Ai-based diamond grading

Abstract

A method is provided that employs artificial intelligence (Al) for predicting a colour grade of either a D-Z colour diamond or a fancy colour diamond, either for an existing polished diamond or for a polished diamond to be cut from a rough diamond. The method includes applying either an IR spectrogram and weight of an existing polished diamond as input data to a first machine learning (ML) model, to predict a D-Z colour grade or a fancy colour grade of the existing polished diamond or applying an IR spectrogram and weight of a rough diamond as input data to a second ML model, to predict a D-Z colour grade or a fancy colour grade of a polished diamond to be cut from the rough diamond.

Description

AI-BASED DIAMOND GRADINGTECHNICAL FIELD

[0001] The present application relates to the field of grading diamonds, and, in particular, to Al-based methods.BACKGROUND

[0002] Diamonds are classified based on their colour, appearance, and physical properties and these classifications have a key role in diamond evaluation. Diamonds are composed of essentially pure carbon, but may be contaminated by other elements, such as nitrogen, or by defects, which are acquired naturally during the course of formation. Diamonds appear in a variety of colours, some of which are very rare and thus highly prized.

[0003] In a colourless diamond, the presence of a weak-intensity colour tint (usually yellow or brown) is considered a defect of the diamond and significantly decreases the price. Very pure colourless diamonds have no body colour, hence the greater value they have. By looking at rough diamonds it is difficult to determine the colour that they will present post polishing process. A diamond's origin, whether it was mined from the earth or synthesised in a lab and subsequent treatment history is also a major factor on its value.

[0004] The colour of natural diamonds is typically caused by atomic-scale features (often termed optical centres, optical defects, or simply defects) that occur within the diamond structure. These defects may include mainly the aforementioned contamination of trace atoms of nitrogen. In some diamonds, trace elements are added as the result of treatment or synthesis in a laboratory.

[0005] Diamonds may also contain other imperfections in the crystal lattice of the carbon atoms, often termed lattice defects. These imperfections are vacant carbon- atom locations (holes or vacancies) in crystal lattices of diamonds, plastic deformation (when there is deviation of some carbon atoms in the crystal from theideal position) or several nitrogen-carbon bonds. The configuration of these defects and their concentration vary with the growth conditions and subsequent geological or treatment history.

[0006] All diamonds may be classified based on the presence or absence of nitrogen atoms within the crystal lattices. The most common Type I diamonds (~99%) contain nitrogen atoms as their main impurity, commonly at a concentration of up to 0.2%. Type I diamonds absorb electromagnetic radiation in both the infrared, visible and ultraviolet region at A > 320 nm. They may a Iso have a characteristic fluorescence and visible absorption spectrum.

[0007] Type II diamonds have no measurable nitrogen atoms in the lattice (no interaction of light with lattice vibrations). These diamonds therefore absorb light not in the visible region but in the far ultraviolet region below 225 nm. They also have different fluorescence characteristics, but no discernible visible absorption spectrum.

[0008] These two types are further subdivided into several other types based on the arrangement of the contaminating nitrogen in the crystal lattice. For example, Type la diamonds, which constitute approximately 95-98% of natural diamonds, contain nitrogen atoms in clusters and normally vary from near-colourless to light yellow. These "cape" diamonds derived their name from diamonds that were initially mined in Cape Province, South Africa. Type la diamonds subdivided into type laA diamonds containing nitrogen atoms aggregated in pairs, and sub-type laB diamonds containing clusters of four nitrogen atoms in their crystal lattice.

[0009] Type lb diamonds also contain nitrogen, but as isolated atoms instead of clusters. These nitrogen atoms are more diffuse and dispersed throughoutthe crystal in isolated sites. Natural type lb diamonds are almost always brown, yellow, or orange and are extremely rare. These diamonds absorb green light in addition to blue and have a much more intense or darker yellow or brown colour than Type la diamonds. Almost all synthetic diamonds, which are manufactured in the high-pressure high temperature (HPHT) method and contaminated by nitrogen, are of Type lb.

[0010] Type Ila diamonds have no measurable nitrogen or boron impurities and chemically are the purest diamonds with the highest thermal conductivity. They are very transparent to ultraviolet light with no absorption at 0 > 230 nm. These diamonds are usually colourless, but they can also be grey, light brown, lightyellowor light pink. Their occasional colour is probably a result of plastic deformations in the crystal.

[0011] While Type Ila diamonds are being extruded towards the surface of the Earth, the pressure and tension could cause structural anomalies arisingthrough the plastic deformation during the growth of the tetrahedral crystal structure, leading to imperfections. These imperfections can conferthe different colours to the stones. A. T. Collins etal in "High-temperature annealing of optical centres in type-1 diamond", Journal of Applied Physics, 97(8), (2015), pp. 083517-1 -10 describes the HPHT process for repairing the structural deformations in the Type Ila diamonds, bleaching much or all of the diamond's colour.

[0012] Type Ila diamonds constitute a great percentage of the Australian production. Many famous large diamonds, like the Cullinan, Koh-I-Noor, and Lesedi La Rona, are Type Ila. Synthetic diamonds grown using chemical vapour deposition (CVD) process typically also belong to this type.

[0013] Type lib diamonds contains boron impurities making them one of the rarest natural, blue-coloured diamonds and extremely valuable. The trace element boron is responsible for most of them being light blue or greyish blue, because the absorption spectrum of boron causes these stones to absorb red, orange, and yellow light, though examples with low levels of boron impurities can also be colourless. The historic Wittelsbach Blue diamond was sold for a record-breaking $23.4 million to London jeweller Laurence Graff at Christie's in December 2008. It was consequently recut to 31 .06 carat and renamed the Wittelsbach-Graff.

[0014] Type lib diamonds show distinctive infrared absorption spectrum and show gradually increasing absorption towards the red side of visible spectrum. Thesediamonds are also p-type semiconductors, unlike other diamond types, due to uncompensated electron holes. Therefore, they are known to conduct electricity. As little as 1 ppm of boron is enough for this effect.

[0015] Every diamond on the market is assigned a colour grade that is certified by a skilled person or a gemmological laboratory. This allows the colour of a diamond to be accurately measured and graded for determining its price on the market. Minor differences in the diamond's colour graded in the laboratories are very difficult if not impossible to detect outside of these laboratories. The diamond industry has adopted the common international colour scale of diamonds, and almost every diamond sold today is rated using this colour scale. Identification of the diamond type, also known as grading, is critical. Diamonds are also classified as natural, treated or synthetic. The GIA (Gemmological Institute of America) first developed a grading system in the 1950s to grade yellow diamonds outside of the usual colour range of colourless to light yellow (D-Z). Over time, the Institute expanded and improved the system, publishing its findings and effectively introducing the Coloured Diamond Grading System in use today. These improvements included improvements to the colour scoring system and the addition of new“Fancy Deep” and “Fancy Vivid” colour descriptions. Thus, diamonds whose colours fall outside the GIA D to Z colour range are described as "fancy colour diamonds". When evaluatingtheir colouration, the GIA considers three attributes: hue is an attribute that we normally perceive as a colour, such as red, yellow, green, blue, or something in between, tone the relative lightness or darkness of a colour and saturation, the relative depth and strength of a colour. GIA recognizes 27 different shades or"colours" of fancy-coloured diamonds.

[0016] Like D-Z diamonds, GIA grades fancy colour diamonds under carefully controlled lightingand viewing conditions. Appraisers look at the diamond in its faceup position and establish that colour’s three attributes, bracketing each of the attributes, using reference stones as colour comparators, and finally assigning one of fancy colour ratings to the coloured diamond.

[0017] Diamonds are generally graded on the D-to-Z colour grading system, which is based on the relative absence of colour, where the difference in colour between D, E and F is negligible and can be detected only by a gemmologist via side- by-side comparisons to master diamonds. When there is colour, it is usually yellowish or brownish. The G, H, I and J grades are considered near colourless or having slight traces of colour that are not easily detectable to the untrained eye, particularly when the diamonds are placed in a mounting orframe in jewellery items. Down the scale, the traces of colour in a diamond become gradually more apparent.

[0018] In general, almost all D- to M-grade diamonds are considered colourless, even though they contain varying degrees of colour (usually yellow). Starting from K colour level, yellow tint can be easily detected by the naked eye. The K, L and M diamonds are therefore said to have "faint" colours, while diamonds in the colour range from N to R have an easily seen yellow or even brown tint and are significantly less expensive than higher grades. True fancy-coloured diamonds, such as pinkand blue, are graded on a separate colour scale.

[0019] Polished single diamonds are much easier to grade than rough (unpolished) diamonds. A rough diamond may be barely transparent, which results in its possibly incorrect grading.

[0020] D-to-Z colour grading is based on the observations of a trained grader, who compares a polished diamond to colour master stones of known position on a grading scale under special lamp of white light. GIA master stones are located at the highest point in their respective grade range. A diamond equal to the G master stone is graded a G. If it has slightly less colour, it would receive a grade of F.

[0021] A diamond with more colourthan the G masterand less than the H master would receive a G grade. A diamond with less colourthan the E master is graded a D. A diamond with more colourthan the Y-Z master stone is graded face-up as a fancy colour (see King et al., “Colour grading “D-to-Z” diamonds”, Gems & Gemmology, Vol. 44, No. 4, pp. 296-321 , 2008). The intermediate grades are usually assessedaccording to the grader’s own judgement. This is a largely subjective process however and is dependent upon the grader’s skill in this area.

[0022] Due to complex treatments and the evolution of synthetic diamond growth techniques, gemmological laboratories require several devices to aid determining diamond colour grades. The Gran Colorimeter, for example, can very accurately and impartially determine a diamond colour, and is even able to produce results that estimate how high or low the diamond is in a grading category. Other instruments that are used to grade diamonds are spectroscopic units, especially in the Visible-UV range, to measure the absorption spectrum of the diamond.

[0023] Fluorescence spectroscopy is used for characterisation of diamonds that are capable of emitting fluorescence when exposed to ultraviolet light. However, this is not a common trait and only about 30% of diamonds exhibit this characteristic to some degree. In addition, because fluorescent glow is usually blue (complementary colour to yellow), diamonds of l-M colour having a blue fluoresce tend to appear to one grade whiter when exposed to daylight. As a result, if a lower-colour diamond with strong fluorescence is chosen, the stone will appear to be more colourless than it is, as some of the yellow body colour disappeared out in the daylight. For this reason, l-M diamonds tend to sell at a slight premium when they possess medium to very strong fluorescence, but colourless (D-F) diamonds having fluorescence sell at up to a 15% discount, since fluorescence is perceived as a defect. Overall, diamond fluorescence cannot be a major factor in grading diamonds for the above reasons.

[0024] The emergence of more complex grading techniques for diamonds and the development of sophisticated synthetic diamonds have meant that the use of standard gemmological techniques described above cannot entirely guarantee the identity and quality of a stone. For this reason, new advanced spectroscopic techniques for grading diamonds have recently been developed. U.S. patents No. 7,652,755 and No. 7,388,656 by Liu Labs describe multifunction dual integrating sphere spectrometer for grading diamonds controlled by artificial intelligence software with several functions of spectral measurement, colour measurement,fluorescence measurement, photoluminescence measurement, colour grading of diamonds, colour grading of coloured diamonds, colour grading of jadeite, and alexandrite effect grading.

[0025] S. R. Boyd et al in "Multiple growth events during diamond genesis: an integrated study of carbon and nitrogen isotopes and nitrogen aggregation state in coated stones", Earth and Planetary Science Letters, 86, (1987), pp. 341-353, described the combination of dynamic and high-sensitivity static mass spectrometry with high-resolution technique to investigate the variance of carbon and nitrogen isotope composition, nitrogen concentration and the degree of nitrogen aggregation within diamonds, employing microgram-sized samples.

[0026] Raman spectroscopy, which is a form ofvibrational spectroscopy that can be used to analyse very small samples, is also a non-destructive testing technique and is very reliable in identifying a diamond by comparing its spectrum to known spectra in a reference library. This can also be done at room temperature or at liquid nitrogen temperature (-195.8 °C). In a photoluminescence mode, a Raman spectrometer is capable of identifying synthetic diamonds and Type II diamonds that have been treated to produce a very high colour grade. Although Raman provides different information from Fourie Transform Infra-Red (FTIR), as the physical way that the techniques analyse a sample spectrum is different, their results are often considered in combination and can provide the final answer regarding a diamond.

[0027] S. Eaton-Magana and C. Breeding in "An Introduction to Photoluminescence Spectroscopy for Diamond and Its Applications in Gemmology", Gems & Gemmology, Spring 2016, Vol. 52, No. 1 , pp. 2-17, reviewed a photoluminescence spectroscopy for grading diamonds. It is a non-destructive analyticaltechnique in which a diamond is illuminated with UV light and the resulting luminescence is recorded as a plot of emitted light intensity versus wavelength. In the last decade, photoluminescence has become an essential tool used by major gemmological laboratories to separate treated and synthetic diamonds from their natural counterparts.

[0028] Notwithstanding the aforementioned developments in optical spectroscopy for grading diamonds, the problems of incorrect grading of rough diamonds prior to their polishing and having reduced transparency, difficulty to separate treated and synthetic diamonds from their natural counterparts and to identify plastic deformations in diamonds still persist. These problems lead to only about 50-55% of rough diamonds being graded correctly today. Therefore, there is a long-felt need to significantly improve the grading of diamonds, particularly rough diamonds, to identify plastic deformations in diamonds, and to separate treated and synthetic diamonds from their natural counterparts. These problems are addressed by the present invention.SUMMARY

[0029] A method is provided that employs artificial intelligence (Al) for predicting a colour grade of a diamond, either for an existing polished diamond or for a polished diamond to be cut from a rough diamond.

[0030] In one embodiment, the method for predicting a D-Z colour grade of a diamond or a fancy colour grade of a diamond, either for an existing polished diamond or for a polished diamond to be cut from a rough diamond, the method comprising steps of:I. Measuring an IR spectrogram of either the existing polished diamond or of the rough diamond;II. Measuringthe weight of the existing polished diamond or estimatingtheweight of the polished diamond to be cut from the rough diamond; andIII. (a) If said diamond is a D-Z colour grade diamond, applyingthe IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a D-Z colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a D-Z colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML modelis trained to correlate input data with a D-Z colour grade of polished diamonds providing the input data, and wherein the second ML model is trained to correlate input data with a D-Z colour grade of polished diamonds to be cut from rough diamonds providingthe input data; or(b) If said diamond is a fancy colour grade diamond, applying the IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a fancy colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a fancy colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML model is trained to correlate input data with a fancy colour grade of polished diamonds providingthe input data, a nd wherein the second ML model is trained to correlate input data with a fancy colour grade of polished diamonds to be cut from rough diamonds providingthe input data.

[0031] The method may further include a step of measuring fluorescence of the existing polished or rough diamond, wherein the input data to the respective first- or second-ML model includes the measure of fluorescence.

[0032] Alternatively, or additionally, the method may further include a step of measuring one or more of nitrogen concentration, boron concentration, and plastic deformations of the existing polished diamond or of the rough diamond, wherein the input data to the respective first- or second-ML model includes the one or more measures of nitrogen concentration, boron concentration, and plastic deformations of the respective diamond.

[0033] Alternatively, or additionally, the input data to the first ML model may include an indication of a shape of the existing polished diamond. Alternatively, or additionally, the input data to the second ML model may include an indication of a shape of the polished diamond to be cut from the rough diamond.

[0034] The input data to the first or the second ML model may also include providing an indication as to whether the respective existing polished diamond orrough diamond is a natural diamond. The input data to the first or the second ML model may further include providing an indication as to whether the respective existing polished diamond or rough diamond is a fancy colour diamond, and also the colour grade assigned to the fancy colour diamond.

[0035] The first- and second-ML models may also be configured to generate from their respective input data an indication as to whetherthe spectrogram is valid.

[0036] The method may also include the training of the first ML model to correlate its input data with a colour grade of a polished diamond bytrainingthe first ML model with multiple data sets, each data set including an IR spectrogram and a weight of a polished diamond, each data set being associated with a label of a pre-determined colour grade of the polished diamond of the data set.

[0037] The method may also include the training of the second MLto correlate its input data with a colour grade of polished diamonds to be cut from rough diamonds by training the second model with multiple data sets, each data set including an IR spectrogram of a rough diamond and an estimated weight of a polished diamond to be cut to from the rough diamond, each data set being associated with a label of a colour grade determined for the polished diamond subsequently cut from the rough diamond.

[0038] In some embodiments, the first- and second-ML models are one or more of a support vector machine (SVM) based process; a decision tree-based process; and a deep neural network (NN) process, wherein the deep neural network is one or more models of a convolutional neural network (CNN), a regional CNN (RCNN), and a long-short term memory recurrent CNN (LSTM Recurrent CNN).BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Disclosed embodiments will be understood and appreciated more fully from the following detailed description taken in conjunction with the appended figures. The drawings included and described herein are schematic and are not limiting the scope of the disclosure. It is also noted that in the drawings, the size ofsome elements may be exaggerated and, therefore, not drawn to scale for illustrative purposes. The dimensions and the relative dimensions do not necessarily correspond to actual reductions to practice of the disclosure. In the figures:

[0040] Fig. 1 is a flowchart of a possible exemplary process for Al prediction of a colour grade of a polished diamond, according to some embodiments of the present invention;

[0041] Fig. 2 is a flowchart of a possible exemplary process for Al analysis of a rough diamond, to predict a colour grade of a polished diamond to be cut from the rough diamond, accordingto some embodiments of the present invention; and

[0042] Fig. 3 is a schematic block diagram of Al modelling to create a machine learning (ML) model for analysis of polished or rough diamonds, according to some embodiments of the present invention;DEFINITIONS

[0043] As described in the Background section, any diamond whose colour falls outside the GIA colour range of D to Z is defined as a “fancy colour diamond”. The method of the present invention is therefore divided into two cases: 1 ) the method for predicting a colour grade of a D-Z colour diamond (based on the D-Z colour grade scale); and 2) the method for predicting a colour grade of a fancy colour diamond (based on the fancy colour grade scale).

[0044] The term "comprising", used in the claims, is "open ended" and means the elements recited, ortheir equivalent in structure orfunction, plus any other element or elements which are not recited. It should not be interpreted as being restricted to the means listed thereafter; it does not exclude other elements or steps. It needs to be interpreted as specifying the presence of the stated features, integers, steps or components as referred to, but does not preclude the presence or addition of one or more other features, integers, steps or components, or groups thereof. Thus, the scope of the expression "a device comprising x and z" should not be limited to devices consisting only of components x and z. Also, the scope of the expression "amethod comprising the steps x and z" should not be limited to methods consisting only of these steps.

[0045] Unless specifically stated, as used herein, the term "about" is understood as within a range of normal tolerance in the art, for example within two standard deviations of the mean. In one embodiment, the term "about" means within 10% of the reported numerical value of the number with which it is being used, preferably within 5% of the reported numerical value. For example, the term "about" can be immediately understood as within 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1 %, 0.5%, 0.1 %, 0.05%, or 0.01 % of the stated value. In other embodiments, the term "about" can mean a highertolerance of variation depending on for instance the experimental technique used. Said variations of a specified value are understood by the skilled person and are within the context of the present invention. As an illustration, a numerical range of "about 1 to about 5" should be interpreted to include not only the explicitly recited values of about 1 to about 5, but also include individual values and sub-ranges within the indicated range. Thus, included in this numerical range are individual values such as 2, 3, and 4 and sub-ranges, for example from 1 -3, from 2-4, and from 3-5, as well as 1 , 2, 3, 4, 5, or 6, individually. This same principle applies to ranges reciting only one numerical value as a minimum or a maximum. Unless otherwise clear from context, all numerical values provided herein are modified by the term "about". Other similar terms, such as "substantially", "generally", "up to" and the like are to be construed as modifying a term or value such that it is not an absolute. Such terms will be defined by the circumstances and the terms that they modify as those terms are understood by those of skilled in the art. This includes, at very least, the degree of expected experimental error, technical error and instrumental error for a given experiment, technique or an instrument used to measure a value.

[0046] As used herein, the term "and / or" includes any and a ll combinations of one or more of the associated listed items. Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonlyunderstood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the specification and relevant art and should not be interpreted in an idealized or overly formal sense unless expressly so defined herein. Well-known functions or constructions may not be described in detail for brevity and / or clarity.

[0047] It will be understood that when an element is referred to as being "on", "attached to", "connected to", "coupled with", "contacting", etc., another element, it can be directly on, attached to, connected to, coupled with or contacting the other element or intervening elements may also be present. In contrast, when an element is referred to as being, for example, "directly on", "directly attached to", "directly connected to", "directly coupled" with or "directly contacting" another element, there are no intervening elements present. It will also be appreciated by those of skill in the art that references to a structure or feature that is disposed "adjacent" another feature may have portions that overlap or underlie the adjacent feature.DETAILED DESCRIPTION

[0048] In the following description, various aspects of the present application will be described. For purposes of explanation, specific configurations and details are set forth in order to provide a thorough understanding of the present application. However, it will also be apparent to one skilled in the art that the present application may be practiced without the specific details presented herein. Furthermore, well- known features may be omitted or simplified in order not to obscure the present application.

[0049] According to the present invention, a method for predicting a D-Z colour grade of a D-Z colour diamond or a fancy colour grade of a fancy colour diamond, either for an existing polished diamond or for a polished diamond to be cut from a rough diamond, comprises the following steps:Step I: Measuring an IR spectrogram of eitherthe existing polished diamond or of the rough diamond;Step II: Measuring the weight of the existing polished diamond or estimating the weight of the polished diamond to be cut from the rough diamond; andStep III:(a) if said diamond is a D-Z colour grade diamond, applyingthe IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a D-Z colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a D-Z colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML model is trained to correlate input data with a D-Z colour grade of polished diamonds providing the input data, and wherein the second ML model is trained to correlate input data with a D-Z colour grade of polished diamonds to be cut from rough diamonds providingthe input data; or(b) if said diamond is a fancy colour grade diamond, applying the IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a fancy colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a fancy colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML model is trained to correlate input data with a fancy colour grade of polished diamonds providing the input data, and wherein the second ML model is trained to correlate input data with a fancy colour grade of polished diamonds to be cut from rough diamonds providingthe input data.

[0050] It is to be understood that the term "grading" is not limited only to diamond colour identification based on the GIA diamond colour scale or on the fancy colour scale, but may also comprise identification of plastic deformations, prediction of fluorescence emission in diamonds and separation of treated and synthetic diamonds from their natural counterparts. Also, the diamonds graded by the method of the present embodiments may be either polished or rough. In a specificembodiment, the diamonds graded by the method of the present invention are fancy colour diamonds.

[0051] Grading the colour of coloured diamonds requires considerable skill, as small gradations of colour can make a big difference in value. In addition, diamonds can be processed to change their colour. Since natural colour diamonds are much more expensive than cut diamonds, expert judgment is required to determine what is referred to as "colour origin". The machine learning model, trained as described herein, significantly simplifies the procedure forgradingdiamonds, particularlyfancy colour diamonds, to improve the throughput and consistency of diamond grading. Results of the ML model can also avoid inaccurate grading that causes significant losses in diamond trades.

[0052] Poor transmission causes an incoherent and diffusive background. Resonant elastic scattering of light may be visualised as a chain of coherent absorption and re-emission events. Photons are then re-emitted in many random directions as the wave-vector is not conserved by the system. The resonant scattering results in a significant increase of the mean trajectory of photons travelling through the medium, meaning absorbed energy increases.

[0053] The light that is not absorbed by colour centres in the crystal but is emitted randomly is “stray light.” Stray light causes incorrect grading, especially for rough diamonds. “Stray light” is dependent on wavelength (A) and is generally proportional to 1 / A4(in accordance with Rayleigh formula for scattering of light).

[0054] Thus, measurement in the IR region of the spectrum, where the wavelength of the electromagnetic radiation is appreciably longer than the UV- Visible region is much better, as the difference between measurement of a rough diamond and a polished one is smaller and results in an improved accuracy of colour grading. Light transmission intensity therefore increases with wavelength. Shorter wavelength irradiation produces more diffusion of photons, meaning more absorption. The optical losses are directly proportional to 1 / A4. IR light in the range of, for example, A = 700 nm to 1 mm therefore has less transmission loss than UV-VIS light. The use of IR spectroscopy is therefore preferable in the method of the present invention.

[0055] One of the major aspects of the present invention is the use of machinelearning methods, based on the artificial intelligence, to analyse results obtained from both the mass measurement (i.e., weighing) of tested diamonds and their IR spectroscopic measurements.

[0056] Fig. 1 is a flowchart of an exemplary process 100 for Al prediction of a colour grade of a polished diamond, according to some embodiments of the present invention. A first step 102 includes testing a polished diamond (that is, an existing polished diamond) by infrared (IR) spectrophotometer and recording the resulting spectrogram, that is, the IR absorption spectrum of the diamond.

[0057] Next at a step 104, the spectrogram is uploaded to an Al prediction system, which is based on a machine learning (ML) model described further herein below (also referred to herein as a “first ML model").

[0058] At a step 106, the Al prediction system may be trained to first analyse whether the spectrogram (i.e., the “scan”) meets basic criteria (typically for criteria for appropriate intensity variation) to be considered a valid scan. If not, the user may be alerted that a new scan is required. Also, the Al prediction system may be trained to determine, according to the spectrogram, whether the diamond is natural, treated, or synthetic.

[0059] At a step 108, the user may also enter into the system the weight of the diamond, aswell as additional, optionalinput data. The ML modeloftheAl prediction system may be trained to apply the additional input data to more accurately predict the colour grade of the polished diamond. The additional input data may include shape and fluorescence level (quantitatively or subjectively measured), as well as known contaminations and the type of diamond, in terms of whether the diamond is natural, synthetic, or treated.

[0060] When optional data is to be entered, the step 108 may be preceded by steps (not shown) of measuringthe optional data, such as thefluorescence levelandthe contamination. Contamination may be measured as nitrogen concentration, boron concentration, and plastic deformations in the diamond.

[0061] At a step 110, the system applies the ML model to analyse the spectrogram and weight and any optional input data and to generate a prediction of a colour grade of the polished diamond, and, optionally, a predicted fluorescence level. At a step 112, the system returns the prediction of the colour grade to the user, and, optionally, the predicted fluorescence level.

[0062] Fig. 2 is a flowchart of an exemplary process 200 for Al analysis of a rough diamond, to predict a colour grade of a polished diamond to be cut from the rough diamond, according to some embodiments of the present invention. Steps 202-212 of process 200 are parallel to steps 102-112 of process 100 described above, the main difference being that the weight input and the optional input of shape refer to estimates of the polished diamond to be cut from the rough diamond, ratherthan to actual measurements made of an existing polished diamond.

[0063] Spectroscopic techniques for grading diamonds are typically based on transmission of UV or visible light through the diamonds. These techniques are less appropriate for grading rough diamonds, because UV and visible light is subject to a significant level of optical interference during transmission through a rough diamond, as compared to the polished one. Indeed, UV-Vis instruments and methods are capable of correctly grading only about 45-47% of all the rough diamonds in the trade. The IR spectroscopy, employed by the methods described herein reduce this problem.

[0064] Fig. 3 is a schematic block diagram of a training process 300 of Al modelling, by which a machine learning (ML) model 302 is generated for analysis of polished or rough diamonds, according to some embodiments of the present invention.

[0065] The generated ML model may include any of the following algorithms: a support vector machine (SVM) based process; a decision tree-based process; and a deep neural network (NN) process, wherein the deep neural network is one or moremodels of a convolutional neural network (CNN), a regional CNN (RCNN), and a long- short term memory recurrent CNN (LSTM Recurrent CNN).

[0066] The Al in the present invention involves a training process shown in diagram 300. The process includes training the ML model 308 with input data sets 310, each data set based on a single polished or rough diamond and associated with a set of labels 320, which represent the predictions that will be made by the trained ML model.

[0067] As described above, an input data set includes a diamond’s IR spectrogram 312, and its weight or mass 314. Additional input 316 may include shape, diamond type (natural, treated, or synthetic), fluorescence level, and defects, such as nitrogen concentration, boron concentration, and plastic deformation. Training of the ML model correlates the input data with pre-determined labels, including colour grade 322, fluorescence 324, and diamond type 326. (Generally, when fluorescence or diamond type are provided as labels, they are not provided as input.)

[0068] After being trained, the ML model (e.g., a deep neural network) predicts a colour grade of a polished diamond (or the level according to other grading scales, such as fancy yellow colour) from the input data set of an IR spectrogram and a mass of a particular diamond (either rough diamond or existing polished diamond). The prediction may also be based on optional input data parameters relating to shape, to defects of the crystalline structure, such as nitrogen concentration, boron concentration, and plastic deformations, and to fluorescence of the diamond.

[0069] In some embodiments, the method of the present invention further comprises a step of measuring fluorescence of the existing polished diamond or rough diamond, wherein the input data to the respective first- or second-ML model includes the measure of fluorescence. In a particular embodiment, the method of the present invention further comprises a step of measuring one or more of nitrogen concentration, boron concentration, and plastic deformations of the existing polished diamond or of the rough diamond, wherein the input data to the respectivefirst- or second-ML model includes the one or more measures of nitrogen concentration, boron concentration, and plastic deformations of the respective polished or rough diamond.

[0070] In another embodiment, the input data to the first ML model includes an indication of a shape of the existing polished diamond. In still another embodiment, the input data to the second ML model includes an indication of a shape of the polished diamond to be cut from the rough diamond. In some embodiments, the input data to the first or the second ML model further includes providing an indication as to whether the respective existing polished diamond or rough diamond is a natural diamond. In a certain embodiment, the input data to the first or the second ML model further includes providing an indication as to whether the respective existing polished diamond or rough diamond is a D-Z colour diamond or a fancy colour diamond. In a further embodiment, the first- and second-ML models are further configured to generate from their respective input data an indication as to whether the spectrogram is valid.

[0071] In a specific embodiment, the method of the present invention further comprises training the first ML model to correlate its input data with a D-Z colour grade or fancy colour grade of polished diamonds by training the first ML model with multiple data sets, each data set including an IR spectrogram and a weight of a polished diamond, each data set being associated with a label of a pre-determined D-Z colour grade or fancy colour grade of the polished diamond. In another specific embodiment, the method of the present invention further comprises training the second ML to correlate its input data with a colour grade of polished diamonds to be cut from rough diamonds by trainingthe second model with multiple data sets, each data set including an IR spectrogram of a rough diamond and an estimated weight of a polished diamond to be cut to from the rough diamond, each data set being associated with a label of a D-Z colour grade or fancy colour grade determined for the polished diamond subsequently cut from the rough diamond.

Claims

CLAIMS1. A method for predicting a D-Z colour grade of a diamond or a fancy colour grade of a diamond, either for an existing polished diamond or for a polished diamond to be cut from a rough diamond, the method comprising steps of: measuring an IR spectrogram of either the existing polished diamond or of the rough diamond; measuring the weight of the existing polished diamond or estimating the weight of the polished diamond to be cut from the rough diamond; and(a) if said diamond is a D-Z colour grade diamond, applying the IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a D-Z colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a D-Z colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML model is trained to correlate input data with a D-Z colour grade of polished diamonds providing the input data, and wherein the second ML model is trained to correlate input data with a D-Z colour grade of polished diamonds to be cut from rough diamonds providing the input data; or(b) if said diamond is a fancy colour grade diamond, applying the IR spectrogram and the weight of the existing polished diamond as input data to a first machine learning (ML) model, to predict a fancy colour grade of the existing polished diamond, or applying the IR spectrogram and the weight of the rough diamond as input data to a second ML model, to predict a fancy colour grade of the polished diamond to be cut from an rough diamond, wherein the first ML model is trained to correlate input data with a fancy colour grade of polished diamonds providing the input data, and wherein the second ML model is trained to correlate input data with a fancy colour grade of polished diamonds to be cut from rough diamonds providing the input data.

2. The method according to claim 1, further comprising a step of measuring fluorescence of the existing polished diamond or rough diamond, wherein the input data to the respective first- or second-ML model includes the measure of fluorescence.

3. The method according to claim 1 or 2, further comprising a step of measuring one or more of nitrogen concentration, boron concentration, and plastic deformations of the existing polished diamond or of the rough diamond, wherein the input data to the respective first- or second-ML model includes the one or more measures of nitrogen concentration, boron concentration, and plastic deformations of the respective polished or rough diamond.

4. The method according to claim 1 or 2, wherein the input data to the first ML model includes an indication of a shape of the existing polished diamond.

5. The method according to claim 1 or 2, wherein the input data to the second ML model includes an indication of a shape of the polished diamond to be cut from the rough diamond.

6. The method according to claim 1 or 2, wherein the input data to the first or the second ML model further includes providing an indication as to whether the respective existing polished diamond or rough diamond is a natural diamond.

7. The method according to claim 1 or 2, wherein the input data to the first or the second ML model further includes providing an indication as to whether the respective existing polished diamond or rough diamond is a D-Z colour diamond orfancy colour diamond.

8. The method according to claim 1 or 2, wherein the first- and second-ML models are further configured to generate from their respective input data an indication as to whether the spectrogram is valid.

9. The method according to claim 1 or 2, further comprising training the first ML model to correlate its input data with a D-Z colour grade or fancy colour grade of polished diamonds by training the first ML model with multiple data sets, each data set including an IR spectrogram and a weight of a polished diamond, each data set being associated with a label of a pre-determined D-Z colour grade or fancy colour grade of the polished diamond.

10. The method accordingto claim 1 or 2, further comprising trainingthe second ML to correlate its input data with a colour grade of polished diamonds to be cut from rough diamonds by trainingthe second modelwith multiple data sets, each data set including an IR spectrogram of a rough diamond and an estimated weight of a polished diamond to be cut to from the rough diamond, each data set being associated with a label of a D-Z colour grade or fancy colour grade determined forthe polished diamond subsequently cut from the rough diamond.

11. The method according to claim 1 or 2, wherein the first- and second-ML models are one or more of a support vector machine (SVM) based process; a decision tree-based process; and a deep neural network(NN) process, wherein the deep neural network is one or more models of a convolutional neural network (CNN), a regional CNN (RCNN), and a long-short term memory recurrent CNN (LSTM Recurrent CNN).

Images