Gold industry e-commerce SaaS cloud service implementation method and system

Abstract

The invention provides a gold industry e-commerce SaaS cloud service realization method and system, and the method comprises the steps: collecting and extracting the dimension features of a hyperspectral feature image, a polarization feature image and a defect mask image of a gold product in real time, and splicing the dimension features to form a dimension reduction fusion feature vector; on the basis of the distribution situation of the dimensionality reduction fusion feature vector in a low-dimensional feature space, constructing a feature manifold reflecting the oxidation evolution continuity of the gold coating, and selecting and calculating the local density entropy and the feature vector field divergence of gold coating sample points along an oxidation principal axis; the invention aims to solve the problem that the development of the gold industry e-commerce is seriously restricted due to the fact that a consumer cannot truly observe the color and texture of a gold product through an image because the space-time difference between the thickness distribution of a gold surface oxide layer and the reflectivity of a plating layer is difficult to analyze by the existing SaaS cloud service implementation method for the gold industry e-commerce.

Description

Technical Field

[0001] This invention relates to the field of e-commerce service technology in the gold industry, specifically to a method and system for implementing e-commerce SaaS cloud services in the gold industry. Background Technology

[0002] E-commerce SaaS cloud services for the gold industry constitute a comprehensive e-commerce business management system encompassing image processing, marketing, and delivery of gold. However, due to environmental factors, the surface of gold is prone to forming a double-layer dielectric film structure consisting of a micron-level oxide layer and an electroplating layer. The presence of this double-layer dielectric film structure causes the acquired gold images to easily exhibit non-uniform rainbow diffraction patterns, severely compromising the reproduction of the gold's true color.

[0003] Although existing e-commerce SaaS cloud services for the gold industry have improved image quality through operations such as correcting image rendering material parameters or replacing materials in the oxide layer distribution area, it is still difficult to analyze the spatiotemporal differences in the thickness distribution of the oxide layer on the gold surface and the reflectivity of the coating. This makes it impossible for consumers to truly observe the color and texture of the gold product itself through the image, which seriously restricts the development of e-commerce in the gold industry.

[0004] Therefore, how to analyze the spatiotemporal differences between the thickness distribution of the oxide layer on the gold surface and the reflectivity of the coating has become a question worthy of consideration by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing e-commerce SaaS cloud service implementation methods in the gold industry are unable to analyze the spatiotemporal differences in the thickness distribution of the oxide layer on the gold surface and the reflectivity of the coating, which makes it impossible for consumers to truly observe the color and texture of the gold product itself through images, thus seriously restricting the development of e-commerce in the gold industry.

[0006] To achieve the above objectives, this invention provides a method and system for implementing e-commerce SaaS cloud services in the gold industry.

[0007] According to a first aspect of the present invention, a method for implementing an e-commerce SaaS cloud service in the gold industry is provided, comprising:

[0008] The dimensional features of hyperspectral feature images, polarization feature images, and defect mask images of gold products are acquired and extracted in real time, and then stitched together to form a dimensionality-reduced fusion feature vector.

[0009] Based on the distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, a feature manifold reflecting the continuity of the oxidation evolution of the gold coating is constructed. The local density entropy and feature vector field divergence of the gold coating sample points are selected and calculated along the oxidation principal axis.

[0010] The transition point of the gold plating thickness variation is defined as the abrupt change point of the local density entropy and the eigenvector field divergence. Several abrupt change points are automatically identified by the Bayesian change point detection algorithm to dynamically classify the order level and core situation point.

[0011] In the feature manifold space, the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value level is calculated, and the order value level of the gold product is determined according to the situation interval in which the geodesic distance is located.

[0012] A reflectivity anomaly detection model is trained and generated by pre-acquired benchmark coating reflectivity feature set, and a dimension-reduced fusion feature vector is input into the reflectivity anomaly detection model to output a reflectivity detection signal.

[0013] The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

[0014] Optionally, the method of constructing a feature manifold reflecting the continuity of the oxidation evolution of the gold coating based on the distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, and selecting and calculating the local density entropy and feature vector field divergence of the gold coating sample points along the oxidation principal axis, specifically includes:

[0015] A nonlinear dimensionality reduction algorithm is used to map the dimensionality-reduced and fused feature vectors into a 2D low-dimensional space, and the spatial neighborhood relationship of the oxidation state of the gold coating is extracted. A local connection structure is constructed based on the K-nearest neighbor graph, and a feature manifold reflecting the continuity of the oxidation evolution of the gold coating is formed through a local linear embedding algorithm.

[0016] Based on the distribution density of sample points in the characteristic manifold, the density peak clustering algorithm is used to automatically identify high-density regions and define them as the core clusters of the oxidation state of the gold coating. The direction corresponding to the largest eigenvalue in the eigenvector of the covariance matrix of the core cluster is defined as the oxidation principal axis.

[0017] For each sample point as the center, the number of sample points in the neighborhood of its distance radius R is counted to obtain the local density. The probability density function of the local density is calculated by the kernel density estimation algorithm, and the local density entropy is calculated based on the probability density function and the information entropy formula.

[0018] Using each sample point in the characteristic manifold space as the base point, the direction of oxidation state change of the K nearest neighbor sample points is calculated to construct the eigenvector field. The divergence of the vector field at each sample point is calculated by the gradient estimation algorithm to obtain the eigenvector field divergence.

[0019] Optionally, after automatically identifying high-density regions and defining them as core clusters of the gold plating oxidation state using a density peak clustering algorithm based on the distribution density of sample points in the characteristic manifold, and defining the direction corresponding to the largest eigenvalue in the covariance matrix eigenvector of the core cluster as the oxidation principal axis according to PCA calculation, the method further includes:

[0020] The oxygen principal axis direction of different samples is aligned using a dynamic time warping algorithm.

[0021] Optionally, the transition point of the gold plating thickness variation is defined as the abrupt change point of the local density entropy and eigenvector field divergence, and several abrupt change points are automatically identified using a Bayesian change point detection algorithm to dynamically classify the order level and core situation points, specifically including:

[0022] A sliding window with a width of 5%-10% of the total number of samples and a step size of 1 is used. For each sample point, the mean and standard deviation of the local density entropy and the eigenvector field divergence within its window are calculated. Sample points whose local density entropy exceeds the sum of the current window mean and twice the standard deviation, and whose eigenvector field divergence exceeds the sum of the current window mean and 1.5 times the standard deviation, are marked as mutation candidate points. The position and number of mutation candidate points are iteratively calculated using a Bayesian change point detection algorithm to automatically identify several mutation points.

[0023] If the number of mutation points is N, the feature manifold is divided into N+1 oxidation stage regions based on the N mutation points. The local density entropy and feature vector field divergence of the sample points in each oxidation stage region are standardized by Z-score, so that the oxidation state can be divided into thin oxygen level, medium oxygen level and thick oxygen level by K-means clustering algorithm.

[0024] The local density of sample points in each oxidation stage region is calculated and sorted based on the DBSCAN algorithm. The sample points with the highest local density in the top 10% are selected as candidate points. The cosine of the divergence direction between each candidate point and its five nearest neighbor sample points is calculated. Each candidate point with a cosine of the angle with its nearest neighbor sample points that is greater than 0.8 and has more than 3 such points is selected as the core situation point.

[0025] Optionally, the step of calculating the geodesic distance between the dimensionality-reduced fused feature vector and the core situation points corresponding to each order level in the feature manifold space, and determining the order level of the gold product based on the situation interval where the geodesic distance is located, specifically includes:

[0026] Map all the output core situation points to the feature manifold space and obtain the coordinates of each core situation point;

[0027] Based on the Dijkstra algorithm, the distance between each sample point in all sample points and each core situation point with the shortest distance is calculated and output as geodesic distance;

[0028] Based on the geodesic distance distribution of all core situation points, the characteristic manifold space is divided into thin oxygen interval, medium oxygen interval and thick oxygen interval using the quantile method.

[0029] Based on the geodesic distance interval of each sample point, the corresponding order value is matched to determine the order value of the gold product.

[0030] Optionally, before dividing the characteristic manifold space into thin-oxygen, medium-oxygen, and thick-oxygen intervals using the quantile method based on the geodesic distance distribution of all core situation points, the method further includes:

[0031] Using the initial geodesic distance of each sample point as the weight, sample points within its radius R neighborhood are selected for local weighted regression processing, and the geodesic distance is corrected through iterative calculation.

[0032] Optionally, in addition to matching the corresponding order value based on the geodesic distance interval of each sample point to determine the order value of the gold product, the method also includes:

[0033] If the geodesic distance of any sample point lies between two interval boundaries, the nearest neighbor principle is used to determine the order value.

[0034] Optionally, the step of training and generating a reflectivity anomaly detection model using a pre-acquired reference coating reflectivity feature set, and inputting a dimension-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal, specifically includes:

[0035] While splicing together to form a dimension-reduced fusion feature vector, a spectrophotometer is used to measure the reflectance value of the gold coating. The dimension-reduced fusion feature vector and the reflectance value are combined to obtain the reflectance feature set of the reference coating.

[0036] Using the baseline coating reflectance feature set as input, a reflectance anomaly detection model is trained and generated through the isolated forest algorithm.

[0037] The dimension-reduced and fused feature vectors are input into the reflectivity anomaly detection model to output the reflectivity detection signal;

[0038] The quality grade of the gold product image is determined based on the grade value and reflectivity detection signal.

[0039] Optionally, the image quality grade of the gold product can be determined based on the grade value and reflectivity detection signal, specifically including:

[0040] When a thin oxygen level value and a normal reflectance detection signal are input, a high grade is output and an image processing termination command is executed.

[0041] When the input is a thin oxygen level value and an abnormal reflectivity detection signal, or a medium oxygen level value and a normal reflectivity detection signal, the output is a medium to high grade and the image rendering material parameter correction command is executed.

[0042] When inputting a medium-grade oxygen level value and an abnormal reflectivity detection signal, or a thick-grade oxygen level value and a normal reflectivity detection signal, output a medium-to-low grade and execute a material replacement command for the oxide layer distribution area;

[0043] When inputting the thick oxide level value and the abnormal reflectivity detection signal, the output is a low grade and simultaneously executes the image rendering material parameter correction command and the oxide layer distribution area material replacement command.

[0044] According to a second aspect of the present invention, a gold industry e-commerce SaaS cloud service implementation system is also provided, for implementing any of the above-described gold industry e-commerce SaaS cloud service implementation methods, wherein the gold industry e-commerce SaaS cloud service implementation system comprises:

[0045] The feature acquisition module is used to acquire and extract the dimensional features of hyperspectral feature images, polarization feature images and defect mask images of gold products in real time, and then stitch them together to form a dimensionality-reduced fusion feature vector.

[0046] The sample parameter calculation module can construct a feature manifold that reflects the continuity of the oxidation evolution of the gold coating based on the distribution of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, and select and calculate the local density entropy and feature vector field divergence of the gold coating sample points along the oxidation principal axis.

[0047] The order value classification module is used to define the transition point of the thickness change of the gold coating as the abrupt change point of the local density entropy and the eigenvector field divergence, and automatically identify several abrupt change points through the Bayesian change point detection algorithm to dynamically classify the order value level and core situation point.

[0048] The order value determination module is used to calculate the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value in the feature manifold space, and determine the order value of the gold product based on the situation interval in which the geodesic distance is located.

[0049] The reflectivity detection signal output module can train and generate a reflectivity anomaly detection model through a pre-acquired reference coating reflectivity feature set, and input a dimension-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal.

[0050] The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] The proposed method for implementing e-commerce SaaS cloud services in the gold industry involves real-time acquisition of hyperspectral, polarization, and defect feature images of gold products, which are then fused into a dimensionality-reduced feature vector. This constructs a feature manifold reflecting the continuity of oxide coating oxidation. By utilizing local density entropy and vector field divergence mutation points, the method accurately locates the inflection points of oxide layer thickness changes. Combined with Bayesian variable point detection, the method dynamically classifies the order levels. Simultaneously, it determines the core situation interval based on geodesic distance matching. Finally, a detection signal is generated through a reflectivity anomaly detection model. This method achieves quantitative analysis of the spatiotemporal differences in oxide layer thickness distribution and coating reflectivity on the gold surface. It effectively solves the problem of image color distortion caused by the inability of traditional methods to capture dynamic changes in microstructure, thereby contributing to the development of e-commerce in the gold industry.

[0053] The gold industry e-commerce SaaS cloud service implementation system proposed in this invention belongs to the same general inventive concept as the aforementioned gold industry e-commerce SaaS cloud service implementation method, and should at least have the same technical effects as the aforementioned gold industry e-commerce SaaS cloud service implementation method. Therefore, this invention will not elaborate further here.

[0054] As can be seen from the above, the present invention can effectively solve the problem that existing e-commerce SaaS cloud service implementation methods in the gold industry are unable to analyze the spatiotemporal differences in the thickness distribution of the oxide layer on the gold surface and the reflectivity of the coating, which makes it impossible for consumers to truly observe the color and texture of the gold products through images, thus seriously restricting the development of e-commerce in the gold industry.

[0055] Other features and advantages of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0057] Figure 1 This is a flowchart illustrating a method for implementing an e-commerce SaaS cloud service in the gold industry, according to an embodiment of the present invention.

[0058] Figure 2 This is a schematic diagram of a gold industry e-commerce SaaS cloud service implementation system according to an embodiment of the present invention. Detailed Implementation

[0059] To enable those skilled in the art to more fully understand the technical solutions of the present invention, exemplary embodiments of the present invention will be described more comprehensively and in detail below with reference to the accompanying drawings. Obviously, the one or more embodiments of the present invention described below are merely one or more specific ways to implement the technical solutions of the present invention, and are not exhaustive. It should be understood that other ways belonging to a general inventive concept can be used to implement the technical solutions of the present invention, and should not be limited to the embodiments described exemplary. Based on one or more embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0060] Reference Figure 1 The present invention provides a method for implementing e-commerce SaaS cloud services in the gold industry, comprising:

[0061] Step S1: Real-time acquisition and extraction of dimensional features from hyperspectral feature images, polarization feature images, and defect mask images of gold products, which are then stitched together to form a dimensionality-reduced fusion feature vector;

[0062] Step S2: Based on the distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, construct a feature manifold that reflects the continuity of the oxidation evolution of the gold coating, select and calculate the local density entropy and feature vector field divergence of the gold coating sample points along the oxidation principal axis;

[0063] Step S3: Define the transition point of the gold plating thickness change as the abrupt change point of the local density entropy and eigenvector field divergence, and automatically identify several abrupt change points through the Bayesian change point detection algorithm to dynamically classify the order level and core situation point.

[0064] Step S4: In the feature manifold space, calculate the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value level, and determine the order value level of the gold product based on the situation interval where the geodesic distance is located.

[0065] Step S5: Train and generate a reflectivity anomaly detection model using the pre-acquired reference coating reflectivity feature set, and input a dimension-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal;

[0066] The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

[0067] In one embodiment, in step S1, the dimensional features of the hyperspectral feature image, polarization feature image, and defect mask image of the gold product are acquired and extracted in real time, and then stitched together to form a dimensionality-reduced fusion feature vector, specifically including:

[0068] Step S11: Perform multi-band spectral scanning on the gold product to obtain and extract hyperspectral feature images based on high-dimensional continuous spectral data;

[0069] Step S12: Collect gold artifacts at 0 -135 Intensity images of M polarization angles within an angular range are obtained. Polarization angle images are calculated and generated based on the degree of polarization of the intensity images. Multi-scale texture analysis is performed on the polarization angle images to extract their polarization texture features and form polarization feature images.

[0070] Step S13: Acquire gold images based on industrial cameras, and perform semantic segmentation of the gold surface defect areas in the gold images using a deep learning model to generate a binary defect mask image. Fill the internal holes of the defect areas using opening and closing operations to obtain the defect mask image.

[0071] Step S14: Scale the hyperspectral feature image, polarization feature image, and defect mask image to the same spatial resolution, and stitch them together to form a multi-channel tensor. Then, process the tensor based on the convolutional block attention module and output a dimension-reduced fused feature vector.

[0072] Specifically, embodiments of the present invention employ hyperspectral scanning to capture the continuous spectral response of gold products from visible light to near infrared, accurately reflecting the absorption peak shift caused by changes in oxide layer thickness. Polarization imaging generates polarization angle images through multi-angle intensity analysis, and combined with polarization features extracted by multi-scale texture analysis, effectively characterizing the micro-roughness differences on the coating surface caused by oxidation. Defect mask images locate surface defects through deep learning semantic segmentation, avoiding their interference with oxide layer analysis.

[0073] Therefore, the output of the dimension reduction and fusion feature vector not only preserves the spatiotemporal correlation of the physical properties of the gold surface, but also suppresses irrelevant noise through the attention mechanism, providing a data foundation with a high signal-to-noise ratio for subsequent processing and significantly improving the accuracy of image color restoration.

[0074] In one specific embodiment, in step S11, the gold product undergoes multi-band spectral scanning to acquire and extract hyperspectral feature images based on high-dimensional continuous spectral data, specifically including:

[0075] Step S111: Perform multi-band spectral scanning on the gold product using a hyperspectral imaging device to obtain high-dimensional continuous spectral data covering the visible to near-infrared bands;

[0076] Step S112: Principal Component Analysis (PCA) is used to reduce the dimensionality of the high-dimensional continuous spectral data, and the first N principal components are extracted from the high-dimensional continuous spectral data to obtain a hyperspectral feature image.

[0077] Specifically, Principal Component Analysis (PCA) projects high-dimensional continuous spectral data onto the N principal component directions with the largest variance through orthogonal transformation. These principal components retain the key spectral features in the high-dimensional continuous spectral data that reflect the composition and surface state of the material, thereby enabling the reconstruction of a representative hyperspectral feature image using low-dimensional feature vectors.

[0078] It is worth noting that the value of N is determined by calculating the variance of each principal component and then calculating the cumulative variance contribution value for each component in descending order. At 95%, the number of principal components selected is N.

[0079] In one specific embodiment, after step S11, where a multi-band spectral scan of the gold product is performed using a hyperspectral imaging device to obtain high-dimensional continuous spectral data covering the visible to near-infrared bands, the method further includes:

[0080] Sensor noise cancellation is performed on high-dimensional continuous spectral data using a non-uniformity correction method.

[0081] Specifically, non-uniformity correction is a technical method that compensates for differences in sensor pixel response through algorithms or hardware means to eliminate image non-uniformity noise and improve data consistency. Its specific denoising principle is existing technology and will not be elaborated upon further here.

[0082] In one specific embodiment, in step S12, the gold product is collected at 0 -135 Intensity images at M polarization angles within an angular range are used. Polarization angle images are calculated and generated based on the degree of polarization of the intensity images. Multi-scale texture analysis is performed on the polarization angle images to extract their polarization texture features, forming a polarization feature image. Specifically, this includes:

[0083] Step S121: Synchronously acquire data on the gold product at 0°C using a split-focus plane polarization camera. 45 90 and 135 Intensity images at four polarization angles;

[0084] Step S122: Calculate 45 sequentially With 135 0 With 90 The intensity difference values ​​recorded by the image sensor within the angular channel are mapped to an arctangent function value table using a lookup table to obtain the initial angle value of the polarization direction, and then the initial angle value is normalized to 0. -180 Within the angular range, to generate a polarization angle image;

[0085] Step S123: Extract texture features from the polarization angle image using a multi-scale Gabor filter bank, and calculate the sum of squares of the filter response amplitudes at each scale and the standard deviation of the main direction of the filter response based on the texture feature extraction results;

[0086] Step S124: Decompose the polarization angle image based on the pyramid decomposition algorithm, and calculate the sum of squares of the filter response amplitude and the standard deviation of the main direction of the filter response at each level to form a multi-scale texture feature map. Then, perform feature fusion on the multi-scale texture feature map based on Principal Component Analysis (PCA) to obtain the polarization feature image.

[0087] Specifically, 0 45 90 and 135 The acquisition of intensity images at four polarization angles can completely cover the spatial distribution of the linear polarization direction of polarized light. By calculation, the degree of polarization and polarization angle of the reflected light on the gold surface can be uniquely determined, so as to accurately analyze the differences in the scattering properties of light by the oxide layer on the gold surface, significantly improve the accuracy of the analysis of the oxide layer thickness distribution, and thus provide consumers with images of the color and texture of gold products that are closer to the real physical state.

[0088] In addition, the arctangent function value table is obtained by directly calculating the arctangent value of each acquisition angle using mathematical library functions and storing the calculation results as offline lookup data, so as to achieve rapid mapping of intensity difference.

[0089] It is worth noting that in step S122, normalization maps the initial angle values ​​calculated for different polarization directions to 0. -180 Within the angular range, the multi-valuedness caused by the periodicity of the arctangent function is eliminated, so that the angle value of each pixel uniquely corresponds to the vibration direction of the local light wave on the gold surface, thereby forming a polarization angle image that can intuitively reflect the microstructure arrangement characteristics.

[0090] In one embodiment, in step S2, based on the distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, a feature manifold reflecting the continuity of the oxidation evolution of the gold coating is constructed. The local density entropy and feature vector field divergence of the gold coating sample points are selected and calculated along the oxidation principal axis. Specifically, this includes:

[0091] Step S21: Use a nonlinear dimensionality reduction algorithm to map the dimensionality-reduced and fused feature vectors into a 2D low-dimensional space, and extract the spatial neighborhood relationship of the oxidation state of the gold coating. Construct a local connection structure based on the K-nearest neighbor graph, and form a feature manifold that reflects the continuity of the oxidation evolution of the gold coating through a local linear embedding algorithm.

[0092] Step S22: Based on the distribution density of sample points in the characteristic manifold, the density peak clustering algorithm is used to automatically identify high-density regions and define them as the core clusters of the oxidation state of the gold plating. The direction corresponding to the largest eigenvalue in the eigenvector of the covariance matrix of the core cluster is calculated and selected according to the principal component analysis (PCA) method and defined as the oxidation principal axis.

[0093] Step S23: Using each sample point as the center, count the number of sample points in the neighborhood of its distance radius R to obtain the local density. Calculate the probability density function of the local density using the kernel density estimation algorithm, and calculate the local density entropy based on the probability density function and the information entropy formula.

[0094] Step S24: Using each sample point in the feature manifold space as the base point, calculate the direction of oxidation state change of the K nearest neighbor sample points, construct the feature vector field, and calculate the divergence of the vector field at each sample point through the gradient estimation algorithm to obtain the feature vector field divergence.

[0095] Specifically, the local linear embedding algorithm enables adjacent sample points within the spatial neighborhood to maintain the same topological relationship in the low-dimensional space as in the original high-dimensional space. This allows the resulting feature manifold to accurately reflect the gradual continuity of the oxidation state of the gold plating, effectively avoiding topological breaks or pseudo-clustering caused by traditional dimensionality reduction methods. This provides a reliable geometric structure basis for subsequent oxidation axis identification and mutation point detection.

[0096] Furthermore, the value of radius R is determined based on the distribution density of gold coating samples in the characteristic manifold and the spatial scale of oxidation state, and is adaptively adjusted based on the manifold curvature. This avoids situations where a radius R that is too small makes the local density estimation sensitive to noise and unable to capture the gradual characteristics of oxide layer thickness, and where a radius R that is too large mixes samples of different oxidation states, thus masking the abrupt change information of local density entropy. In this way, the calculated local density entropy can effectively quantify the disorder of oxidation state on the gold surface.

[0097] The embodiments of this invention construct a low-dimensional representation space for the oxidation evolution of gold plating through nonlinear dimensionality reduction and manifold learning, effectively solving the technical challenge of intuitively analyzing the spatiotemporal differences in oxidation states in high-dimensional features. Furthermore, by combining the calculation of local density entropy and eigenvector field divergence, it highlights regions of oxide layer peeling or abrupt thickness changes and reveals the diffusion pattern of the oxidation front, providing key features that reflect the dynamic oxidation process for subsequent abrupt point detection, significantly improving the accuracy of gold color and texture restoration in images.

[0098] In one specific embodiment, in step S22, based on the distribution density of sample points in the characteristic manifold, a density peak clustering algorithm is used to automatically identify high-density regions and define them as core clusters of the gold plating oxidation state. After calculating and selecting the direction corresponding to the largest eigenvalue in the eigenvector of the covariance matrix of the core cluster according to Principal Component Analysis (PCA) and defining it as the oxidation principal axis, the method further includes:

[0099] The oxygen principal axis direction of different samples is aligned using a dynamic time warping algorithm.

[0100] Specifically, aligning the oxygen principal axis directions of different samples can eliminate directional deviations caused by shooting angles or surface curvature, thereby improving the selection accuracy of the oxidation principal axis.

[0101] In one specific embodiment, after the feature vector field is constructed, the following steps are also included:

[0102] The eigenvector field is smoothed by natural neighborhood interpolation.

[0103] Specifically, the smoothed eigenvector field can avoid interference from local noise, thereby improving the calculation accuracy of the eigenvector field divergence.

[0104] In one embodiment, in step S3, the transition point of the gold plating thickness change is defined as the abrupt change point of the local density entropy and eigenvector field divergence, and several abrupt change points are automatically identified by a Bayesian change point detection algorithm to dynamically classify the order level and core situation points, specifically including:

[0105] Step S31: Using a sliding window with a width of 5%-10% of the total number of samples and a step size of 1, calculate the mean and standard deviation of the local density entropy and the eigenvector field divergence within the window for each sample point. Mark sample points whose local density entropy exceeds the sum of the current window mean and twice the standard deviation, and whose eigenvector field divergence exceeds the sum of the current window mean and 1.5 times the standard deviation as mutation candidate points. Iteratively calculate the position and number of mutation candidate points using the Bayesian change point detection algorithm to automatically identify several mutation points.

[0106] Step S32: Let the number of mutation points be N. Then, based on the N mutation points, the feature manifold is divided into N+1 oxidation stage regions. The local density entropy and feature vector field divergence of the sample points in each oxidation stage region are standardized by Z-score. The oxidation state is divided into thin oxygen level, medium oxygen level and thick oxygen level by K-means clustering algorithm.

[0107] Step S33: Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is used to calculate and sort the local density of sample points within each oxidation stage region. Sample points with the highest local density in the top 10% are selected as candidate points. The cosine of the divergence direction between each candidate point and its five nearest neighbor sample points is calculated. Each candidate point with a cosine of the angle with its nearest neighbor sample points that is greater than 0.8 and has more than 3 such points is selected as the core situation point.

[0108] Specifically, embodiments of the present invention utilize the dynamic fluctuation characteristics of local density entropy and divergence statistically using a sliding window, and screen for abrupt change candidate points through a dual threshold, thereby preserving the key signal of abrupt changes in oxide layer thickness while suppressing the interference of measurement noise.

[0109] Meanwhile, by combining the Bayesian change point detection algorithm with iterative optimization of the posterior distribution of the number and location of change points, it automatically adapts to the differences in oxidation characteristics of different batches of gold products, and the dynamically divided oxidation stage regions are more in line with the actual oxidation evolution process.

[0110] The extraction of core state points ensures that the selected points can stably represent the typical features of each oxidation stage, ultimately outputting the quantization level of the oxidation state, providing an interpretable physical grading standard for image color restoration. This design directly addresses the image distortion problem caused by the inability of traditional methods to distinguish the spatiotemporal differences in oxidation states. Through multi-level feature selection and probabilistic modeling, it significantly improves the accuracy of visual evaluation of gold products in e-commerce scenarios, enabling consumers to obtain high-fidelity visual information that truly reflects the physical properties of gold through a SaaS platform.

[0111] In one embodiment, in step S4, in the feature manifold space, the geodesic distance between the dimensionality-reduced fused feature vector and the core situation points corresponding to each order value level is calculated, and the order value level of the gold product is determined according to the situation interval where the geodesic distance is located, specifically including:

[0112] Step S41: Map all the output core situation points to the feature manifold space and obtain the coordinates of each core situation point;

[0113] Step S42: Calculate the distance between each sample point in all sample points and each core situation point with the shortest distance to it based on Dijkstra's Algorithm, and output it as geodesic distance;

[0114] Step S43: Based on the geodesic distance distribution of all core situation points, the characteristic manifold space is divided into thin oxygen interval, medium oxygen interval and thick oxygen interval using the quantile method;

[0115] Step S44: Based on the geodesic distance interval of each sample point, match the corresponding order value to determine the order value of the gold product.

[0116] Specifically, the embodiments of the present invention use the core state point as a reference benchmark and accurately reflect the true geometric relationship between the sample point and each oxidation stage through geodesic distance. Combined with the interval division achieved by the quantile method, it can adaptively match the oxidation differences of different batches of gold products based on the statistical distribution characteristics of the geodesic distance of the core point, so as to output the order value determined by the interval matching, which is directly related to the physical meaning of the thickness of the oxide layer on the gold surface. This provides an interpretable quantitative standard for image color restoration, effectively solving the problem of distance distortion caused by manifold curvature in the quantitative evaluation of the oxidation state of gold plating.

[0117] In one specific embodiment, before dividing the characteristic manifold space into thin-oxygen, medium-oxygen, and thick-oxygen intervals using the quantile method based on the geodesic distance distribution of all core situation points in step S43, the method further includes:

[0118] Using the initial geodesic distance of each sample point as the weight, sample points within its radius R neighborhood are selected for local weighted regression processing, and the geodesic distance is corrected through iterative calculation.

[0119] Specifically, the correction of geodetic distance can eliminate distance deviations caused by local curvature, thereby improving the accuracy of geodetic distance.

[0120] In one specific embodiment, step S45, which involves matching the corresponding order value based on the geodesic distance interval of each sample point to determine the order value of the gold product, also includes:

[0121] If the geodesic distance of any sample point lies between two interval boundaries, the nearest neighbor principle is used to determine the order value.

[0122] Specifically, by determining the order value of any sample point at the boundary of two intervals through voting, the problem of misjudgment of order value caused by the ambiguity of geodesic distance of boundary samples is effectively solved, and the fault tolerance and robustness of oxidation state classification are improved.

[0123] In one embodiment, in step S5, a reflectivity anomaly detection model is trained and generated using a pre-acquired reference coating reflectivity feature set. A dimension-reduced fusion feature vector is then input into the reflectivity anomaly detection model to output a reflectivity detection signal. Specifically, this includes:

[0124] Step S51: While splicing to form a dimension-reduced fusion feature vector, a spectrophotometer is used to measure the reflectance value of the gold coating, and the dimension-reduced fusion feature vector and reflectance value are combined to obtain the reflectance feature set of the reference coating.

[0125] Step S52: Using the baseline coating reflectivity feature set as input, train and generate a reflectivity anomaly detection model through the isolated forest algorithm;

[0126] Step S53: Input the dimension-reduced and fused feature vector into the reflectivity anomaly detection model to output the reflectivity detection signal;

[0127] Step S54: Determine the image quality grade of the gold product based on the grade value and reflectivity detection signal.

[0128] In one specific embodiment, in step S54, the image quality grade of the gold product is determined based on the grade value and the reflectivity detection signal, specifically including:

[0129] When a thin oxygen level value and a normal reflectance detection signal are input, a high grade is output and an image processing termination command is executed.

[0130] When the input is a thin oxygen level value and an abnormal reflectivity detection signal, or a medium oxygen level value and a normal reflectivity detection signal, the output is a medium to high grade and the image rendering material parameter correction command is executed.

[0131] When inputting a medium-grade oxygen level value and an abnormal reflectivity detection signal, or a thick-grade oxygen level value and a normal reflectivity detection signal, output a medium-to-low grade and execute a material replacement command for the oxide layer distribution area;

[0132] When inputting the thick oxide level value and the abnormal reflectivity detection signal, the output is a low grade and simultaneously executes the image rendering material parameter correction command and the oxide layer distribution area material replacement command.

[0133] Specifically, the above embodiments of the present invention systematically solve the technical problem of the disconnect between physical characteristics and visual representation in the quality assessment of gold product images. Its reflectance anomaly detection model, constructed using the isolated forest algorithm and trained with spectrophotometer measurement data, accurately captures reflectance anomaly signals caused by sudden changes in oxide layer thickness. Based on the order value, it quantifies the continuous change process of the oxidation state, and the two combined form a two-dimensional assessment system. This preserves the repairable value of gold products while preventing the online dissemination of low-quality images. Furthermore, by incorporating dynamic termination or correction commands, it achieves closed-loop control from assessment to optimization, significantly improving the authenticity and transaction trust of gold product images in e-commerce scenarios.

[0134] It is worth noting that the algorithms used in the above embodiments of the present invention are all existing technologies. Therefore, the specific application principles will not be elaborated further here.

[0135] The proposed method for implementing e-commerce SaaS cloud services in the gold industry involves real-time acquisition of hyperspectral, polarization, and defect feature images of gold products, which are then fused into a dimensionality-reduced feature vector. This constructs a feature manifold reflecting the continuity of oxide coating oxidation. By utilizing local density entropy and vector field divergence mutation points, the method accurately locates the inflection points of oxide layer thickness changes. Combined with Bayesian variable point detection, the method dynamically classifies the order levels. Simultaneously, it determines the core situation interval based on geodesic distance matching. Finally, a detection signal is generated through a reflectivity anomaly detection model. This method achieves quantitative analysis of the spatiotemporal differences in oxide layer thickness distribution and coating reflectivity on the gold surface. It effectively solves the problem of image color distortion caused by the inability of traditional methods to capture dynamic changes in microstructure, thereby contributing to the development of e-commerce in the gold industry.

[0136] Accordingly, refer to Figure 2 The present invention also provides a gold industry e-commerce SaaS cloud service implementation system, used to implement the gold industry e-commerce SaaS cloud service implementation method in any of the above embodiments. The gold industry e-commerce SaaS cloud service implementation system includes:

[0137] The feature acquisition module is used to acquire and extract the dimensional features of hyperspectral feature images, polarization feature images and defect mask images of gold products in real time, and then stitch them together to form a dimensionality-reduced fusion feature vector.

[0138] The sample parameter calculation module can construct a feature manifold that reflects the continuity of the oxidation evolution of the gold coating based on the distribution of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, and select and calculate the local density entropy and feature vector field divergence of the gold coating sample points along the oxidation principal axis.

[0139] The order value classification module is used to define the transition point of the thickness change of the gold coating as the abrupt change point of the local density entropy and the eigenvector field divergence, and automatically identify several abrupt change points through the Bayesian change point detection algorithm to dynamically classify the order value level and core situation point.

[0140] The order value determination module is used to calculate the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value in the feature manifold space, and determine the order value of the gold product based on the situation interval in which the geodesic distance is located.

[0141] The reflectivity detection signal output module can train and generate a reflectivity anomaly detection model through a pre-acquired reference coating reflectivity feature set, and input a dimension-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal.

[0142] The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

[0143] The gold industry e-commerce SaaS cloud service implementation system proposed in this invention belongs to the same general inventive concept as the aforementioned gold industry e-commerce SaaS cloud service implementation method, and should at least have the same technical effects as the aforementioned gold industry e-commerce SaaS cloud service implementation method. Therefore, this invention will not elaborate further here.

[0144] While one or more embodiments of the present invention have been described above, those skilled in the art will recognize that the present invention can be implemented in any other form without departing from its spirit and scope. Therefore, the embodiments described above are illustrative and not restrictive, and many modifications and substitutions will be apparent to those skilled in the art without departing from the spirit and scope of the invention as defined in the appended claims.

Claims

1. A method for implementing e-commerce SaaS cloud services in the gold industry, characterized in that, include: The dimensional features of hyperspectral feature images, polarization feature images, and defect mask images of gold products are acquired and extracted in real time, and then stitched together to form a dimensionality-reduced fusion feature vector. Based on the distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, a feature manifold reflecting the continuity of the oxidation evolution of the gold coating is constructed. The local density entropy and feature vector field divergence of the gold coating sample points are selected and calculated along the oxidation principal axis. The transition point of the gold plating thickness variation is defined as the abrupt change point of the local density entropy and the eigenvector field divergence. Several abrupt change points are automatically identified by the Bayesian change point detection algorithm to dynamically classify the order level and core situation point. In the feature manifold space, the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value level is calculated, and the order value level of the gold product is determined according to the situation interval in which the geodesic distance is located. A reflectivity anomaly detection model is trained and generated by pre-acquired benchmark coating reflectivity feature set, and a dimension-reduced fusion feature vector is input into the reflectivity anomaly detection model to output a reflectivity detection signal. The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

2. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 1, characterized in that, The distribution pattern of the dimensionality-reduced fused feature vectors in the low-dimensional feature space is used to construct a feature manifold reflecting the continuity of the oxidation evolution of the gold coating. The local density entropy and feature vector field divergence of the gold coating sample points are selected and calculated along the principal oxidation axis. Specifically, this includes: A nonlinear dimensionality reduction algorithm is used to map the dimensionality-reduced and fused feature vectors into a 2D low-dimensional space, and the spatial neighborhood relationship of the oxidation state of the gold coating is extracted. A local connection structure is constructed based on the K-nearest neighbor graph, and a feature manifold reflecting the continuity of the oxidation evolution of the gold coating is formed through a local linear embedding algorithm. Based on the distribution density of sample points in the characteristic manifold, the density peak clustering algorithm is used to automatically identify high-density regions and define them as the core clusters of the oxidation state of the gold coating. The direction corresponding to the largest eigenvalue in the eigenvector of the covariance matrix of the core cluster is defined as the oxidation principal axis. For each sample point as the center, the number of sample points in the neighborhood of its distance radius R is counted to obtain the local density. The probability density function of the local density is calculated by the kernel density estimation algorithm, and the local density entropy is calculated based on the probability density function and the information entropy formula. Using each sample point in the characteristic manifold space as the base point, the direction of oxidation state change of the K nearest neighbor sample points is calculated to construct the eigenvector field. The divergence of the vector field at each sample point is calculated by the gradient estimation algorithm to obtain the eigenvector field divergence.

3. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 2, characterized in that, The method of automatically identifying high-density regions and defining them as core clusters of the gold plating oxidation state based on the distribution density of sample points in the characteristic manifold using the density peak clustering algorithm, and defining the direction corresponding to the largest eigenvalue in the covariance matrix eigenvector of the core cluster as the oxidation principal axis according to PCA calculation, further includes: The oxygen principal axis direction of different samples is aligned using a dynamic time warping algorithm.

4. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 3, characterized in that, The transition point of the gold plating thickness variation is defined as the abrupt change point of local density entropy and eigenvector field divergence. Several abrupt change points are automatically identified using a Bayesian change point detection algorithm to dynamically classify order levels and core situation points, specifically including: A sliding window with a width of 5%-10% of the total number of samples and a step size of 1 is used. For each sample point, the mean and standard deviation of the local density entropy and the eigenvector field divergence within its window are calculated. Sample points whose local density entropy exceeds the sum of the current window mean and twice the standard deviation, and whose eigenvector field divergence exceeds the sum of the current window mean and 1.5 times the standard deviation, are marked as mutation candidate points. The position and number of mutation candidate points are iteratively calculated using a Bayesian change point detection algorithm to automatically identify several mutation points. If the number of mutation points is N, the feature manifold is divided into N+1 oxidation stage regions based on the N mutation points. The local density entropy and feature vector field divergence of the sample points in each oxidation stage region are standardized by Z-score, so that the oxidation state can be divided into thin oxygen level, medium oxygen level and thick oxygen level by K-means clustering algorithm. The local density of sample points in each oxidation stage region is calculated and sorted based on the DBSCAN algorithm. The sample points with the highest local density in the top 10% are selected as candidate points. The cosine of the divergence direction between each candidate point and its five nearest neighbor sample points is calculated. Each candidate point with a cosine of the angle with its nearest neighbor sample points that is greater than 0.8 and has more than 3 such points is selected as the core situation point.

5. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 4, characterized in that, In the characteristic manifold space, the geodesic distance between the dimensionality-reduced fused feature vector and the core situation points corresponding to each order level is calculated, and the order level of the gold product is determined based on the situation interval where the geodesic distance lies. Specifically, this includes: Map all the output core situation points to the feature manifold space and obtain the coordinates of each core situation point; Based on the Dijkstra algorithm, the distance between each sample point in all sample points and each core situation point with the shortest distance is calculated and output as geodesic distance; Based on the geodesic distance distribution of all core situation points, the characteristic manifold space is divided into thin oxygen interval, medium oxygen interval and thick oxygen interval using the quantile method. Based on the geodesic distance interval of each sample point, the corresponding order value is matched to determine the order value of the gold product.

6. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 5, characterized in that, Before dividing the characteristic manifold space into thin-oxygen, medium-oxygen, and thick-oxygen intervals using the quantile method based on the geodesic distance distribution of all core situation points, the method further includes: Using the initial geodesic distance of each sample point as the weight, sample points within its radius R neighborhood are selected for local weighted regression processing, and the geodesic distance is corrected through iterative calculation.

7. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 6, characterized in that, The process of matching the corresponding order value level based on the geodesic distance interval of each sample point to determine the order value level of gold products also includes: If the geodesic distance of any sample point lies between two interval boundaries, the nearest neighbor principle is used to determine the order value.

8. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 7, characterized in that, The process of training and generating a reflectivity anomaly detection model using a pre-acquired reference coating reflectivity feature set, and inputting a dimensionality-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal specifically includes: While splicing together to form a dimension-reduced fusion feature vector, a spectrophotometer is used to measure the reflectance value of the gold coating. The dimension-reduced fusion feature vector and the reflectance value are combined to obtain the reflectance feature set of the reference coating. Using the baseline coating reflectance feature set as input, a reflectance anomaly detection model is trained and generated through the isolated forest algorithm. The dimension-reduced and fused feature vectors are input into the reflectivity anomaly detection model to output the reflectivity detection signal; The quality grade of the gold product image is determined based on the grade value and reflectivity detection signal.

9. The method for implementing e-commerce SaaS cloud services in the gold industry according to claim 8, characterized in that, The quality grade of gold products is determined based on the grade value and reflectivity detection signal, specifically including: When a thin oxygen level value and a normal reflectance detection signal are input, a high grade is output and an image processing termination command is executed. When the input is a thin oxygen level value and an abnormal reflectivity detection signal, or a medium oxygen level value and a normal reflectivity detection signal, the output is a medium to high grade and the image rendering material parameter correction command is executed. When inputting a medium-grade oxygen level value and an abnormal reflectivity detection signal, or a thick-grade oxygen level value and a normal reflectivity detection signal, output a medium-to-low grade and execute a material replacement command for the oxide layer distribution area; When inputting the thick oxide level value and the abnormal reflectivity detection signal, the output is a low grade and simultaneously executes the image rendering material parameter correction command and the oxide layer distribution area material replacement command.

10. A SaaS cloud service implementation system for the gold industry e-commerce sector, characterized in that, The gold industry e-commerce SaaS cloud service implementation system, used to implement the gold industry e-commerce SaaS cloud service implementation method as described in any one of claims 1-9, comprises: The feature acquisition module is used to acquire and extract the dimensional features of hyperspectral feature images, polarization feature images and defect mask images of gold products in real time, and then stitch them together to form a dimensionality-reduced fusion feature vector. The sample parameter calculation module can construct a feature manifold that reflects the continuity of the oxidation evolution of the gold coating based on the distribution of the dimensionality-reduced fused feature vectors in the low-dimensional feature space, and select and calculate the local density entropy and feature vector field divergence of the gold coating sample points along the oxidation principal axis. The order value classification module is used to define the transition point of the thickness change of the gold coating as the abrupt change point of the local density entropy and the eigenvector field divergence, and automatically identify several abrupt change points through the Bayesian change point detection algorithm to dynamically classify the order value level and core situation point. The order value determination module is used to calculate the geodesic distance between the dimension-reduced fused feature vector and the core situation point corresponding to each order value in the feature manifold space, and determine the order value of the gold product based on the situation interval in which the geodesic distance is located. The reflectivity detection signal output module can train and generate a reflectivity anomaly detection model through a pre-acquired reference coating reflectivity feature set, and input a dimension-reduced fusion feature vector into the reflectivity anomaly detection model to output a reflectivity detection signal. The quality grade of gold products can be determined by analyzing the grade value and reflectivity detection signal.

Images