2006/054154 PCT/IB2005/003443 -15 Spatial calibration system As indicated above, a spatial calibration system is used to ensure that the world coordinates determined by the hyperspectral scanning system coincide with the coordinate system used by the robot. There are several ways in which this can be done. A simple spatial calibration system used in the current machine is described here, with reference to Figure 5. Four calibration points 60A, 60B, 60C and 60D, at the corners of a rectangle, are placed in the camera scan field, straddling the sample tray 12. The z coordinate of the points coincides with the sample tray surface level. The calibration points are laser printed onto the steel plate 30, to relative positional accuracy of 100 micron, so as to define four dimples/notches. One of the points 60A, 60B, 60C or 60D is chosen as the origin of coordinates. The relative position of the other three points can be used to test for misalignment of the camera track, or deviations in the angle of the slit, caused, for example, by twisting the camera assembly. The robot 16 is then calibrated to this point as follows. A set of calibration particles, with diameters from 300 micron to 2000 micron, is used in this regard. To calibrate the planar (x,y) coordinates, a 300 micron particle is placed in the dimple/notch of the chosen one point. The robot is manually placed so that the nozzle is directly over the particle, with, in one version, this process being aided by placing an LED light source in the nozzle, so that the position of the nozzle relative to the dimple can be monitored by the light emitted from the nozzle aperture. The z-position is calibrated by placing a particle corresponding to the top end of the material size fraction to be treated. The robot is then returned to (0,0)woad and the height of the nozzle is manually adjusted to just above the particle. This z-coordinate is recorded by the robot as the height at which the nozzle will operate for the batch. The planar (x,y) and z setting is tested with representative particles from the size fraction, to WO 2006/054154 PCT/IB2005/003443 -16 ensure that particles at both the lower and top end of the size fraction to be treated are successfully picked. Spectral calibration system Additional calibration systems of the present invention comprise a wavelength calibration system and a spectrometer response calibration (Is). A mercury vapour Spectral Calibration Lamp is used to provide a vapour line spectrum. This spectrum is used to calibrate the spectral dimension of the datacube. A calibration tray, with Spectralon@ standard white reflectance material is used for the spectrometer response calibration. The camera is moved to any point above the tray, and a single frame is acquired. This provides the spectrometer response spectra for each pixel in the line element. Due to the geometry, this response will be invariant to the position of the camera over its scan range. 2. Second embodiment Turning now to the second embodiment of the present invention shown in Figures 6A to 6C, a reflectance spectroscopy apparatus 62 comprises a feed presentation sub-system 64 and an optical detection sub-system 66. The feed presentation sub-system 64 comprises two vibratory feeders 68 and 70 and a grooved belt conveyor 72. Mineral grains are fed onto a tray 74 of the first vibratory feeder 68 via a hopper 76, and these are fed onto a tray 78 of the second vibratory feeder 70, imparting a first level of separation onto the grains. The second feeder 70 separates the grains further, and at the end of the tray 78 has five grooves 80 in order to constrain the grains in one direction, indicated by arrow 82. These grooves 80 are aligned with the grooves 84 in the rubber conveyor belt 72 such that the mineral grains are fed onto these grooves 84 whilst minimising spillage.

WO 2006/054154 PCT/IB2005/003443 -17 The conveyor 72 comprises a rubber belt with five V grooves 84 that are used to constrain and transport the mineral grains, two pulleys 86 and 88, and an associated motor, gearbox, and controller (not shown for the sake of clarity). To ensure a monolayer of separated grains, we assume a feedrate of 1 grain / mm. This gives a required belt speed using a single groove only of, (30 grains/s) x (I mm/grain) = 30 mm/s. Using all five grooves results in a required belt speed of 30 mm/s / 5 = 6 mm/s. The grains are fed past the optical detection module 86 and are allowed to fall onto a tray 90 placed under the conveyor 72. A plastic scraper is mounted at the rear of the conveyor 72 to dislodge any grains that remain stuck in the grooves 84. The grooves 84 in the belt 72 are required for the optical detection module 66, which collects light only from a single point on the belt, rather than scanning the entire belt, as will be described in more detail further on in the specification. The feeder trays 74, 78 are machined from aluminum, and have been sand-blasted, which provides the best results for feeding of mineral grains of this size. Binder vibrators 92A and 92B for tray 74 and 94A and 94B for tray 78 are used, together with associated Binder controllers. Adjusting the voltage output of the controllers changes the level of vibration and therefore the speed of transport of the grains. The purpose of the two feeders 68, 70 in series is to achieve a monolayer of mineral grains, whilst feeding at as high a rate as possible. In one version of this embodiment, the belt 72 is 28 mm wide, with each groove 3 mm wide at the top and 4 mm deep. The belt is 1070 mm in WO 2006/054154 PCT/IB2005/003443 -18 length (circumference) and 6.5 mm thick. The pulleys are 105 mm in diameter, leaving approximately 280 mm of the belt flat on the upper side of the conveyor. The optical detection sub-system 66 comprises an illumination and collection module 96, together with a spectral acquisition module. The grains are illuminated on the belt 72, the reflected light then being collected and focused into an optical fiber associated with the groove carrying the grains, which guides the light into a spectrometer 78. The spectrometer disperses the grain's reflected light onto a CCD detector, which acquires and stores the spectrum on a computer. In one version of this second embodiment, only one of the grooves 84 is used to carry grains, and thus only one optical fibre would be needed. If however, more than one groove is used, a corresponding number of optical fibres would be needed. In an alternate version of this second embodiment, the hyperspectral camera 42 described above with reference to the first embodiment of the invention could be used, which would then not only replace the optical fibre/s, but would also allow the grooved belt 78 to be replaced with a flat, non-grooved belt. The optical collection module 96 comprises a base and an optical focuser, the focuser being an off-the-shelf component purchased from OZ Optics Ltd. It was selected to have a field-of-view (FOV) of 100 microns, with an object distance of 100 mm. The illumination means takes the form of two 50 W tungsten halogen lamps. The preferred spectrometer for the spectral acquisition module 98 is an Ocean Optics S2000 Fiber Optic Spectrometer. The S2000 contains both the spectrometer and the detector in a single unit, and is therefore relatively small and robust. The data acquisition rate is also relatively fast, with a theoretical spectral acquisition time of 2 ms. The grating used had a WO 2006/054154 PCT/IB2005/003443 -19 resolution of 300 lines I mm, with a blaze wavelength of 500 nm. The detector is a Sony CCD, linear array with 2048 pixels. A dedicated 2 MHz A/D acquisition card (ADC2000-PCI) was supplied with the S2000, and was housed in a PCI slot in a PC. Light enters the S2000 through an interchangeable fiber optic, and this was selected to be an Ocean Optics fiber with an aperture of 100 micron (P100-2-VIS/NIR). This connected to the optical focuser through an SMA 905 connector. A 25 micron slit was included in the S2000, giving a spectral resolution of approximately 1 nm. In this second embodiment, a pneumatic mini-cyclone 98 for mineral grain extraction is used. The basic design of the extractor is that of a cyclone. Air is forced trough a valve at the top of the mini-cyclone, creating a vacuum inside, and along a pick-up arm. The arm terminates in a nozzle 100, which gets positioned just above the belt 72. The vacuum is sufficient to extract the mineral grain from the belt groove 84, and the grain is then transported to a cylindrical portion of the cyclone. The extracted grains are released by opening the valve at the base of the cylinder. This second embodiment of the present invention also requires spectral calibration and the use of a classifier, both of which are similar to what has been described above with reference to the first embodiment. The primary advantage of the present invention is thus to enable analysis of heavy mineral concentrates at a much faster rate than was previously possible, shortening the time required for the processing of exploration samples.

Claims (18)

1. A hyperspectral imaging apparatus for identifying and sorting target particles within a batch of particles, the apparatus comprising: a tray for supporting the batch of particles; leveling means for leveling the batch of particles into substantially a monolayer; a hyperspectral scanning system for scanning the batch of particles, to produce a hyperspectral image of the batch of particles; a classifier for determining the pixel coordinates of target particles in the hyperspectral image; converter means for converting the pixel coordinates to world coordinates of the target particles on the tray; and target particle extraction means for picking the target particles based on the calculated world coordinates and for transferring the picked target particles to a storage arrangement.

2. A hyperspectral imaging apparatus according to claim 1, wherein the size range of the particles is -2.0+0.3 mm.

3. A hyperspectral imaging apparatus according to either claim 1 or claim 2, wherein the target particles are kimberlitic indicator minerals. WO 2006/054154 PCT/IB2005/003443 -21

4. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the tray comprises a tapered rim and a ridge provided at the bottom of the tapered rim.

5. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the tray is part of a batch sample preparation assembly, with the apparatus further including a vibrating stage assembly comprising a spring plate and a linear electromagnetic vibrator, the electromagnetic vibrator being used to produce the required vibration in the plate so as to yield the monolayer of batch particles.

6. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the hyperspectral scanning system comprises a spectrograph and a camera, the spectrograph being arranged to produce a line element in the field of view of the camera, the spectrograph and camera being arranged to move linearly along the length of the tray so as to produce a hyperspectral image of a batch of particles.

7. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the hyperspectral image is stored as a three-dimensional array, comprising the two spatial dimensions of the tray and one spectral dimension.

8. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the apparatus is aimed at the visible (VIS) to the near infrared (NIR) spectral range, with the classifier derived from a database in which reflectance spectra for target and non-target particles are stored.

9. A hyperspectral imaging apparatus according to any one of the preceding claims, wherein the tray comprises a plurality of calibration points that form part of a spatial calibration system, for WO 2006/054154 PCT/IB2005/003443 -22 ensuring that the world coordinates determined by the converter means coincide with the coordinate system used by the target particle extraction means.

10. A reflectance spectroscopy apparatus for identifying target particles comprises: a moving conveyor belt for carrying particles including target particles, the conveyor belt defining a plurality of grooves. a feed presentation sub-system comprising at least one vibratory feeder that is arranged to define a monolayer of particles and for feeding the monolayer of particles to the moving conveyor belt; an optical detection sub-system located operatively above the conveyor belt for illuminating the particles and storing the reflected spectra; a classifier for determining the location and mineral type of the particles within the grooves of the conveyor belt; and target particle extraction means for picking the target particles based on the classifier's determined location of target particles.

11. A hyperspectral imaging apparatus according to claim 10, wherein the size range of the particles is -2.0+0.3 mm.

12. A hyperspectral imaging apparatus according to either claim 10 or claim 11, wherein the target particles are kimberlitic indicator minerals. WO 2006/054154 PCT/IB2005/003443 -23

13. A reflectance spectroscopy apparatus according to any one of the preceding claims 10 to 12, wherein the feed presentation sub system comprises two vibratory feeders, each vibratory feeder comprising a tray for carrying the particles.

14. A reflectance spectroscopy apparatus according to claim 13, wherein a first feeder extends between a hopper and a second feeder, with the tray of the second feeder defining a plurality of grooves that are arranged to line up with the plurality of grooves in the conveyor belt.

15. A reflectance spectroscopy apparatus according to either claim 13 or claim 14, wherein the feeder trays are machined from aluminum, and are sand-blasted.

16. A hyperspectral imaging method of identifying target particles within a batch of particles, the method comprising: leveling a batch of particles into substantially a monolayer; hyperspectrally scanning the batch of particles to produce a hyperspectral image of the batch of particles; determining the pixel coordinates of target particles in the hyperspectral image; converting the pixel coordinates to world coordinates of the target particles on the tray; extracting the target particles based on the calculated world coordinates; and transferring the picked target particles to a storage arrangement. WO 2006/054154 PCT/IB2005/003443 -24

17. A method according to claim 16, which further includes the step of vibrating the batch of particles to produce the monolayer of batch particles.

18. A reflectance spectroscopic method of identifying target particles comprises: providing a moving conveyor belt for carrying particles including target particles, the conveyor belt defining a plurality of grooves. providing a feed presentation sub-system comprising at least one vibratory feeder that is arranged to define a monolayer of particles and for feeding the monolayer of particles to the moving conveyor belt; providng an optical detection sub-system located operatively above the conveyor belt; illuminating the particles on the conveyor belt; storing the reflected spectra; determining the location and mineral type of the particles within the grooves of the conveyor belt; and extracting the target particles based on the determined location of target particles. -