Hyperspectral-based jewelry gemstone color intelligent grading system and method

Abstract

The present application relates to the technical field of gemstone grading, and specifically relates to a gemstone color intelligent grading system and method based on hyperspectrum, which comprises a uniform illumination module for uniformly illuminating a gemstone sample placed on an illumination station; a spectrum acquisition module for acquiring a visible light spectrum of the gemstone sample; a spectrum analysis module for performing spectrum analysis on the visible light spectrum of the gemstone sample to obtain spectrum reflectivity; an information acquisition module for acquiring basic information of the gemstone sample; a model matching module for matching a color model of the gemstone sample according to the basic information of the sample; a spectrum conversion module for converting the spectrum reflectivity into corresponding color parameters according to the color model matched by the gemstone sample; and a grading module for outputting a color grade corresponding to the gemstone sample based on a preset color grading model according to the color parameters.

Description

Technical Field

[0001] The present invention relates to the technical field of jewelry and jade grading, and in particular to a jewelry and jade color intelligent grading system and method based on hyperspectral analysis. Background Art

[0002] The color grade of jewelry and gemstones is a key determinant of their value. For precious gemstones like rubies and sapphires, a specific color grade can command a price per carat of nearly 100,000 RMB, significantly higher than similar gemstones of other color grades. Therefore, accurate color grading is crucial for assessing the value of jewelry and gemstones.

[0003] Currently, the color grading of gemstones and gemstones primarily relies on visual colorimetry. This method requires experienced gemstone color graders to work in a stable, uniform environment, using standard color charts and standard samples to measure the gemstone's color parameters, such as hue, lightness, and chroma, to ultimately determine the sample's color grade. Although visual colorimetry is relatively simple and intuitive, it still presents some challenges:

[0004] Different graders may assign different color grades to the same gemstone due to individual visual differences or varying levels of experience, leading to inconsistent and subjective results.

[0005] Visual colorimetry needs to be performed in a stable, uniform, and non-interfering light environment. Any changes in lighting conditions may affect the accuracy of grading. It is highly dependent on the environment and has poor grading reliability.

[0006] The manual grading process is slow, especially when dealing with a large number of samples, which is inefficient, increases costs, and is time-consuming and labor-intensive.

[0007] Based on this, there is an urgent need for a hyperspectral-based intelligent grading system and method for the color of jewelry and jade, which can realize the intelligent and quantitative grading of jewelry and jade, and greatly improve the grading efficiency of jewelry and jade. Summary of the Invention

[0008] One of the purposes of the present invention is to provide a hyperspectral-based intelligent grading system and method for the color of jewelry and jade, which can realize the intelligent and quantitative grading of jewelry and jade, and greatly improve the grading efficiency of jewelry and jade.

[0009] In order to achieve the above objectives, a hyperspectral-based intelligent grading system for gemstone color is provided, comprising:

[0010] Uniform lighting module, used to evenly illuminate the jewelry and jade samples placed on the lighting station;

[0011] Spectrum acquisition module, used to collect visible light spectra of jewelry and jade samples on the lighting station;

[0012] Spectral analysis module, used to perform spectral analysis on the visible light spectrum corresponding to the collected jewelry and jade samples and obtain the corresponding spectral reflectance;

[0013] An information acquisition module is used to acquire basic information of jewelry and jade samples placed on the lighting station;

[0014] The model matching module is used to match the color model corresponding to the jewelry and jade samples based on the obtained basic information of the samples and the preset model matching strategy;

[0015] A spectrum conversion module is used to convert the spectral reflectance corresponding to the jewelry and jade sample into the color parameters corresponding to the color model matched by the jewelry and jade sample according to the color model matched by the jewelry and jade sample;

[0016] The grading module is used to output the color grade corresponding to the jewelry and jade sample based on the color parameters corresponding to the color model matched by the jewelry and jade and based on the preset color grading model.

[0017] The technical principles and effects of this solution: First, the jewelry and jade samples are placed on a lighting station for uniform illumination, eliminating errors caused by uneven light sources and improving the accuracy of spectral acquisition. The spectrum acquisition module then collects the corresponding visible light spectrum of the jewelry and jade samples on the lighting station, capturing a spectral image of the jewelry and jade samples. Spectral analysis of the collected visible light spectrum is then performed to determine the spectral reflectance of the jewelry and jade samples.

[0018] Then the next step is to match the corresponding color model. Specifically, it is to obtain the basic information of the jewelry and jade samples, and match the color model corresponding to the jewelry and jade samples through the basic information of the samples and the preset model matching strategy.

[0019] After matching the corresponding color model, the corresponding spectral reflectance will be used as input, and the spectral reflectance will be converted into the color parameters corresponding to the matching model. Then, the color parameters are combined with the preset color grading model to determine the color grade corresponding to the jewelry and jade sample.

[0020] Compared with the traditional visual colorimetry method, this solution does not require human intervention throughout the entire process, from sample placement, information acquisition, spectral collection and analysis, to color model matching, parameter conversion and final grading. This greatly improves work efficiency and realizes intelligent and quantitative identification of jewelry and jade. It can accurately and intelligently determine the color of jewelry and jade samples without damaging the samples.

[0021] Furthermore, the preset model matching strategy is:

[0022] Identify the sample type corresponding to the jewelry and jade samples based on their basic information;

[0023] The corresponding jewelry model matching table is retrieved from the database, and the color model corresponding to the sample type corresponding to the jewelry and jade sample is identified based on the sample type corresponding to the jewelry and jade sample and the retrieved jewelry model matching table; the jewelry model matching table records the matching suitability of each sample type and each of the three color models.

[0024] Beneficial effects: In this solution, by identifying the specific type of sample and combining it with the jewelry model matching table in the database, the most suitable color model can be selected for each type of jewelry and jade. This targeted selection ensures the accuracy of color parameter conversion. At the same time, the jewelry model matching table records the matching suitability of each sample type with each of the three color models, which makes the model selection process more scientific and quantitative, avoiding errors caused by subjective judgment.

[0025] The entire model matching process is highly automated, reducing the need for manual intervention and improving evaluation efficiency. Technicians only need to input basic information about the gemstone, and the system automatically identifies the type and selects the optimal color model.

[0026] Furthermore, the color model includes a CIE color model, an HSB color model, and a Munsell color model.

[0027] Furthermore, the uniform lighting module is a dome lighting LED light source, which provides a uniform diffuse reflection light source, and the illumination angle corresponding to the dome lighting LED light source is between 0 degrees and 90 degrees.

[0028] Furthermore, the system also includes a light source calibration module for placing a standard whiteboard on the lighting station before placing the jewelry and jade samples on the lighting station, and calibrating and calculating the environment and light source corresponding to the lighting station based on a preset calibration formula until the calibrated spectral image meets the preset requirements;

[0029] The preset calibration formula is:

[0030]

[0031] Where R c is the spectral image after calibration, I c is the spectral image before calibration, A c is the full brightness spectrum image, B c It is a full dark spectrum image.

[0032] Beneficial effects: The light source calibration module uses a standard white plate to calculate the calibration formula before each measurement, ensuring that the environment and light source at the lighting station meet the preset requirements. This eliminates errors caused by changes in light sources or environmental factors, ensuring consistency in each measurement condition.

[0033] The calibration formula ensures that the calibrated spectral image can accurately reflect the color characteristics of the jewelry, reducing measurement deviations caused by uneven light source intensity or environmental light interference.

[0034] Further, it further comprises a model training construction module for constructing and training the color grading model according to the model parameters corresponding to the color models matched by each historical jewelry sample in the historical database.

[0035] Further, the model training construction module comprises:

[0036] A historical data retrieval module for retrieving the model parameters corresponding to each historical jewelry sample output by the same color model in the historical database;

[0037] A labeling module for labeling the model parameters corresponding to each historical jewelry sample output by the same color model, and the labeling content is the color grade corresponding to the corresponding historical jewelry sample;

[0038] A division module for dividing the model parameters corresponding to each historical jewelry sample output by the same color model with labeling content into a training set and a validation set according to a preset division ratio;

[0039] A construction module for constructing a color grading model corresponding to each color model;

[0040] A random generation module for randomly generating a plurality of parameter groups corresponding to the color grading model corresponding to a certain color model when training the color grading model corresponding to the color model;

[0041] A training module for inputting the model parameters corresponding to each historical jewelry sample output by the same color model in the training set as input data into the color grading model corresponding to the corresponding color model, and outputting a training prediction grade result corresponding to each parameter group;

[0042] An evaluation module for evaluating the prediction reliability corresponding to each parameter group based on the training prediction grade result and the corresponding labeling content, and sorting each parameter group in descending order of prediction reliability to form a reliability ranking list;

[0043] An update module is used to eliminate some parameter groups in the reliability ranking list based on a preset ratio, pair the remaining parameter groups in the reliability ranking list, and randomly select a number of parameters to exchange them two by two to form a new parameter group. The training module is then re-executed until a preset number of iterations is reached. The color grading model corresponding to the parameter group ranked at a preset ranking in the reliability ranking list corresponding to the evaluation module at this time is regarded as the excellent model set;

[0044] The selection module is used to input the model parameters corresponding to each historical jewelry sample corresponding to the corresponding color model in the verification set as input data into the color grading model corresponding to the corresponding color model in the excellent model set, input the verification prediction grade results corresponding to each color grading model, and based on the verification prediction grade results corresponding to each color grading model and the annotation content corresponding to each model parameter in the corresponding verification set, select the color grading model with the highest prediction reliability as the final color grading model corresponding to the corresponding color model.

[0045] Beneficial Effects: This solution leverages existing data resources through the historical data retrieval module, providing a rich foundation for model training. This not only reduces data collection costs but also improves the starting point for model training. The labeling module ensures that each sample has an accurate color grade label, while the partitioning module divides the data into training and validation sets according to a preset ratio, ensuring the independence and fairness of the training and validation processes. The construction module can simultaneously process multiple color models (such as CIE, HSB, and Munsell), constructing corresponding color grading models for each, enabling multi-model parallel training and improving overall efficiency. The random generation module generates multiple parameter combinations during the initial training phase, increasing the possibilities for model exploration and facilitating the search for optimal solutions. The evaluation module quantitatively assesses the prediction reliability of each parameter group based on the training prediction grade results and annotations, generating a reliability ranking table. This evaluation mechanism ensures the scientific and objective nature of model selection. The update module removes inefficient parameter groups based on the reliability ranking table and continuously optimizes parameter combinations through pairwise pairing and random swapping. This process helps to gradually improve model performance.

[0046] The selection module further tests the color grading models from the excellent model set on the validation set to ensure their continued strong performance on unseen data. Based on the predictions from the validation set, the color grading model with the highest prediction reliability is selected as the final model. This process ensures that the model not only performs well on the training data but also has good generalization capabilities.

[0047] The present invention also provides a hyperspectral-based intelligent grading method for jewelry and jade colors, using the aforementioned hyperspectral-based intelligent grading system for jewelry and jade colors. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a logic block diagram of the hyperspectral-based intelligent grading system for jewelry and jade colors in Example 1 of the present invention. DETAILED DESCRIPTION

[0049] The following is further described in detail through specific implementation methods:

[0050] Example 1

[0051] The intelligent grading system for jewelry and jade color based on hyperspectral is basically as follows Figure 1 Shown, including:

[0052] A uniform lighting module is used to uniformly illuminate jewelry and jade samples placed on the lighting station; in this embodiment, the uniform lighting module is a dome lighting LED light source, which provides a uniform diffuse reflection light source, and the illumination angle corresponding to the dome lighting LED light source is between 0 degrees and 90 degrees.

[0053] The system also includes a light source calibration module, which is used to place a standard whiteboard on the lighting station before placing the jewelry and jade samples on the lighting station, and calibrate and calculate the environment and light source corresponding to the lighting station based on a preset calibration formula until the calibrated spectral image meets the preset requirements;

[0054] The preset calibration formula is:

[0055]

[0056] Where R c is the spectral image after calibration, I c is the spectral image before calibration, A c is the full brightness spectrum image, B c It is a full dark spectrum image.

[0057] Spectrum acquisition module, used to collect visible light spectra of jewelry and jade samples on the lighting station;

[0058] Spectral analysis module, used to perform spectral analysis on the visible light spectrum corresponding to the collected jewelry and jade samples and obtain the corresponding spectral reflectance;

[0059] An information acquisition module is used to acquire basic information of jewelry and jade samples placed on the lighting station;

[0060] The model matching module is used to match the color model corresponding to the jewelry and jade samples based on the obtained basic information of the samples and the preset model matching strategy;

[0061] The preset model matching strategy is:

[0062] Identify the sample type corresponding to the jewelry and jade samples based on their basic information;

[0063] The corresponding jewelry model matching table is retrieved from the database, and the color model corresponding to the sample type corresponding to the jewelry and jade sample is identified based on the sample type corresponding to the jewelry and jade sample and the retrieved jewelry model matching table; the jewelry model matching table records the matching suitability of each sample type and each of the three color models.

[0064] The color models include the CIE color model, the HSB color model, and the Munsell color model.

[0065] In this embodiment, the CIE color model converts the spectral reflectance into a three-dimensional uniform color space CIE1976L * a * b * , the calculation formula is:

[0066] L * =116f(Y / Y n )-16

[0067] a * =500[f(X / X n )-f(Y / Y n )]

[0068] b * =500[f(Y / Y n )-f(Z / Z n )]

[0069]

[0070] In the formula, X, Y, and Z are the three stimulus values ​​of the color stimulus of the object to be tested, X n 、Y n , Z n is the tristimulus value of the light reflected by the same light source as the object to be measured when it is illuminated by a completely diffuse reflector. n 、Y n , Z n That is the tristimulus value of the light source and Y n =100. In this color space, L * Represents brightness, +a * The axis approximately represents the direction of red stimulation, -a * The axis approximately represents the direction of green stimulus, +b * The axis approximately represents the direction of the yellow stimulus, -b * The axes approximately represent the blue stimulus directions.

[0071] In the HSB color model, the spectral reflectance is converted into the color parameters corresponding to the HSB color model, and the RGB color parameters are first normalized;

[0072]

[0073] Then perform HSB conversion on the normalized RGB color parameters according to the following formula:

[0074]

[0075] V(B)=MAX×100%

[0076] Where MAX is the maximum value of the RGB color parameters, and MIN is the minimum value of the RGB color parameters.

[0077] A spectrum conversion module is used to convert the spectral reflectance corresponding to the jewelry and jade sample into the color parameters corresponding to the color model matched by the jewelry and jade sample according to the color model matched by the jewelry and jade sample;

[0078] The grading module is used to output the color grade corresponding to the jewelry and jade sample based on the color parameters corresponding to the color model matched by the jewelry and jade and based on the preset color grading model.

[0079] It also includes a model training construction module for constructing and training a color grading model based on the model parameters corresponding to the color model matched by each historical jewelry and jade sample in the historical database.

[0080] The model training building block includes:

[0081] A historical data retrieval module is used to retrieve the model parameters corresponding to each historical jewelry and jade sample output by the same color model in the historical database;

[0082] A labeling module is used to label the model parameters corresponding to each historical jewelry sample output by the same color model, and the labeling content is the color grade corresponding to the corresponding historical jewelry and jade sample;

[0083] A partitioning module is used to divide the model parameters corresponding to each historical jewelry sample output by the same color model with annotated content into a training set and a validation set according to a preset partitioning ratio;

[0084] A construction module is used to construct a color grading model corresponding to each color model; the color grading model can be constructed using BP neural network, KNN, Bayesian and other methods. In this embodiment, a BP neural network model is used to construct the corresponding model. Specifically, a three-layer BP neural network model is first constructed, including an output layer, a hidden layer and an output layer. In this embodiment, the model parameters are input as input layer inputs, and the corresponding output is the color grade. BP neural networks usually use Sigmoid differentiable functions and linear functions as the excitation function of the network. This paper selects the S-type tangent function tansig as the excitation function of the hidden layer neurons. The prediction model selects the S-type logarithmic function tansig as the excitation function of the output layer neurons.

[0085] A random generation module, for randomly generating a plurality of parameter groups corresponding to the color grading model corresponding to a certain color model when training the color grading model corresponding to the color model;

[0086] The training module is used to input the model parameters corresponding to each historical jewelry sample output by the same color model in the training set as input data into the color grading model corresponding to the corresponding color model, and output the training prediction grade results corresponding to each parameter group;

[0087] The evaluation module is used to evaluate the prediction reliability corresponding to each parameter group based on the training prediction level results and the corresponding annotation content, and to sort the parameter groups in descending order of prediction reliability to form a reliability arrangement table; in this embodiment, the calculation logic corresponding to the prediction reliability evaluation is: through all the training prediction level results of the same parameter group and the corresponding annotation content, the corresponding accuracy is statistically calculated, which is used as the corresponding prediction reliability.

[0088] An update module is used to eliminate some parameter groups in the reliability ranking list based on a preset ratio, pair the remaining parameter groups in the reliability ranking list, and randomly select a number of parameters to exchange them two by two to form a new parameter group. The training module is then re-executed until a preset number of iterations is reached. The color grading model corresponding to the parameter group ranked at a preset ranking in the reliability ranking list corresponding to the evaluation module at this time is regarded as the excellent model set;

[0089] The selection module is used to input the model parameters corresponding to each historical jewelry sample corresponding to the corresponding color model in the verification set as input data into the color grading model corresponding to the corresponding color model in the excellent model set, input the verification prediction grade results corresponding to each color grading model, and based on the verification prediction grade results corresponding to each color grading model and the annotation content corresponding to each model parameter in the corresponding verification set, select the color grading model with the highest prediction reliability as the final color grading model corresponding to the corresponding color model. For example, a turquoise sample is placed in a uniform illumination module, focused using a microscope, and then the visible light spectrum of the sample is collected using a spectrum acquisition module. The collected optical signal is converted into an electrical signal and ultimately into a digital signal. The spectral information is converted into reflectance information, and the spectral reflectance is converted into tristimulus values ​​XYZ. The color parameters of the turquoise are digitized using the HSB color model. The hue H of the sample is 180.2°, the saturation S is 25.2%, and the lightness B is 62.3%. Using the corresponding color grading model, the color grade of the turquoise is intelligently graded. The hue of the turquoise is blue B, the saturation is medium S3, and the lightness is bright B2. Comprehensively, the turquoise color grade is determined to be the second grade. In this embodiment, the color grade is divided into four levels. Of course, the color grade in other embodiments can also be changed, such as three or five levels.

[0090] This embodiment also discloses a hyperspectral-based intelligent grading method for jewelry and jade colors, using the aforementioned hyperspectral-based intelligent grading system for jewelry and jade colors.

[0091] The above is only an embodiment of the present invention. Common knowledge such as the known specific structures and characteristics in the scheme is excessively described here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the enlightenment given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. Hyperspectral-based intelligent grading system for gemstone color, characterized by: include: Uniform lighting module, used to evenly illuminate the jewelry and jade samples placed on the lighting station; Spectrum acquisition module, used to collect visible light spectra of jewelry and jade samples on the lighting station; Spectral analysis module, used to perform spectral analysis on the visible light spectrum corresponding to the collected jewelry and jade samples and obtain the corresponding spectral reflectance; An information acquisition module is used to acquire basic information of jewelry and jade samples placed on the lighting station; The model matching module is used to match the color model corresponding to the jewelry and jade samples based on the obtained basic information of the samples and the preset model matching strategy; A spectrum conversion module is used to convert the spectral reflectance corresponding to the jewelry and jade sample into the color parameters corresponding to the color model matched by the jewelry and jade sample according to the color model matched by the jewelry and jade sample; The grading module is used to output the color grade corresponding to the jewelry and jade sample based on the color parameters corresponding to the color model matched by the jewelry and jade and based on the preset color grading model.

2. The hyperspectral-based intelligent grading system for gemstone colors according to claim 1, characterized in that: The preset model matching strategy is: Identify the sample type corresponding to the jewelry and jade samples based on their basic information; The corresponding jewelry model matching table is retrieved from the database, and the color model corresponding to the sample type corresponding to the jewelry and jade sample is identified based on the sample type corresponding to the jewelry and jade sample and the retrieved jewelry model matching table; the jewelry model matching table records the matching suitability of each sample type and each of the three color models.

3. The hyperspectral-based intelligent grading system for gemstone colors according to claim 2, characterized in that: The color models include the CIE color model, the HSB color model, and the Munsell color model.

4. The hyperspectral-based intelligent grading system for gemstone colors according to claim 3, characterized in that: The uniform lighting module is a dome lighting LED light source, which provides a uniform diffuse reflection light source. The illumination angle corresponding to the dome lighting LED light source is between 0 degrees and 90 degrees.

5. The hyperspectral-based intelligent grading system for gemstone colors according to claim 4, characterized in that: The system also includes a light source calibration module, which is used to place a standard whiteboard on the lighting station before placing the jewelry and jade samples on the lighting station, and calibrate and calculate the environment and light source corresponding to the lighting station based on a preset calibration formula until the calibrated spectral image meets the preset requirements; The preset calibration formula is: Where R c is the spectral image after calibration, I c is the spectral image before calibration, A c is the full brightness spectrum image, B c It is a full dark spectrum image.

6. The hyperspectral-based intelligent grading system for gemstone colors according to claim 5, characterized in that: It also includes a model training construction module for constructing and training a color grading model based on the model parameters corresponding to the color model matched by each historical jewelry and jade sample in the historical database.

7. The hyperspectral-based intelligent grading system for gemstone color according to claim 6, characterized in that: The model training building block includes: A historical data retrieval module is used to retrieve the model parameters corresponding to each historical jewelry and jade sample output by the same color model in the historical database; A labeling module is used to label the model parameters corresponding to each historical jewelry sample output by the same color model, and the labeling content is the color grade corresponding to the corresponding historical jewelry and jade sample; A partitioning module is used to divide the model parameters corresponding to each historical jewelry sample output by the same color model with annotated content into a training set and a validation set according to a preset partitioning ratio; A construction module for constructing a color grading model corresponding to each color model; A random generation module, for randomly generating a plurality of parameter groups corresponding to the color grading model corresponding to a certain color model when training the color grading model corresponding to the color model; The training module is used to input the model parameters corresponding to each historical jewelry sample output by the same color model in the training set as input data into the color grading model corresponding to the corresponding color model, and output the training prediction grade results corresponding to each parameter group; The evaluation module is used to evaluate the prediction reliability corresponding to each parameter group based on the training prediction level results and the corresponding annotation content, and to sort the parameter groups in descending order of prediction reliability to form a reliability ranking table; An update module is used to eliminate some parameter groups in the reliability ranking list based on a preset ratio, pair the remaining parameter groups in the reliability ranking list, and randomly select a number of parameters to exchange them two by two to form a new parameter group. The training module is then re-executed until a preset number of iterations is reached. The color grading model corresponding to the parameter group ranked at a preset ranking in the reliability ranking list corresponding to the evaluation module at this time is regarded as the excellent model set; The selection module is used to input the model parameters corresponding to each historical jewelry sample corresponding to the corresponding color model in the verification set as input data into the color grading model corresponding to the corresponding color model in the excellent model set, input the verification prediction grade results corresponding to each color grading model, and based on the verification prediction grade results corresponding to each color grading model and the annotation content corresponding to each model parameter in the corresponding verification set, select the color grading model with the highest prediction reliability as the final color grading model corresponding to the corresponding color model.

8. A hyperspectral-based intelligent grading method for gemstone color, characterized by: A hyperspectral-based intelligent grading system for gemstone color using any one of claims 1 to 7.

Images