1. INTRODUCTION

Gemstone identification and evaluation in gemological laboratories reveal the type of gemstones and their corresponding value factors, such as color and size, which are the key references for maintaining transparency in the gemology industry. However, such testing is rarely applied to jewelry pieces because the surrounding metal mount or adjacent gemstones may interfere with gemstone examination. In addition, owing to recent technical advances in gemstone synthesis and treatments, the complexity of natural versus synthetic gemstone identification has significantly increased. For example, natural diamonds, their laboratory-grown counterparts, and colorless non-diamond materials share a similar appearance and are almost impossible to identify visually. As a result, fraudulent activity can take place where lower value gemstones are attempted to be passed off as higher value gemstones. Separating natural gemstones from synthetic counterparts in fine jewelry has become an urgent challenge for the approximately 100 billion US dollar retail diamond industry [1]. Advanced spectroscopy is necessary to conclusively identify and separate valuable gemstones from their lower value counterparts or simulants.

Raman and photoluminescence (PL) spectroscopy is an effective tool for gemstone identification and screening because of its nondestructive nature and high sensitivity [2–7]. Raman scattering is widely used for unknown gemstone material analysis, and the PL features concerned with atomic impurities can be used for gemstone screening [3,8,9], identification, and treatment detection [10–17]. These advanced optical methods are widely used in gemological laboratories for routine loose gemstone examination but not in jewelry testing because of the challenges in sample alignment. Raman/PL spectroscopy requires an accurate sample alignment to overlap the tested sample with the focal point of the spectroscopy system. Failure to achieve this might result in a weak signal level or even incorrect results from neighboring samples. This limitation slows down the measurement and prevents the application of Raman spectroscopy in mounted gemstone measurement because most mounted samples are relatively small, further complicating the alignment. In addition, lasers that can generate Raman/PL signals significantly exceed the exposure limit for eye safety. Therefore, the entire measurement system and the sample need to be enclosed to meet safety requirements [18].

A typical alignment strategy is to couple the laser spectroscopy system with an optical microscope [2]. Microscope based Raman spectroscopy with a motorized stage can be used for sample alignment in a fully enclosed environment. Although the microscope provides a spatial resolution of ${\sim}1\;{\unicode{x00B5}{\rm m}}$, its small field of view limits the user when attempting to see multiple gemstones on the entire jewelry piece. An additional labeling process is required to record the results, thus limiting its application in jewelry measurements. An alternative solution is to apply a mapping technique for the automatic collection of spectra across a small area with micrometer-level spatial resolution to study the distribution of PL features [19]. However, the data acquisition of such Raman and PL mapping spectroscopy is time consuming. The scanning area is limited by the field of view of the objective lens in the system, which is usually less than $1\;{\rm mm} \times 1\;{\rm mm}$, making it insufficient to cover the entire jewelry sample.

Color and carat weight are the major value factors of gemstones. Traditionally, the color of the diamond is visually evaluated by a gemologist, and the carat weight is weighed using scales before mounting on a jewelry piece. Unfortunately, it is difficult to evaluate these parameters once the gemstone is mounted, as part of the gemstone is blocked by the metal mount. In addition, the color of the metal also interferes with judgment [20,21]. Owing to the low commercial value and typically high volumes of melee-sized (${\lt}{0.2}\;{\rm carat}$, ${1}\;{\rm carat} = {200}\;{\rm mg}$) gemstones, the size is manually measured using calipers, and the color of each miniature sample is visually evaluated, but this is still a labor-intensive process [13]. The cost of time, labor, and equipment for gemstone evaluation may exceed the value of such melee-sized samples.

An ideal jewelry identification device should positively identify the mineral type of the gemstone and comprehensively distinguish natural diamonds from their laboratory-grown counterparts. To optimize flexibility, instruments should be able to measure a wide variety of jewelry, such as rings, earrings, waistbands, pendants, necklaces, watches, and brooches, in a wide size and shape range. Further, automatic identification and result labeling, rather than user interpretation and recording, are desirable. Finally, the material, maintenance, and operational costs should be reasonable. Although there are many gemstone testing instruments on the market, their design and technology define their advantages and limitations. For example, absorption based systems detect the transmission of ultraviolet (UV) light or analyze the UV-visible or infrared absorption spectrum. However, jewelry settings usually interfere with the absorption measurement, and high-end absorption spectroscopy is usually slow and expensive. In addition, some low-cost transmission type systems have been reported to misread laboratory-grown diamond as natural [8]. Fluorescence and phosphorescence imaging based systems compare the color and brightness of samples’ fluorescence and phosphorescence features. Although these types of instruments can positively identify natural diamond from laboratory-grown diamond, they cannot be used to identify simulants [8,9]. Finally, while fluorescence spectroscopy based systems analyze the fluorescence spectrum to positively identify the majority of natural diamonds, many laboratory-grown diamonds, simulants, and a small percentage of natural diamonds have to be identified by additional testing [8]. Generally, most of this information is not disclosed to the public.

This paper proposes an imaging-assisted Raman/PL spectroscopy system to aid gemological laboratories and the jewelry industry in analyzing diamond jewelry samples. The system can be used for screening natural diamonds out from laboratory-grown diamonds and diamond simulants and estimating gemstone weight and color. The system consists of a laser spectroscopy probe pre-aligned with a low-distortion imaging camera and a three-axis motorized translation stage to realize automatic measurement and accurate sample alignment. In addition, such a design maintains an enclosed laser beam path to meet radiation safety requirements. The imaging camera provides minimal distortion with a field of view that covers the entire jewelry sample. The sample size and scanning distance can be calculated based on the image pixel-to-distance conversion; therefore, user selected samples on the jewelry piece can be sequentially aligned and measured. The color image of the sample can be used for the color evaluation of gemstones. Preliminary results prove the capability of using a wide spectral range of Raman/PL spectroscopy for diamond identification, including natural (mined) diamonds, laboratory-grown diamonds, and commonly encountered diamond simulants. Combined with a diffused lighting environment, the primary application of this system includes color evaluation, size/weight estimation, and identification of every gemstone on a mounted jewelry piece.

2. BACKGROUND

Raman spectroscopy detects the vibrational and rotational energy of the chemical structure fingerprint, which has been proven effective in mineral identification and gemstone testing [2–4]. However, as the crystal structure and chemistry of natural and laboratory-grown diamond is the same, natural and laboratory-grown materials share the same Raman scattering spectrum and cannot be distinguished via Raman spectroscopy. However, PL spectroscopy can be used to distinguish natural diamonds from their laboratory-grown and treated counterparts by evaluating their atomic defects [22–25]. Major gemological laboratories apply Raman and PL spectroscopy to gemstone analysis as they are nondestructive, contactless, and highly sensitive [2,10].

Both Raman and luminescence features can be collected from the same measurement. The fundamental difference between Raman and PL signals is their dependency on the excitation wavelength. As Raman scattering results from the inelastic scattering of photons, the emission wavelength depends on the excitation energy. Conversely, the majority of gemstones’ PL features are based on the crystallographic defects or impurity in the mineral, resulting in the emission wavelength remaining unchanged with varying excitations. To achieve consistent analysis of gemstone Raman and PL spectra, Raman scattering is usually recorded in wavenumbers, targeting the wavenumber region of 200 to 2000 to obtain the material’s structural fingerprint. In contrast, PL is recorded in wavelength and covers a range of several hundred nanometers.

Currently, Raman and PL spectroscopy probes are commercially available. They enable optimal signal collection using optical fibers in restricted spaces or hostile environments and are easily adaptable to other sensing systems [26–29]. The 532 and 785 nm Raman spectroscopy probes are the most popular configurations; however, many important PL features in natural and laboratory-grown diamonds cannot be excited by these wavelengths [30]. For example, the N3 defect, consisting of three nitrogen atoms surrounding a vacancy, has a zero phonon line (ZPL) at 415 nm and is detectable in most natural diamonds but not in laboratory-grown diamonds under ambient temperature or diamond simulants [13]. Neither 532 nor 785 nm can excite N3 fluorescence for diamond identification. In addition, the strong fluorescence background usually overwhelms the weaker Raman scattering signals from commonly encountered minerals on the market, such as corundum, beryl, zoisite, jadeite, and chrysoberyl [31,32]. A customized 405 nm Raman spectroscopy probe was used to overcome these problems. Further, 405 nm laser spectroscopy, combined with a wide spectral range spectrometer, can detect major diamond PL features in the visible to near-infrared region.

 

Fig. 1. System design of the jewelry identification system. (a) System with the pre-aligned spectroscopy probe, imaging camera, and motorized translation stage to assist the gemstone sample alignment for laser spectroscopy probe. LED light panels and reflectors have been used here to provide a homogeneous diffused lighting environment for the gemstone color analysis. (b) Experimental prototype of the imaging-assisted spectroscopy.

Download Full Size | PDF

The vast majority of modern jewelry pieces include a number of melee-sized gemstones ranging from 0.9 to 3.8 mm (0.01 to 0.2 carat) in diameter as part of the design. As these miniature gemstones are low in value but high in quantity, their identification using various spectroscopic techniques in tandem with the traditional testing protocol becomes expensive. However, 405 nm Raman and PL spectroscopy is a cost-effective analytical technique because of its capability to differentiate natural diamonds from laboratory-grown diamonds and simulants from one spectral measurement per sample.

Detecting diamond treatment is generally not a concern for melee-sized diamonds due to their low value, as treating them is not a cost-effective application for diamond treatment facilities.

As a result, color-treated melee-sized diamonds rarely appear on the jewelry market. To perform comprehensive diamond treatment detection usually requires PL measurement under multiple excitation wavelengths at liquid nitrogen temperature to detect fine spectral features [5].

The imaging camera can effectively assist sample alignment during spectroscopy measurement as it can cover a sufficient field of view for the large jewelry piece and maintain the required spatial resolution for melee gemstone alignment. A field of view of approximately $20 \times 20 \; {\rm mm}$ is sufficient to cover the vast majority of jewelry pieces; this can be achieved by using a commercially available ${{2/3}^\prime}$ camera with an imaging lens of ${0.3} \times$ magnification while maintaining ${15 +}$ line pairs per mm spatial resolution to detect 0.9 mm diameter gemstones.

The color and carat weight evaluation of melee-sized gemstones can be performed using a low-distortion imaging system in a homogeneous diffused lighting environment. Round brilliant-cut diamond is the most popular type of gemstone cut used in jewelry design because it emphasizes the brilliance and sparkle of jewelry [33]. As round-cut gemstones follow cutting ratios of diameter to depth, similar to that of diamond, the volume of any round-cut gemstone can be estimated using its diameter. The proposed weight estimation assumes all samples have fairly standard, modern proportions and reasonably symmetrical designs. This assumption can be applied to diamond jewelries but not to other colored gemstones, which generally do not follow standard cuts.

Owing to the mounted gemstones on jewelry, the color of the metal can affect the gemstone’s apparent color. For example, gold usually works as a yellow background to enhance the yellowish color of the diamond. Therefore, gemological laboratories have traditionally performed color evaluation only on unmounted gemstones. A homogeneous diffused lighting environment can be used for evaluating the color of mounted gemstones to minimize the saturated reflection of facets and emphasize body color [21]. A simple color evaluation can be performed by converting the average red, green, and blue values within a gemstone image to lightness, chroma, and hue values. Although the visual appearance of white-colored gemstones, such as diamonds and diamond simulants, on jewelry pieces is highly related to the metal color, this method can separate gemstone colors into several groups based on their chroma values.

3. EXPERIMENTAL DETAILS

This study proposes image-assisted Raman/PL spectroscopy, which integrates an imaging system and a three-axis motorized translation stage along with a spectroscopy probe, for automatic gemstone screening and identification, as shown in Fig. 1. A 405 nm laser Raman and PL spectroscopy probe (SPC-R405, Spectra Solution) with a focal length of 15 mm was used in the system for spectral analysis of the gemstone. The probe was mounted approximately 20° down and toward the sample stage. Owing to the large focal length of the spectroscopy probe, blocking the camera viewing area by the metal barrel lens can be minimized in this system. A machine vision imaging camera (BFS-U3-50S5C-C, FLIR) with a low-distortion imaging lens (VS-LDA35, VS Technology) created an approximately ${30.6} \times {25.6}\;{\rm mm}$ field of view to cover the majority types of jewelry samples, including rings, pendants, earrings, and brooches. A 410 nm longpass filter (410LP RapidEdge, Omega filter) was used to block the scattering laser light from entering the imaging camera. A wide spectral range modular spectrometer (Avaspec-mini, Avantes) with 1.2 nm (${70}\;{{\rm cm}^{- 1}}$) spectral resolution and 400–950 nm sensing range was used to separate natural diamonds from laboratory-grown diamonds and diamond simulants because many characteristic PL features are located in the near-infrared region [30]. The sample stage was controlled using a three-axis motorized translation stage to support automatic sample alignment. The motorized translation stage provided a 20 mm traveling range along all three axes for the sample’s translation and alignment. Multiple gemstones on the same jewelry piece were sequentially measured by moving the selected stones underneath the laser spot. Finally, four light-emitting diode (LED) panel lights surrounded the sample chamber, and two white light reflectors were placed below and above the sample area to provide a homogeneous diffused lighting environment for color analysis. The LED panel light was built using LED strips (CRI-MAX, YUJILEDS) with a color temperature of 5600 K and a color rendering index (CRI) of over 95.

Accurate sample alignment is critical for this system. The spectroscopy probe was pre-aligned with the camera and motorized translation stage to assist in gemstone sample alignment. First, the vertical alignment overlapped the focal planes of the spectroscopy probe and the imaging camera, where the strongest spectroscopy signal from the tested sample showed sharp surface facet junctions from the image. Therefore, the imaging system can use the sharpness of the image to evaluate the vertical-axis alignment of the sample. The horizontal alignment localized the laser spot center on the image plane of the camera. The targeted laser spot position, combined with the information on the camera’s pixel pitch and magnification of the imaging system, can serve as a reference point for sample alignment. After the calibration process, the imaging system was used to support the user to localize the positions of the tested samples and calculate the physical distance between the laser spot and the sample. Therefore, the motorized stage could decipher the required movement and automatically move to the user selected sample to reach the laser spot for the Raman/PL measurements. The tolerance of the alignment was limited by the diameter of the smallest sample, which is approximately $\pm {0.45}\;{\rm mm}$ or $\pm {36}$ pixels laterally in the system. The camera’s depth of field, which is approximately $\pm {1}\;{\rm mm}$ along the vertical direction, results in $\pm {0.36}\;{\rm mm}$ lateral variation, which is still within tolerance.

One advantage of this system is the flexibility of adapting different spectrometers based on the requirements of the spectral resolution and sensing range. For example, colored stones can be studied with a high-resolution spectrometer with a narrow sensing range, and fine Raman scattering spectra can be resolved. As the sensing range is away from the colored gemstone’s strong red and green fluorescence regions, the same 405 nm spectroscopy probe can effectively minimize the unwanted fluorescence background from the gemstone minerals [32].

4. RESULTS AND DISCUSSION

A. Natural and Laboratory-Grown Diamonds

Examples of characteristic Raman and PL spectra for natural, high-pressure, high-temperature (HPHT)-, and chemical vapor deposition (CVD)-grown diamonds under 405 nm laser excitation are presented in Fig. 2. Figure 2(a) shows the PL features of natural diamonds. The N3 fluorescence with ZPL at 415 nm and vibronic sidebands extending to 452 nm appears in the vast majority of natural diamonds, where strong fluorescence usually dominates the natural diamond’s PL spectra [12–14]. N3 fluorescence is not detected in some rare types of natural diamonds. These low-nitrogen-concentration diamonds account for less than 2% of natural gem diamonds [34,35].

 

Fig. 2. Typical diamond PL spectra. (a) PL spectra for natural diamond and weak fluorescence natural diamond. (b) PL spectra for HPHT-grown, CVD-grown, and treated CVD-grown diamonds.

Download Full Size | PDF

Some natural diamonds do not contain detectable N3 fluorescence because of the high concentration of aggregated nitrogen pair defects that can quench this fluorescence. The PL spectra of this group of natural diamonds usually contain a broad emission band centered at 520 nm and a narrow peak at 788 nm, which are related to the nitrogen pair concentration in natural diamonds [7,30]. Figure 2(b) shows the typical PL spectra of laboratory-grown diamonds. Conversely, laboratory-grown diamonds did not exhibit the above-mentioned fluorescence features in the PL spectra recorded at room temperature. Several important PL features can be used to identify laboratory-grown diamonds. For example, the nickel-related emission features at 883 nm are detectable from HPHT-grown diamonds with a nickel catalyst [25,30,36]. The 468 nm emission system and neutral nitrogen vacancy (NV) centers with ZPL at 575 nm are fairly common in nitrogen-doped CVD-grown diamonds [30,37]. The silicon vacancy (SiV) center at 737 nm is also a common feature in the PL spectra of CVD-grown diamonds [5,30]. Finally, a fluorescence band centered at 450 nm can be detected in CVD-grown diamonds that have been irradiated and annealed [30]. In addition to the emission features, the Raman peak of diamond at ${1332}\;{{\rm cm}^{- 1}}$ (428 nm under 405 nm excitation) can be detected from natural diamond with weak N3 fluorescence and laboratory-grown diamonds, but can be swamped by the strong N3 fluorescence exhibited by many natural diamonds.

Compared with existing diamond screening devices that use fluorescence spectroscopy, the proposed system has a higher natural diamond detection rate owing to the stronger excitation power and better light collection efficiency from the laser spectroscopy probe, which can effectively generate the N3 fluorescence of natural diamond. The laboratory-grown diamond related features, which are undetectable by ${\lt}{10}\;{\rm mW}$ LED light sources, can be easily detected using a 50 mW 405 nm spectroscopy probe. As a result, the vast majority of laboratory-grown diamonds can be positively identified.

B. Diamond Simulants

The characteristic PL and Raman scattering spectra of colorless diamond simulants are shown in Figs. 3 and 4, respectively, under 405 nm laser excitation. As diamond simulants are chemically different from diamonds, they do not have the same Raman and PL features. Figure 3 shows the PL spectra of fluorescent diamond simulants, including gadolinium gallium garnet (GGG), synthetic sapphire, synthetic spinel, yttrium aluminum garnet (YAG), and zircon. According to the results, none of these fluorescent diamond simulants emitted PL features similar to diamond fluorescence. Figure 4 shows the Raman spectra of non-fluorescent diamond simulants, including cubic zirconium (CZ), strontium titanate, synthetic moissanite, synthetic rutile, and topaz. These simulants did not have detectable PL features; therefore, their Raman scattering became the dominant feature in the spectra. Owing to the discrepancy in the PL and Raman spectra, the identification of diamond simulants based on spectral analysis can be realized. The identification of diamond simulants based on spectral analysis can be realized owing to the discrepancy in the PL and Raman spectra. This method outperforms the fluorescence spectroscopy based diamond screening devices, which lack the capability to identify non-fluoresce diamond simulants.

 

Fig. 3. Typical PL spectra for fluorescent diamond simulants, including gadolinium gallium garnet (GGG), synthetic sapphire, synthetic spinel, yttrium aluminum garnet (YAG), and zircon.

Download Full Size | PDF

 

Fig. 4. Typical Raman spectra for non-fluorescent diamond simulants, including CZ, strontium titanate, synthetic moissanite, synthetic rutile, and topaz, under 405 nm laser excitation.

Download Full Size | PDF

C. Jewelry Identification and Evaluation

To validate the performance of the system for jewelry identification, we measured several jewelry samples and present five of them in this section. The tested samples were from GIA’s research sets and client-submitted stones from the grading laboratory operations. All samples were previously conclusively identified by experienced gemologists using various gemological and advanced techniques, including Fourier transform infrared (FTIR) absorption microscopy, Raman and PL spectroscopy, fluorescence spectroscopy, and fluorescence imaging.

In the prototype, multiple stones can be selected from the camera’s preview image for measurement by the user; subsequently, the system sequentially calculates the distances between stones and converts them into $x - y$ stage movements, aligns the stones under the spectroscopy probe for measurement, analyzes the Raman/PL spectra, and displays the result with color-coded output. Each selected stone took approximately 5 s, including sample alignment, data collection, and spectral analysis. Four jewelry pieces were tested under this system to prove their capability to positively identify natural diamonds within the pieces, as shown in Fig. 5. Multiple gemstones were selected for testing from each jewelry piece. The stones of interest are highlighted by coloring the adjacent samples with blue ink. Finally, many laboratory-grown diamonds were identified by using this system based on their Raman scattering and characteristic PL features.

 

Fig. 5. Example of jewelry identification, showing the tested jewelry pieces, including three rings and one pin. Multiple stones were measured and detected in each piece, including (a) four HPHT-grown diamonds, (b) one natural diamond and one HPHT-grown diamond, (c) seven natural diamonds, and (d) two HPHT-grown diamonds. In the figures, the natural diamond and HPHT-grown diamond are colored in cyan and dark green, respectively.

Download Full Size | PDF

A pendant was tested to demonstrate the capability of this gemstone identification system for jewelry testing, as shown in Fig. 6. The jewelry sample included 54 colorless to near-colorless gemstones, as presented in Fig. 6(a). The different gemstone samples were aligned in rows, except for the second sample in the CZ (sixth) row, which was a natural diamond that was unintentionally misplaced by the jeweler. The sample included five moissanite stones (synthetic moissanite) in the first row, followed by nine HPHT-grown diamonds, 10 CVD-grown diamonds, two rows of 10 natural diamonds, and a mixture of one natural diamond and nine CZs in the last row. Under the homogeneously diffused white light, an image of this pendant was captured by the camera, as shown in Fig. 6(b). To analyze all samples, the motorized stage sequentially moved each sample underneath the laser spot for Raman and PL measurements. Figures 6(c)–6(h) show the Raman and PL spectra of the six selected samples on this pendant. The features presented in the spectra can be used to identify the mineral types of samples among moissanite, HPHT-grown diamond, CVD-grown diamond, natural diamond, and CZ, based on the features presented in the previous sections. The system also successfully identified that the second sample in the sixth row was an incorrectly placed natural diamond. The carat weight and color estimation were evaluated based on the images collected from the camera, as shown in Fig. 6(b). The diameter of each selected sample can be converted from the number of occupied pixels multiplied by the magnification of the camera and pixel pitch of the detector.

 

Fig. 6. Mounted jewelry measurement and the Raman and PL spectra of six selected samples on this pendant. (a) A pendant with 54 pieces of samples, including natural diamonds, laboratory-grown diamonds, and diamond simulants, was used to test the system for mounted jewelry. The types of gemstone samples are labeled on the side of the image. (b) Image of this pendant under homogeneous diffused light. Raman and PL spectra of (c) moissanite, (d) HPHT-grown diamond, (e) CVD-grown diamond, (f) natural diamond, (g) natural diamond in the CZ row, and (h) CZ.

Download Full Size | PDF

The carat weight of the melee-sized diamond sample can be estimated based on the spectral analysis results of mineral identification and the following equation:

The lightness of the sample (quality of light and dark), chroma (saturation of color), and hue (object’s basic color) values can be converted from the average red, green, and blue values of the image. The color conversion of the image was achieved by averaging the pixels in a cropped region of each diamond sample. The extreme dark and bright areas in the image were excluded from the analysis by setting a brightness threshold in the calculation to minimize the impact caused by internal reflection in faceted diamonds. As the melee diamond samples are miniature in size, they were assumed to be homogeneous in color. The results of the diameters, estimated carat weights, and color analysis of all samples are presented in Supplement 1. The results of the color evaluation and visual observation are consistent. Based on these values, the moissanites in the first row and the CZs in the sixth row are close to colorless owing to the low chroma value, whereas the HPHT-grown diamonds in the second row are near colorless; further, all the natural diamonds in the fourth and fifth rows are slightly colored, and the majority of them have a yellowish hue. Finally, the CVD-grown diamonds in the third row are slightly colored with a brownish hue, consistent with the characteristics of the as-grown CVD-grown diamonds. A comprehensive analysis of each sample’s material type, diameter, estimated carat weight, and color information is presented in Supplement 1. Compared with conventional weight estimation, the difference in total weight is 0.02 ct., which is 1.3% of the total carat weight. The conventional estimation performed by caliber measurement is also presented as reference in Supplement 1. Since the hue of the metal acts as a colored background, this color evaluation can represent the observation of stones only after being mounted, not their natural color as loose stones. For samples mounted on uneven surfaces, such as ring samples, where the height distribution of the gemstone exceeds the system’s $\pm {2}\;{\rm mm}$ depth of field, it is necessary to split the testing into multiple measurements to maintain that all targeted samples remain in focus.

Compared with existing natural diamond screening devices, the proposed system can positively identify natural diamonds, the majority of laboratory-grown diamonds, and diamond simulants based on Raman and PL analysis, thereby reducing the requirement of using additional testing by multiple inspection techniques to analyze the undetermined samples. With the aid of sample positioning and labeling, the time requirement for sample preparation, testing, and report generation can be significantly shortened. In addition, the carat weight and color of the diamonds can be simultaneously evaluated; this minimizes the requirement of additional labor and provides further merit to this identification device.

5. CONCLUSIONS

This study proposed imaging-assisted Raman and PL spectroscopy to simplify and accelerate gemstone identification and evaluation of mounted jewelry and experimentally demonstrated its functionality. The proposed systems included a 405 nm spectroscopy probe that was pre-aligned with an imaging camera and a three-axis motorized translation stage to perform automatic Raman and PL spectroscopy analysis on jewelry pieces. This design allows laser spectroscopy to be operated under the laser safety standard while maintaining a sufficient field of view for all jewelry pieces. Multiple user selected gemstone samples were sequentially measured to simplify the identification process. Further, this study demonstrated that the proposed system could accurately identify natural diamonds, laboratory-grown diamonds, and diamond simulants in different types of jewelry pieces. Finally, gemstone identification and evaluation of color and size were performed on a pendant piece to demonstrate the capability of this system for jewelry analysis.

Future research will focus on integrating multiple spectroscopy probes into one system to perform high-resolution Raman spectroscopy and wide-spectral-range PL spectroscopy sequentially. The feasibility of applying a 405 nm Raman/PL spectroscopy probe to other gemstone materials, such as corundum (ruby and sapphire), emerald, tourmaline, spinel, and zoisite, will also be evaluated to extend the application to all popular gemstones on the jewelry market. To support the spectroscopy analysis, a special sample holder similar to other mature products is under development to support the sample orientation and rotation [8,9]. Finally, automatic sample localization and focusing are crucial in further accelerating the process of jewelry identification and maintaining consistency in spectroscopy analysis. The recent development in deep learning provides an alternative way to solve sample localization and image focusing problems in a knowledge-driven way [39,40]. Similar technologies will be investigated to achieve fully automatic jewelry analysis.