Jewelry appraisal information integration management system based on visual image acquisition

Abstract

The invention discloses a jewelry appraisal information integration management system based on visual image acquisition. The system comprises a multi-dimensional visual acquisition module, a multi-factor adaptive correction module, a coupling interference decoupling and feature extraction module and a dynamic weight association management module. According to the system, multi-dimensional parameters of jewelry are synchronously collected, image correction, interference decoupling and information association are achieved through a three-layer linkage formula, and the problems of ambient light-material superposition distortion, coupling interference and association deviation are solved. According to the technical scheme, accurate integration and efficient tracing of jewelry identification information are achieved, information association reliability and query accuracy are guaranteed, and the method is suitable for a jewelry identification whole-process information management scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of visual image processing and information management, and particularly relates to a jewelry identification information integration management system based on visual image acquisition. BACKGROUND

[0002] In the jewelry identification process, image acquisition is a core link for obtaining jewelry characteristic information, and the quality thereof directly determines the accuracy of subsequent identification results and the reliability of information management. In the prior art, fixed-parameter visual acquisition equipment is often used for jewelry image acquisition, without fully considering the interaction between environmental light and the material characteristics of jewelry. The intensity and spectral distribution of environmental light are often in a dynamic state of change, while the optical characteristics such as the refractive index and surface roughness of the material of jewelry have significant individual differences, and the superposition of the two will cause nonlinear distortion of the acquired image. This distortion is not a simple change in brightness or contrast, but will distort the presentation form of key features such as the surface texture and internal inclusions of jewelry, making it difficult for subsequent feature extraction algorithms to accurately capture effective information. The existing fixed-parameter acquisition scheme cannot adaptively compensate for such dynamic superimposed interference, and can only alleviate it by manually adjusting the acquisition environment or equipment parameters, which is not only cumbersome and inefficient, but also relies on the experience of the operator, and it is difficult to ensure the consistency of the acquisition quality. This image acquisition distortion problem directly affects the subsequent identification information association link, resulting in a lack of accurate basis for the association of images and identification results and category information, and when querying and tracing, matching errors are prone to occur, which seriously affects the overall reliability of jewelry identification information management.

[0003] Based on the above problems, there is an urgent need for a technical solution that can adaptively compensate for the superimposed interference of environmental light and material, and ensure the accuracy of acquisition quality and information association. SUMMARY

[0004] The invention aims to solve the problems existing in the prior art and proposes a jewelry identification information integration management system based on visual image acquisition, which comprises a visual acquisition module for acquiring jewelry appearance feature images and internal feature images, an information management module for storing jewelry identification results and category information, and further comprises a multi-dimensional visual acquisition module, a multi-factor adaptive correction module, a coupling interference decoupling and feature extraction module, and a dynamic weight correlation management module; the multi-dimensional visual acquisition module synchronously acquires jewelry appearance feature images, internal feature images, environmental light parameters, jewelry geometric parameters, and jewelry material related parameters; the multi-factor adaptive correction module receives the parameters transmitted by the multi-dimensional visual acquisition module, calculates an image clarity correction value, dynamically adjusts the acquisition parameters according to the clarity correction value, and stops until the clarity correction value meets the preset requirements; the coupling interference decoupling and feature extraction module obtains corrected images based on the adjusted acquisition parameters, calculates an effective feature extraction rate, dynamically adjusts the feature extraction network layer weight according to the effective feature extraction rate, separates the coupling interference, and extracts the core features of the jewelry; the dynamic weight correlation management module compares the extracted core features with the identification standard library to obtain the identification results, calculates the identification information correlation confidence, and stores the images, identification results, category information, acquisition parameters, and correlation confidence in association; when receiving a query request, the information with a correlation confidence higher than a set threshold is matched preferentially.

[0005] Preferably, the multi-dimensional visual acquisition module comprises a multi-spectral camera, an angle-adjustable lens, an environmental light sensor, and a material parameter detection unit; the multi-spectral camera is used to acquire the appearance feature images and internal feature images of the jewelry under different wavelengths; the angle-adjustable lens is used to adjust the acquisition angle and adapt to jewelry of different shapes; the environmental light sensor is used to acquire the real-time environmental light intensity and the environmental light spectrum distribution coefficient; and the material parameter detection unit is used to obtain the jewelry surface roughness, material refractive index, and material optical uniformity coefficient.

[0006] Further preferably, the acquisition parameters adjusted by the multi-factor adaptive correction module include the camera exposure, the spectral channel gain, and the acquisition lens angle; during the adjustment process, the multi-factor adaptive correction module receives the multi-dimensional data corresponding to the adjusted acquisition parameters in real time, recalculates the image clarity correction value, and stops until the image clarity correction value reaches the clarity threshold range corresponding to the optimal feature extraction.

[0007] Further preferably, the coupling interference decoupling and feature extraction module adopts a deep learning model with a fusion attention mechanism; the feature extraction network layer weight adjustment direction includes convolution layer weight enhancement, pooling layer sampling frequency adjustment, and full connection layer neuron activation threshold optimization; and the core features include the jewelry surface texture feature, the internal inclusion distribution feature, the refractive index distribution feature, and the spectral response feature.

[0008] Further preferably, the image clarity correction value is calculated by the following formula:

[0009] ;

[0010] Where L0 is the standard ambient light intensity reference value, in lux (lx); L is the real-time ambient light intensity, in lux (lx); ε is the minimum value, which is 1×10⁻⁶. -6 lx is used to avoid a denominator of 0; K is the real-time ambient light spectral distribution coefficient, which is dimensionless; K n The standard spectral distribution coefficient adapted to the target jewelry category is dimensionless; r is the surface roughness of the jewelry, dimensionless; n is the refractive index of the target jewelry material, dimensionless; n0 is the standard refractive index reference value, dimensionless; δ is the optical homogeneity coefficient of the material, dimensionless; φ is the angle between the acquisition lens and the normal of the jewelry surface, in degrees (°).

[0011] Further optimization yields an effective feature extraction rate calculated using the following formula:

[0012] ;

[0013] Where S is the image sharpness correction value, dimensionless; θ is the real-time reflection angle of the jewelry surface, in degrees (°); ρ is the density of inclusions inside the jewelry, in pieces / m³; α is the sharpness deviation attenuation coefficient, dimensionless; S0 is the sharpness threshold corresponding to the optimal feature extraction, dimensionless; μ is the inclusion optical interference coefficient, dimensionless; λ is the wavelength of the collected light, in nanometers (nm); and d is the average particle size of the inclusions, in nanometers (nm).

[0014] Furthermore, the confidence level of the identification information association is calculated using the following formula:

[0015] ;

[0016] Where E is the effective feature extraction rate, dimensionless; ω is the importance weight of the identification index, dimensionless, dynamically allocated according to the jewelry category, with an allocation range of 0 to 1; β is the information timeliness impact coefficient, dimensionless; τ is the current information storage duration, in days (d); τ0 is the standard timeliness threshold, in days (d); γ is the clarity-extraction rate synergistic gain coefficient, dimensionless; ζ is the information attenuation coefficient, in 1 / day (1 / d); Δt is the time difference between the identification result and the image acquisition, in days (d).

[0017] Furthermore, the dynamic weight association management module also includes an identification standard library update unit; the identification standard library update unit regularly updates the feature comparison benchmarks and the importance weights of identification indicators in the identification standard library based on newly added jewelry identification cases and changes in industry standards; the update cycle is dynamically adjusted according to the number of jewelry categories and the frequency of industry standard updates.

[0018] Further optimized, during the query matching process, the dynamic weight association management module combines image clarity correction value and effective feature extraction rate to assist in verifying the matching accuracy; when there are multiple pieces of information with association confidence higher than the set threshold, they are sorted from high to low according to the association confidence value, and the information with the highest ranking is output first; when the association confidence is lower than the set threshold, the three pieces of information with the highest feature similarity are output, and a message indicating insufficient confidence is added.

[0019] Further preferably, it also includes an anomaly handling module; the anomaly handling module is used to monitor the operating status of each module, and when it detects abnormal acquisition parameters, interrupted data transmission, or calculation results exceeding a reasonable range, it triggers an anomaly alarm and starts a backup plan; the backup plan includes calling historical acquisition parameters, switching backup data transmission channels, or using a simplified calculation model to ensure the basic functions of the system.

[0020] The technical advantages of this invention are as follows: The core inventive technology lies in constructing a multi-dimensional parameter collaborative acquisition mechanism, a three-layer linkage quantitative calculation model, and a full-link dynamic collaborative structure. The multi-dimensional parameter collaborative acquisition mechanism breaks through the limitations of existing single-image acquisition, simultaneously acquiring multiple key parameters of the environment and the jewelry itself; the three-layer linkage formula realizes full-process quantitative control from acquisition and correction to information association, with the progressive influence between parameters ensuring accurate decision-making; the full-link dynamic collaborative structure enables closed-loop interaction among modules. This technical solution effectively solves the image distortion problem caused by the superposition of ambient light and material in the background technology, while also resolving coupling interference and information association deviations, ensuring the accuracy and reliability of jewelry appraisal information integration, and improving query and traceability efficiency. Attached Figure Description

[0021] Figure 1 This is a connection diagram of the jewelry appraisal information integration and management system based on visual image acquisition, as described in this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0023] Traditional technical solutions have the following technical problems: the superposition of dynamic changes in ambient light and optical properties of jewelry materials leads to nonlinear distortion of the acquired images. Existing fixed-parameter acquisition schemes cannot adaptively compensate for this, which seriously affects the accuracy of subsequent feature extraction and information association.

[0024] Based on this, please refer to Figure 1This embodiment provides a jewelry appraisal information integration and management system based on visual image acquisition. It includes a visual acquisition module for acquiring images of jewelry's external and internal features, an information management module for storing jewelry appraisal results and category information, and further includes a multi-dimensional visual acquisition module, a multi-factor adaptive correction module, a coupling interference decoupling and feature extraction module, and a dynamic weight association management module. The multi-dimensional visual acquisition module simultaneously acquires images of jewelry's external and internal features, ambient light parameters, jewelry geometric parameters, and jewelry material-related parameters. The multi-factor adaptive correction module receives the parameters transmitted by the multi-dimensional visual acquisition module and calculates the image sharpness. The system dynamically adjusts the acquisition parameters based on the sharpness correction value until the sharpness correction value meets the preset requirements. The coupling interference decoupling and feature extraction module acquires the corrected image based on the adjusted acquisition parameters, calculates the effective feature extraction rate, and dynamically adjusts the weights of the feature extraction network layer based on the effective feature extraction rate to separate coupling interference and extract the core features of the jewelry. The dynamic weight association management module compares the extracted core features with the identification standard library to obtain the identification result, calculates the association confidence of the identification information, and stores the image, identification result, category information, acquisition parameters, and association confidence as a whole. When receiving a query request, it prioritizes matching information with an association confidence higher than a set threshold.

[0025] This technical solution, centered on multi-module collaboration and quantitative calculation, constructs a comprehensive end-to-end management system. The multi-dimensional visual acquisition module, serving as the core data input, breaks away from the traditional single-image acquisition mode. It captures jewelry features at different wavelengths using a multispectral camera, with an adjustable lens adaptable to different jewelry shapes such as round and square. An ambient light sensor captures real-time changes in light intensity and spectrum. The material parameter detection unit directly acquires key parameters of the jewelry's surface and internal materials. All data is synchronized to subsequent modules via a high-speed transmission interface. The multi-factor adaptive correction module calculates sharpness correction values ​​based on multi-dimensional data, dynamically adjusting acquisition parameters to form a closed-loop control, ensuring that the quality of acquired images is unaffected by environmental and material changes. The coupling interference decoupling module quantifies interference impact through effective feature extraction rates, adjusting the weights of the deep learning model to achieve interference separation and accurately extract core features. The dynamic weight association management module, centered on association confidence, achieves accurate binding and efficient querying of multi-dimensional information. All modules form an organic whole through data transmission and feedback mechanisms, rather than being a simple superposition of independent operations.

[0026] The technical effects achieved by the above embodiments include: adaptive compensation for ambient light and material superposition interference, effective suppression of image distortion, improved accuracy of core feature extraction, quantitative basis for identification information association, accurate and efficient query and traceability, and significantly enhanced overall system reliability.

[0027] Traditional technical solutions have the following technical problems: existing acquisition modules have limited functionality, only able to acquire jewelry image information, and lack synchronous acquisition of key environmental and material parameters, resulting in a lack of data support for subsequent correction and feature extraction.

[0028] Based on this, the multi-dimensional visual acquisition module includes a multispectral camera, an angle-adjustable lens, an ambient light sensor, and a material parameter detection unit. The multispectral camera is used to acquire images of the appearance and internal features of jewelry at different wavelengths. The angle-adjustable lens is used to adjust the acquisition angle to adapt to jewelry of different shapes. The ambient light sensor is used to acquire real-time ambient light intensity and ambient light spectral distribution coefficient. The material parameter detection unit is used to obtain the surface roughness, refractive index, and optical uniformity coefficient of the jewelry.

[0029] This technical solution employs a structured design for the multi-dimensional visual acquisition module, clearly defining the functions and division of labor of each component. The multispectral camera utilizes an industrial-grade camera covering the visible to near-infrared bands, capable of switching between 10 different wavelength acquisition modes to optimally image different features such as jewelry surface textures and internal inclusions. The image resolution is set at 4096×3072 pixels to ensure detailed capture capabilities. The angle-adjustable lens adopts an integrated design of motorized zoom and angle adjustment, with an adjustment range of -45° to 45°. It can achieve 0.1° precision angle fine-tuning via motor drive, adapting to the placement of jewelry of different sizes and shapes. The ambient light sensor uses a high-sensitivity photodiode array, with a light intensity acquisition range of 10 lx to 10000 lx, and a spectral distribution coefficient acquisition covering the 400nm to 760nm visible light band. The sampling frequency is 10Hz, ensuring real-time capture of dynamic changes in ambient light. The material parameter detection unit integrates a laser roughness measurement module, a refractive index measurement module, and an optical uniformity detection module. The laser roughness measurement module acquires surface roughness data through the principle of laser scattering, the refractive index measurement module achieves non-contact measurement using a prism coupling method, and the optical uniformity detection module completes parameter acquisition based on interferometric imaging technology. All three work synchronously to ensure the temporal consistency between material parameters and image data.

[0030] The technical effects achieved by the above embodiments include: realizing the synchronous and accurate acquisition of multi-dimensional parameters of the environment and jewelry, providing comprehensive data support for subsequent correction and feature extraction, enhancing the adaptability of the acquisition module, and meeting the acquisition needs of different types of jewelry.

[0031] Traditional technical solutions have the following technical problems: the existing calibration process only adjusts a single acquisition parameter and lacks quantitative judgment standards, making it impossible to achieve precise control of acquisition quality.

[0032] Based on this, the multi-factor adaptive correction module adjusts the acquisition parameters including camera exposure, spectral channel gain, and acquisition lens angle. During the adjustment process, the multi-factor adaptive correction module receives multi-dimensional data corresponding to the adjusted acquisition parameters in real time, recalculates the image sharpness correction value, and continues until the image sharpness correction value reaches the sharpness threshold range corresponding to the optimal feature extraction.

[0033] This technical solution clearly defines the parameter adjustment range and closed-loop control logic of the correction module. The camera exposure adjustment range is from 1ms to 100ms, supporting step adjustments with 0.1ms accuracy, adapting to different light intensity environments by changing the exposure time. The spectral channel gain is adjusted independently for different wavelength channels, with an adjustment range of 1x to 10x, precisely optimizing the matching degree between the ambient light spectral distribution and the jewelry spectral response. The acquisition lens angle adjustment is linked with the angle-adjustable lens of the multi-dimensional vision acquisition module, achieving synchronous control through data interaction between modules. The adjustment process follows a closed-loop logic of calculation, adjustment, and recalculation. Each time the multi-factor adaptive correction module receives new multi-dimensional data, it immediately substitutes it into the formula to calculate the image sharpness correction value, compares it with the preset optimal feature extraction sharpness threshold range, and if it does not reach the threshold, it generates parameter adjustment instructions based on the direction and magnitude of the deviation. The adjustment magnitude is positively correlated with the deviation value until the sharpness correction value falls within the threshold range, ensuring that the acquired image quality reaches the optimal state.

[0034] The technical effects achieved by the above embodiments include: enabling coordinated and precise adjustment of multiple acquisition parameters, providing clear quantitative standards for the correction process, improving the stability of acquired image quality, and laying a good foundation for subsequent feature extraction.

[0035] Traditional technical solutions have the following technical problems: existing feature extraction models use a fixed network structure, which cannot cope with the changes in feature distribution caused by coupling interference, resulting in low efficiency of effective feature extraction.

[0036] Based on this, the coupling interference decoupling and feature extraction module adopts a deep learning model that integrates attention mechanism; the adjustment of the weights of the feature extraction network layer includes the enhancement of the weights of the convolutional layer, the adjustment of the sampling frequency of the pooling layer, and the optimization of the activation threshold of the neurons in the fully connected layer; the core features include the surface texture features of the jewelry, the distribution features of the internal inclusions, the refractive index distribution features, and the spectral response features.

[0037] This technical solution details the model structure and adjustment logic of the feature extraction module. The deep learning model incorporating the attention mechanism includes an input layer, convolutional layers, attention layers, pooling layers, fully connected layers, and an output layer. The input layer receives corrected image data with dimensions of 4096×3072×3. Five convolutional layers are used: the first three are feature extraction convolutional layers, and the last two are feature fusion convolutional layers. The kernel sizes are 7×7, 5×5, 3×3, 3×3, and 3×3, with 64, 128, 256, 256, and 512 neurons respectively. The attention layer employs a spatial attention mechanism, enhancing the response of key feature regions by calculating the importance weights of each location in the feature map. The pooling layer uses alternating max pooling and average pooling, with a 2×2 kernel size and a stride of 2. Three fully connected layers are used, with 1024, 512, and 256 neurons respectively. When adjusting the weights of the feature extraction network layers, the convolutional layer weights are enhanced by increasing the weight coefficients of the convolutional kernels corresponding to the core features of jewelry, thus strengthening the feature extraction capability. The pooling layer sampling frequency is adjusted by dynamically changing the sampling interval according to the effective feature extraction rate, increasing the sampling frequency when the extraction rate is low. The activation threshold of neurons in the fully connected layer is optimized by adjusting the threshold of the sigmoid function to select more discriminative features. In the process of extracting core features, surface texture features are extracted through the first three convolutional layers, internal inclusion distribution features are obtained by combining the attention layer and the last two convolutional layers, and refractive index distribution features and spectral response features are generated by fusing multispectral data through the fully connected layer.

[0038] The technical effects achieved by the above embodiments include: the model structure has the ability to dynamically adjust, effectively separates coupling interference, improves the accuracy and efficiency of core feature extraction, and provides high-quality feature data for identification results and information association.

[0039] Traditional technical solutions have the following technical problems: they lack quantitative methods for evaluating image sharpness and cannot accurately reflect the combined impact of ambient light and material characteristics on acquisition quality.

[0040] Based on this, the image sharpness correction value is calculated using the following formula:

[0041] ;

[0042] Where L0 is the standard ambient light intensity reference value, in lux (lx); L is the real-time ambient light intensity, in lux (lx); ε is the minimum value, which is 1×10⁻⁶. -6 lx is used to avoid a denominator of 0; K is the real-time ambient light spectral distribution coefficient, which is dimensionless; K nThe standard spectral distribution coefficient adapted to the target jewelry category is dimensionless; r is the surface roughness of the jewelry, dimensionless; n is the refractive index of the target jewelry material, dimensionless; n0 is the standard refractive index reference value, dimensionless; δ is the optical homogeneity coefficient of the material, dimensionless; φ is the angle between the acquisition lens and the normal of the jewelry surface, in degrees (°).

[0043] This technical solution achieves accurate evaluation of image sharpness through a multi-factor coupled quantitative formula. All parameters in the formula are derived from real-time data acquired by the multi-dimensional visual acquisition module or from a preset standard parameter library. The standard ambient light intensity benchmark value L0 is set to 500 lx according to the jewelry appraisal industry standard, suitable for most jewelry acquisition scenarios. The real-time ambient light intensity L is directly acquired by an ambient light sensor, ensuring a true reflection of the current ambient light intensity. The setting of the minimum value ε avoids the calculation anomaly of a denominator of 0 when the real-time ambient light intensity L approaches 0, balancing calculation accuracy and numerical stability. The real-time ambient light spectral distribution coefficient K is obtained by normalizing the spectral data acquired by the ambient light sensor, and the standard spectral distribution coefficient K... n Based on the differences in optical properties among different gemstones such as diamonds, rubies, and sapphires, corresponding values ​​are preset in a standard parameter library, such as the K value for diamonds. n The value is set to 0.85, the K value of the ruby. n The value is set to 0.72. The surface roughness r of the jewelry is obtained by dimensionless processing after being collected by the material parameter detection unit, and the value ranges from 0 to 1. The larger the value, the rougher the surface. The refractive index n of the target jewelry material is directly measured by the material parameter detection unit. The standard refractive index reference value n0 is set according to the type of jewelry material. For example, the n0 value is 0.37 for gold and 0.15 for silver. The optical uniformity coefficient δ of the material is detected by the material parameter detection unit, and the value ranges from 0 to 1. The larger the value, the better the optical uniformity. The angle φ between the acquisition lens and the normal of the jewelry surface is collected by the angle sensor of the angle-adjustable lens, which directly reflects the influence of the acquisition angle on the imaging. The formula couples the various influencing factors into a dimensionless sharpness correction value S through operations such as division, multiplication, absolute value, square root, and cosine function. The value ranges from 0 to 1. The closer the value is to 1, the higher the image sharpness, which is more conducive to subsequent feature extraction.

[0044] The core design logic of this formula is to specifically address the nonlinear image distortion caused by the superposition of dynamic changes in ambient light and the optical properties of jewelry materials. Through multi-factor coupled quantitative calculation, it achieves an accurate evaluation of the sharpness of the acquired image, providing a scientific basis for subsequent adjustment of acquisition parameters. The formula is not a simple accumulation of parameters, but a deep integration based on optical imaging principles and the characteristics of jewelry materials. Each calculation term corresponds to a specific interference compensation requirement, and the parameters work synergistically.

[0045] The standard ambient light intensity benchmark value L0 in the numerator is a benchmark parameter set according to the standard acquisition environment commonly used in the jewelry appraisal industry. Its function is to provide a reference anchor point for the correction of ambient light intensity, ensuring that the clarity evaluation in different environments has a unified standard. The design of summing the real-time ambient light intensity L and the minimum value ε in the denominator directly introduces the dynamic influence of real-time ambient light, and avoids the calculation anomalies that occur when L approaches 0 by controlling the value of ε, thus ensuring the numerical stability of the formula. This design fully considers the extreme cases of ambient light that may occur in actual acquisition scenarios.

[0046] The computation focuses on compensating for the coupling interference between the ambient light spectral distribution and the surface roughness of the jewelry. Absolute values ​​are used. This is because deviations in spectral distribution, whether positive or negative, negatively impact image sharpness, and the absolute value form accurately quantifies the degree of this deviation. Multiplying this deviation by the surface roughness *r* of the jewelry is because rough surfaces are more sensitive to spectral differences, and the coupling effect between the two exacerbates image distortion. Subtracting this product from 1 achieves a reverse correction of this coupling interference. The higher the match between the spectrum and the roughness, the closer the value of this operation term is to 1, and the better the effect on sharpness correction.

[0047] The calculation term corrects for individual differences in the refractive index of jewelry materials. Refractive index *n* is one of the core optical characteristics of jewelry, directly affecting light refraction and imaging effects. Using a square root form mitigates the excessive impact of refractive index deviations on sharpness evaluation, preventing drastic changes in the correction value due to small fluctuations in refractive index, and ensuring the stability of the evaluation results. The standard refractive index reference value *n0* is preset based on the characteristics of different jewelry materials, allowing this correction term to adapt to the differences in optical properties across various jewelry categories.

[0048] The computational terms achieve coordinated correction of material optical uniformity and acquisition angle. (Cosine function) The introduction of this method is based on the principle of optical imaging. Changes in the angle φ between the acquisition lens and the normal to the jewelry surface will cause changes in the intensity and direction of reflected light. The cosine function can accurately quantify the impact of this angle on imaging. When φ is 0°, cosφ takes the value of 1, with the angle having the least impact, which meets the optimal imaging conditions for orthogonal acquisition. The material optical uniformity coefficient δ is multiplied by cosφ and then added to 1. This reflects that the better the optical uniformity of the jewelry, the smaller the impact of angle changes on sharpness. The two form a synergistic correction logic to ensure that the sharpness evaluation under different uniformities and acquisition angles can accurately reflect the actual imaging quality.

[0049] The entire formula couples the correction terms of each dimension into a dimensionless sharpness correction value S through multiplication operations. The value ranges from 0 to 1, with the closer the value is to 1, the higher the image sharpness, which is more conducive to subsequent feature extraction. The design of this formula fully considers the dynamic changes and coupling effects of ambient light and the properties of jewelry materials. The selection of each parameter and the design of the calculation form are based on clear optical principles and practical applications, and can be understood and reproduced by those skilled in the art.

[0050] The technical effects achieved by the above embodiments include: realizing multi-factor quantitative evaluation of image clarity, accurately reflecting the comprehensive impact of environment and material on acquisition quality, and providing clear quantitative basis for adjusting acquisition parameters.

[0051] Traditional technical solutions have the following technical problems: existing feature extraction rate evaluation does not consider the coupling interference between reflection and inclusions, and cannot accurately reflect the degree of extraction of effective features.

[0052] Based on this, the effective feature extraction rate is calculated using the following formula:

[0053] ;

[0054] Where S is the image sharpness correction value, dimensionless; θ is the real-time reflection angle of the jewelry surface, in degrees (°); ρ is the density of inclusions inside the jewelry, in pieces / m³; α is the sharpness deviation attenuation coefficient, dimensionless; S0 is the sharpness threshold corresponding to the optimal feature extraction, dimensionless; μ is the inclusion optical interference coefficient, dimensionless; λ is the wavelength of the collected light, in nanometers (nm); and d is the average particle size of the inclusions, in nanometers (nm).

[0055] This technical solution introduces coupling interference factors on top of the sharpness correction value to achieve precise quantification of the effective feature extraction rate. The image sharpness correction value S in the formula directly uses the calculation results from the previous formula, ensuring the consistency of the evaluation. The real-time reflection angle θ of the jewelry surface is obtained collaboratively by the angle sensor of the angle-adjustable lens and the ambient light sensor, and is determined by analyzing the relationship between reflected light intensity and angle, with a value range of 0° to 90°. The density ρ of inclusions inside the jewelry is statistically obtained by the material parameter detection unit through image analysis technology, reflecting the density of inclusions; for example, the density ρ of inclusions inside a diamond may be 0.5 inclusions / mm³ to 5 inclusions / mm³. The sharpness deviation attenuation coefficient α is calibrated to 0.8 based on a large amount of experimental data, used to quantify the impact of the sharpness correction value S deviating from the optimal threshold S0 on the feature extraction rate. The sharpness threshold S0 corresponding to optimal feature extraction is set to 0.8 based on experimental data; that is, when S≥0.8, the impact of sharpness on feature extraction is relatively small. The inclusion optical interference coefficient μ is set based on the difference in optical properties between the inclusion material and the main jewelry material, ranging from 0 to 1; the greater the difference, the larger the value of μ. The acquisition wavelength λ is determined according to the current acquisition mode of the multispectral camera. For example, λ is set to 550nm when acquiring the surface texture of the jewelry, and 650nm when acquiring internal inclusions. The average particle size d of the inclusions is measured by the material parameter detection unit through image analysis technology, reflecting the size of the inclusions, for example, it may be between 100nm and 1000nm. The formula quantifies the coupling interference between reflection and inclusions through calculations such as tangent function, exponential function, and square root, finally obtaining the dimensionless effective feature extraction rate E, ranging from 0 to 1; the closer the value is to 1, the easier it is to extract the effective features.

[0056] It is worth mentioning that the real-time reflection angle θ on the jewelry surface is obtained by the angle sensor of the adjustable lens, the ambient light sensor, and the newly added reflection detection module. The reflection detection module receives the reflected light signal from the jewelry surface and converts it into an electrical signal. Combined with the lens angle data collected by the angle sensor and the light intensity distribution data collected by the ambient light sensor, the precise reflection angle θ is calculated by the data fusion algorithm. The reflection detection module maintains data acquisition synchronization with other sensors to ensure the accuracy and real-time nature of the θ value.

[0057] The core design purpose of this formula is to address the difficulty in effective feature extraction caused by the coupling interference between surface reflections and internal inclusions in jewelry. Based on the image sharpness correction value S, it further quantifies the impact of coupling interference on feature extraction, providing a precise quantitative basis for the feature extraction module to dynamically adjust network layer weights. The formula design follows a logical chain of "basic quality - interference compensation - collaborative optimization," with each operation term progressing layer by layer. It not only connects to the calculation results of previous formulas but also introduces targeted correction factors to address new technical problems.

[0058] The image sharpness correction value S calculated by the preceding formula is used as the basic factor of the formula because image sharpness is a prerequisite for effective feature extraction. A clear image can provide a good data foundation for feature extraction. This design ensures the coherence and logical consistency of the technical solution, and makes the sharpness correction and feature extraction rate evaluation form a closed loop.

[0059] The computational term is the core correction term for the coupling interference between reflection and inclusion. It employs the tangent function. Based on the nonlinear relationship between reflection angle and reflection intensity, when the real-time reflection angle θ of the jewelry surface increases, The tangent value grows non-linearly, accurately quantifying the variation of reflective intensity with angle. This non-linear quantification method better reflects the changing characteristics of reflective light in actual acquisition scenarios than a linear relationship. Multiplying this tangent value by the density ρ of inclusions within the jewelry is because the coupling interference between reflective light and inclusions is not independent. The higher the inclusion density, the easier it is for the reflected light spot to obscure the inclusion features. This coupling significantly reduces the extractability of effective features. By subtracting this product from 1, this coupling interference is quantitatively canceled out. The smaller the reflective angle and the lower the inclusion density, the closer the value of this operation term is to 1, resulting in better fundamental conditions for effective feature extraction.

[0060] The computation focuses on the attenuation effect of sharpness deviation on feature extraction rate. An exponential function is used because when the sharpness correction value S deviates from the optimal feature extraction threshold S0, the attenuation of the feature extraction rate exhibits a non-linear pattern: slight deviations result in small attenuations, while severe deviations lead to a sharp increase in attenuation. The exponential function can accurately fit this attenuation characteristic. The sharpness deviation attenuation coefficient α has been calibrated through extensive experiments to ensure that the quantification of the attenuation degree conforms to the actual feature extraction pattern, avoiding distortion of evaluation results due to inappropriate coefficient values. (Absolute value) The adoption of this method ensures that the degree of deviation can be accurately quantified regardless of whether S is higher or lower than S0, reflecting the rigor of the formula design.

[0061] The computational terms achieve coordinated optimization and correction of the acquisition light wavelength and the particle size of the inclusions. The ratio reflects the matching relationship between the acquisition wavelength λ and the inclusion particle size. When the acquisition wavelength λ is close to the average particle size d of the inclusion, the inclusion is more easily imaged clearly, thus improving the effective feature extraction rate. When the difference between the two is large, the inclusion features may be blurred or over-highlighted, affecting the recognition of effective features. The inclusion optical interference coefficient μ is introduced to quantify the difference in optical properties between the inclusion material and the main jewelry material. The greater the difference, the larger the value of μ, and the more obvious the interference on feature extraction. This coefficient is used to compare with... Multiplying the ratios allows for differentiated correction of interference from inclusions of different materials. The square root form is designed to mitigate the impact of this correction term on the feature extraction rate, preventing excessive deviations in the extraction rate evaluation due to parameter fluctuations and ensuring the stability of the evaluation results.

[0062] The entire formula, through the coupling of multiple operational terms, ultimately yields a dimensionless effective feature extraction rate E, ranging from 0 to 1. A value closer to 1 indicates that effective features are more easily extracted. This formula not only builds upon the achievements of previous technical solutions but also provides a quantitative solution to the new coupling interference problem. The physical meaning of each parameter and the selection of its operational form are based on clear technical principles, allowing those skilled in the art to substitute the parameter measurement results from actual data acquisition scenarios into the calculations.

[0063] The technical effects achieved by the above embodiments include: accurately quantifying the impact of the coupling interference between reflection and inclusion on feature extraction; the evaluation of effective feature extraction rate is more in line with the actual scenario; and providing a scientific basis for model weight adjustment.

[0064] Traditional technical solutions suffer from the following problems: existing identification information association relies solely on a single matching dimension, failing to consider factors such as feature validity and information timeliness, resulting in a lack of quantitative support for association reliability. Therefore, the confidence level of identification information association is calculated using the following formula:

[0065] ;

[0066] Where E is the effective feature extraction rate, dimensionless; ω is the importance weight of the identification index, dimensionless, dynamically allocated according to the jewelry category, with an allocation range of 0 to 1; β is the information timeliness impact coefficient, dimensionless; τ is the current information storage duration, in days (d); τ0 is the standard timeliness threshold, in days (d); γ is the clarity-extraction rate synergistic gain coefficient, dimensionless; ζ is the information attenuation coefficient, in 1 / day (1 / d); Δt is the time difference between the identification result and the image acquisition, in days (d).

[0067] This technical solution achieves precise quantification of the confidence level of identification information association through multi-dimensional factor coupling calculation. The effective feature extraction rate E in the formula follows the results of the previous formula, ensuring that the association evaluation is directly linked to the quality of feature extraction. The importance weight ω of the identification indicators is dynamically allocated according to the identification focus of the jewelry category. For example, in diamond identification, the weight indicator weight ω is set to 0.3, the color indicator weight to 0.25, the clarity indicator weight to 0.25, and the cut indicator weight to 0.2; in ruby ​​identification, the color indicator weight is set to 0.4, the clarity indicator weight to 0.3, the weight indicator weight to 0.2, and the cut indicator weight to 0.1. The information timeliness impact coefficient β is calibrated to 0.1 based on experimental data and is used to quantify the impact of information storage duration on association reliability. The current information storage duration τ is recorded by the dynamic weight association management module and accumulates from the moment the information storage is completed. The standard timeliness threshold τ0 is set to 365 days based on the validity period of jewelry identification information; after this period, the information association may decrease due to factors such as environmental changes. The sharpness-extraction rate synergistic gain coefficient γ is set to 0.5 to quantify the positive impact of the synergistic effect of S and E on the association confidence. The synergistic gain is more significant when both S and E are high. The information attenuation coefficient ζ is set to 0.00051 / d to quantify the impact of the time difference between the identification result and the image acquisition on the association reliability. The larger the time difference, the more obvious the information attenuation. The time difference Δt between the identification result and the image acquisition is recorded by the dynamic weight association management module, representing the duration from the completion of image acquisition to the generation of the identification result. The formula couples multiple factors such as feature validity, indicator weight, and information timeliness into a dimensionless association confidence C through operations such as division, multiplication, exponential function, and square root. The value ranges from 0 to 1, with a value closer to 1 indicating a higher association reliability between the image and the identification information.

[0068] The core design goal of this formula is to address the problem in existing technologies where the association between identification information and images relies solely on a single dimension and lacks quantitative evidence, leading to association bias. By integrating multiple dimensions such as effective feature extraction rate, identification index weight, and information timeliness, it achieves accurate quantification of the reliability of the association between images and identification results and category information, providing a scientific basis for query matching. The formula design follows the logic of "core quality - dimensional expansion - dynamic correction," organically integrating the calculation results of previous formulas with new association influencing factors to form a comprehensive association confidence evaluation system.

[0069] The product of the effective feature extraction rate E and the importance weight ω of the identification indicators forms the core basis of the association confidence. E directly reflects the effectiveness of the extracted jewelry features and is the core basis of the association. The higher the feature effectiveness, the stronger the reliability of the association. ω is dynamically allocated according to the identification focus of different jewelry categories, reflecting the differentiated importance of identification indicators. For example, cut and clarity indicators have higher weights in diamond identification, while color indicators have higher weights in ruby ​​identification. This dynamic allocation method makes the association evaluation more in line with the identification needs of different jewelry. The product of the two accurately quantifies the core impact of feature quality and indicator importance on the reliability of the association.

[0070] denominator The operational term is a non-linear correction term for information timeliness. An sigmoid exponential function is used because the impact of information storage time τ on the reliability of the association exhibits non-linear characteristics, within the standard timeliness threshold. Within τ0, information decays slowly, and the reliability of association does not decrease significantly; beyond τ0, information decay accelerates, and the reliability of association decreases significantly. The sigmoid function can accurately fit this change pattern. The value of the information timeliness impact coefficient β has been experimentally calibrated to ensure that the degree of timeliness correction conforms to the information decay pattern in actual scenarios. The difference reflects the degree of deviation of the information storage time from the standard threshold. Through the transformation of the exponential function, the reliability of the information association for different storage times is differentiated and corrected, so that the value of this operation term is between 1 and 2. This ensures the effectiveness of the timeliness correction and avoids the weakening of the impact of the core feature quality due to overcorrection.

[0071] The computational term achieves synergistic gain correction between image sharpness and effective feature extraction rate. The ratio reflects the matching degree between image sharpness and effective feature extraction rate. When both S and E are high and the ratio is close to 1, it indicates that the image acquisition quality and feature extraction quality are improving simultaneously, and their synergistic effect can further enhance the reliability of the association. When the ratio deviates from 1, it indicates that either the image acquisition quality has not been fully converted into feature effectiveness, or the feature extraction quality exceeds the support range of the image acquisition quality, and the synergistic effect weakens. The sharpness-extraction rate synergistic gain coefficient γ is introduced to quantify the strength of this synergistic effect. Experimental calibration ensures that the quantification of the synergistic gain conforms to the actual association law. The square root form design mitigates the impact of the synergistic gain on the association confidence, ensuring the stability of the evaluation results and avoiding drastic changes in confidence due to ratio fluctuations.

[0072] The computational term is a linear correction term applied to the time difference between the identification result and the image acquisition. The larger the time difference Δt between the identification result and the image acquisition, the more likely the jewelry's condition may change due to environmental variations, wear and tear, etc., leading to a decrease in the consistency between the image and the identification result, and a reduction in the reliability of the association. The information attenuation coefficient ζ was determined through extensive experiments, quantifying the degree of attenuation of association reliability per unit time difference. The linear quantification of the impact of the time difference is achieved through the product of ζ and Δt. The form 1 minus this product accurately reflects the rule that the larger the time difference, the lower the association reliability. This linear correction method is simple and conforms to the gradual nature of the time difference's impact on association in real-world scenarios.

[0073] The entire formula, through the coupled calculation of multiple dimensions, ultimately yields a dimensionless confidence score C for the identification information, ranging from 0 to 1. A value closer to 1 indicates higher reliability. The formula integrates various influencing factors such as feature quality, indicator weight, timeliness, and collaborative matching degree. The selection of each parameter and the design of the calculation method have clear technical logic and practical application basis. Those skilled in the art can substitute the parameter measurement and setting results from actual scenarios into the calculation.

[0074] The technical effects achieved by the above embodiments include: realizing multi-dimensional quantitative evaluation of the reliability of identification information association, the association confidence degree has a clear physical meaning, providing accurate basis for query matching, and improving the scientificity and reliability of information association.

[0075] Traditional technical solutions have the following technical problems: the existing identification standard library is fixed and cannot adapt to new identification cases and changes in industry standards, resulting in a decrease in the accuracy of identification results.

[0076] Based on this, the dynamic weight association management module also includes an identification standard library update unit; the identification standard library update unit regularly updates the feature comparison benchmarks and the importance weights of identification indicators in the identification standard library according to newly added jewelry identification cases and changes in industry standards; the update cycle is dynamically adjusted according to the number of jewelry categories and the frequency of industry standard updates.

[0077] This technical solution employs a dynamic update mechanism for its standard library to ensure the timeliness and accuracy of the identification standards. The standard library update unit includes a case acquisition module, a standard interpretation module, a weight optimization module, and an update execution module. The case acquisition module receives new jewelry identification cases through a data interface. Cases must include complete jewelry feature images, material parameters, identification results, and identification basis. After format verification and duplicate filtering, they are stored in the case database. The standard interpretation module monitors the release and update information of jewelry identification industry standards in real time, provides technical interpretations of newly released industry standards, extracts core content related to feature comparison and indicator weights, and transforms it into standard data recognizable by the system. The weight optimization module, based on the new case data, uses statistical analysis methods to calculate the discrimination of each identification indicator. Higher discrimination indicates a greater impact of the indicator on the identification result, and the importance weight of the identification indicators is dynamically adjusted. For example, when the discrimination of a certain type of jewelry's clarity indicator significantly improves in a new case, the corresponding weight will be appropriately increased. The update execution module triggers standard library updates based on the update cycle, which is initially set to 90 days. When the number of new cases exceeds 1000 or there is a major update to the industry standard, the update cycle is automatically shortened to 30 days. When the number of new cases is less than 100 and there is no industry standard update, the update cycle is extended to 180 days. During the update process, the system retains historical standard data, supports comparative analysis of identification results before and after the update, and ensures the traceability of the update process.

[0078] The technical effects achieved by the above embodiments include: the identification standard library has dynamic update capability, adapts to industry development and case accumulation, improves the accuracy and timeliness of identification results, and continuously expands the scope of system application.

[0079] Traditional technical solutions have the following technical problems: existing query matching only sorts based on a single correlation indicator and does not consider multi-dimensional auxiliary verification, which affects matching accuracy and user experience.

[0080] Based on this, during the query matching process, the dynamic weight association management module combines image clarity correction value and effective feature extraction rate to assist in verifying the matching accuracy; when there are multiple pieces of information with association confidence higher than the set threshold, they are sorted from high to low according to the association confidence value, and the information with the highest ranking is output first; when the association confidence is lower than the set threshold, the three pieces of information with the highest feature similarity are output, and a message indicating insufficient confidence is added.

[0081] This technical solution optimizes the query matching logic, improving matching accuracy and user experience. The query matching process is divided into three stages: preliminary screening, accuracy verification, and sorted output. In the preliminary screening stage, the dynamic weight association management module receives user-input query conditions, including jewelry category, feature keywords, and authentication time. It filters based on association confidence, setting a threshold of 0.7 to select candidate information higher than this threshold. In the accuracy verification stage, candidate information is further verified using image clarity correction values ​​and effective feature extraction rates. A verification score is calculated as: Verification Score = 0.6 × Association Confidence + 0.2 × Image Clarity Correction Value + 0.2 × Effective Feature Extraction Rate. Information with verification scores higher than 0.75 is further filtered to ensure matching accuracy. In the sorted output stage, when the number of filtered candidate information is greater than one, it is sorted from highest to lowest association confidence value. Information with the highest ranking is output to the user first, and key indicators such as association confidence, image clarity correction value, and effective feature extraction rate are displayed for user reference. When no relevant information with a confidence level higher than the set threshold is found after initial screening, the system activates the feature similarity matching mechanism to calculate the feature similarity between the query conditions and the information in the database. The system then selects the three pieces of information with the highest similarity and outputs them. Each piece of information is marked with a message indicating insufficient confidence and for reference only. The system also provides a feature difference comparison interface to help users analyze the rationality of the matching results.

[0082] The technical effects achieved by the above embodiments include: query matching combined with multi-dimensional indicator verification improves matching accuracy, output results are more in line with user needs, and user experience and query efficiency are significantly improved.

[0083] Traditional technical solutions have the following technical problems: existing systems lack a sound exception handling mechanism, and function interruption is easily caused when modules malfunction, resulting in insufficient system stability.

[0084] Based on this, an exception handling module is also included; the exception handling module is used to monitor the operating status of each module. When an abnormality in the collected parameters, interruption of data transmission, or calculation result exceeding the reasonable range is detected, an exception alarm is triggered and a backup plan is started; the backup plan includes calling historical collected parameters, switching to a backup data transmission channel, or using a simplified calculation model to ensure the basic functions of the system.

[0085] This technical solution establishes a comprehensive anomaly handling mechanism to enhance system stability and fault tolerance. The anomaly handling module includes a status monitoring unit, an anomaly determination unit, an alarm triggering unit, and a backup plan execution unit. The status monitoring unit collects the operating parameters of each module in real time, including the acquisition parameters of the multi-dimensional visual acquisition module, the data transmission rate of each module, and the output results of the calculation module, with a sampling frequency of 1Hz to ensure timely detection of anomalies.

[0086] The anomaly detection unit presets reasonable ranges for each operating parameter. For example, the reasonable range for exposure in the acquisition parameters is 1ms to 100ms, the reasonable range for data transmission rate is 10Mbps to 100Mbps, and the reasonable range for sharpness correction value in the calculation results is 0 to 1. When a parameter is detected to be outside the reasonable range for more than 2 seconds, it is judged as an anomaly. The alarm triggering unit immediately triggers an alarm after an anomaly is detected. The alarm forms include a system interface pop-up notification, an audible alarm, and log recording. Simultaneously, an anomaly notification is sent to the system administrator, specifying the abnormal module, anomaly type, and occurrence time. The backup plan execution unit activates the corresponding backup plan based on the anomaly type: when an abnormal acquisition parameter is detected, historical parameters from the most recent normal acquisition are used as temporary acquisition parameters, and the user is prompted to check the acquisition equipment; when a data transmission interruption is detected, automatic switching to the backup data transmission channel is initiated. The backup channel uses a different communication protocol than the main channel to ensure data transmission continuity; when a calculation result is detected to be outside the reasonable range, a simplified calculation model is used to replace the original model. The simplified model retains the core calculation logic, reduces complex calculation steps, and ensures the system's basic acquisition, feature extraction, and information association functions, while continuously monitoring the anomaly status. Once the anomaly is resolved, automatic switching back to the original plan is initiated.

[0087] It is worth mentioning that the information management module reserves a storage area for historical acquisition parameters, which is dedicated to storing complete parameter data for each normal acquisition. The stored content includes all acquisition-related parameters such as camera exposure, spectral channel gain, and acquisition lens angle. The stored data is arranged in order by acquisition timestamp for easy retrieval. When abnormal acquisition parameters are detected, the system retrieves the historical parameters from the most recent normal acquisition as temporary acquisition parameters from this storage area, and prompts the user to check the acquisition equipment.

[0088] The technical effects achieved by the above embodiments include: the system has complete exception handling and fault tolerance capabilities, the function is not interrupted when the module is running abnormally, the system stability and reliability are significantly improved, and the user experience is guaranteed.

[0089] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A jewelry identification information integration management system based on visual image acquisition, comprising a visual acquisition module for acquiring images of external features and internal features of jewelry, and an information management module for storing jewelry identification results and category information, characterized in that, Also include multi-dimensional visual acquisition module, multi-factor adaptive correction module, coupled interference decoupling and feature extraction module and dynamic weight correlation management module; multi-dimensional visual acquisition module synchronously acquires jewelry appearance feature image, internal feature image, environmental light parameter, jewelry geometric parameter and jewelry material related parameter; multi-factor adaptive correction module receives the parameter transmitted by the multi-dimensional visual acquisition module, calculates the image sharpness correction value, dynamically adjusts the acquisition parameter according to the image sharpness correction value, until the image sharpness correction value meets the preset requirement; the coupled interference decoupling and feature extraction module obtains the corrected image based on the adjusted acquisition parameter, calculates the effective feature extraction rate, dynamically adjusts the feature extraction network layer weight according to the effective feature extraction rate, separates the environmental light and material coupling interference and extracts the jewelry core feature; The dynamic weight correlation management module compares the extracted jewelry core feature with the identification standard library to obtain the identification result, calculates the identification information correlation confidence, and stores the corrected image, identification result, category information, acquisition parameter and identification information correlation confidence in association, and when receiving a query request, the information with identification information correlation confidence higher than the set threshold is matched preferentially.

2. The jewelry identification information integration management system based on visual image acquisition according to claim 1, characterized in that, The multi-dimensional visual acquisition module includes a multispectral camera, an angle adjustable lens, an environmental light sensor and a material parameter detection unit; the multispectral camera is used for acquiring the appearance feature image and internal feature image of the jewelry under different wavelengths; the angle adjustable lens is used for adjusting the acquisition angle to adapt to different shapes of jewelry; the environmental light sensor is used for acquiring real-time environmental light intensity and environmental light spectral distribution coefficient; the material parameter detection unit is used for acquiring the surface roughness, material refractive index and material optical uniformity coefficient of the jewelry.

3. The visual image acquisition based jewelry identification information integration management system according to claim 1, wherein, The acquisition parameters adjusted by the multi-factor adaptive correction module include camera exposure, spectral channel gain and acquisition lens angle; during the adjustment process, the multi-factor adaptive correction module receives the multi-dimensional data corresponding to the adjusted acquisition parameters in real time, recalculates the image sharpness correction value, until the image sharpness correction value reaches the sharpness threshold range corresponding to the best feature extraction.

4. The jewelry identification information integration management system based on visual image acquisition according to claim 1, characterized in that, The coupled interference decoupling and feature extraction module adopts a deep learning model with fusion attention mechanism; the feature The feature extraction network layer weight adjustment direction includes convolution layer weight enhancement, pooling layer sampling frequency adjustment and full connection layer neuron activation threshold optimization; the jewelry core feature includes jewelry surface texture feature, internal inclusion distribution feature, refractive index distribution feature and spectral response feature.

5. The visual image acquisition based jewelry identification information integration management system according to claim 1, wherein, The image sharpness correction value (S) is calculated by the following formula: ； Wherein, S is the image definition correction value, dimensionless, the value range is 0 to 1, the value is closer to 1, the higher the image definition, the more conducive to subsequent feature extraction;L0 is the standard ambient light intensity reference value, the dimension is lux (lx);L is the real-time ambient light intensity, the dimension is lux (lx);ε is the minimum value, the value is 1×10⁻ 6 lx, to avoid the denominator being 0;K is the real-time ambient light spectral distribution coefficient, dimensionless;K n is the standard spectral distribution coefficient adapted to the target jewelry category, dimensionless;r is the jewelry surface roughness, dimensionless;n is the refractive index of the target jewelry material, dimensionless;n0 is the standard refractive index reference value, dimensionless;δ is the material optical uniformity coefficient, dimensionless;φ is the angle between the collection lens and the jewelry surface normal, the dimension is degree (°).

6. The jewelry identification information integration management system based on visual image acquisition according to claim 5, characterized in that, The effective feature extraction rate (E) is calculated by the following formula: ； Wherein, E is the effective feature extraction rate, dimensionless, the value range is 0 to 1, the value is closer to 1, the effective feature is easier to extract; S is the image definition correction value, dimensionless; θ is the real-time reflection angle of the jewelry surface, the dimension is degree (°); ρ is the density of the inclusions in the jewelry, the original dimension is individual per cubic millimeter (individual / mm3), and the dimensionless processing is dimensionless; α is the definition deviation attenuation coefficient, dimensionless; S0 is the definition threshold corresponding to the optimal feature extraction, dimensionless; μ is the inclusion optical interference coefficient, dimensionless; λ is the wavelength of the collected light, the dimension is nanometer (nm); d is the average particle size of the inclusion, the dimension is nanometer (nm).

7. The jewelry identification information integration management system based on visual image acquisition according to claim 6, characterized in that, The identification information correlation confidence (C) is calculated by the following formula: ； Wherein, C is the identification information correlation confidence, dimensionless, the value range is 0 to 1, the value is closer to 1, the higher the reliability of the image and the identification information; E is the effective feature extraction rate, dimensionless; ω is the importance weight of the identification index, dimensionless, dynamically allocated according to the jewelry category, the allocation range is 0 to 1; β is the information timeliness influence coefficient, the dimension is 1 per day (1 / d); τ is the current information storage time, the dimension is day (d); τ0 is the standard timeliness threshold, the dimension is day (d); γ is the definition-extraction rate synergistic gain coefficient, dimensionless; ζ is the information attenuation coefficient, the dimension is 1 per day (1 / d); Δt is the time difference between the identification result and the image collection, the dimension is day (d).

8. The visual image acquisition based jewelry identification information integration management system according to claim 1, wherein, The dynamic weight correlation management module also includes an identification standard library updating unit; the identification standard library updating unit updates the feature comparison reference and the importance weight of the identification index in the identification standard library according to the newly added jewelry identification cases and the changes of the industry standards; The update cycle is dynamically adjusted according to the number of jewelry categories and the industry standard update frequency.

9. The visual image acquisition based jewelry identification information integration management system according to claim 1, wherein, In the query matching process, the dynamic weight correlation management module combines the image definition correction value and the effective feature extraction rate to assist in verifying the matching accuracy; when there are multiple information with correlation confidence higher than the set threshold, the information is sorted from high to low according to the correlation confidence value, and the information with high ranking is preferentially output; when the correlation confidence is lower than the set threshold, the three information with the highest feature similarity are output, and the insufficient confidence prompt is marked.

10. The jewelry identification information integration management system based on visual image acquisition according to claim 1, characterized in that, It also includes an abnormality processing module; the abnormality processing module is used for monitoring the running state of each module, triggering an abnormal alarm and starting a backup scheme when detecting that the collection parameters are abnormal, the data transmission is interrupted or the calculation result is out of the reasonable range; the backup scheme includes calling the historical collection parameters, switching the backup data transmission channel or adopting a simplified calculation model to ensure the basic function of the system.

Images