A knowledge graph-based intelligent identification method for true and false of south red agate

Abstract

The present application relates to the technical field of true and false identification of South Red Agate, and particularly relates to a true and false intelligent identification method of South Red Agate based on a knowledge graph. The method comprises the following steps: collecting original data of South Red Agate samples, and preprocessing the original data to obtain preprocessed data; processing the preprocessed data through a multi-dimensional feature convolutional neural network to obtain comprehensive feature data of the South Red Agate samples; introducing a graph reasoning layer into the multi-dimensional feature convolutional neural network, and reasoning through a dynamic topology reasoning algorithm based on the comprehensive feature data of the South Red Agate samples to obtain an optimal path; and based on the optimal path, intelligently identifying the true and false of the South Red Agate samples. The method solves the technical problems of insufficient accuracy and low efficiency of traditional true and false identification methods of South Red Agate.

Description

Technical Field

[0001] This invention relates to the field of authenticity identification technology for Southern Red Agate, and in particular to an intelligent method for authenticity identification of Southern Red Agate based on knowledge graph. Background Technology

[0002] Red agate, a precious gemstone, is widely loved for its unique color and texture. However, the proliferation of imitations on the market has created a dilemma for consumers and merchants, making it difficult to distinguish between genuine and fake products. Traditional methods for identifying genuine red agate mainly rely on manual identification, chemical reagent testing, microscopic observation, and physical testing. While these methods can provide some assistance, they often have limitations. Manual identification depends on expert experience, may be influenced by subjective factors, and is difficult to deal with various variations of imitations. Chemical reagent testing and microscopic observation typically require complex equipment and techniques, and the testing process is cumbersome, time-consuming, and difficult to operate. Furthermore, with the continuous advancement of imitation technology, the effectiveness of traditional methods in distinguishing some high-quality imitations is gradually weakening.

[0003] Therefore, traditional methods for distinguishing genuine from fake Southern Red Agate suffer from technical problems such as insufficient accuracy and low efficiency. Summary of the Invention

[0004] This invention provides a knowledge graph-based intelligent method for distinguishing the authenticity of Southern Red Agate, which solves the technical problems of insufficient accuracy and low efficiency in traditional methods for distinguishing the authenticity of Southern Red Agate.

[0005] The present invention provides an intelligent method for distinguishing the authenticity of Southern Red Agate based on knowledge graphs, which specifically includes the following technical solutions:

[0006] A knowledge graph-based intelligent method for distinguishing genuine from fake Southern Red Agate includes the following steps:

[0007] S1. Collect the raw data of the Southern Red Agate sample and preprocess the raw data to obtain the preprocessed data; use a multidimensional feature convolutional neural network to process the preprocessed data to obtain the comprehensive feature data of the Southern Red Agate sample.

[0008] S2. A graph inference layer is introduced into the multidimensional feature convolutional neural network. Based on the comprehensive feature data of the Southern Red Agate samples, the optimal path is obtained through dynamic topological inference algorithm. Based on the optimal path, the authenticity of the Southern Red Agate samples is intelligently judged.

[0009] Preferably, S1 specifically includes:

[0010] The multidimensional feature convolutional neural network specifically includes: an input layer, a convolutional layer, a pooling layer, and a fully connected layer.

[0011] Preferably, S1 specifically includes:

[0012] In the input layer of the multidimensional feature convolutional neural network, preprocessed data is assigned to different channels, and the separated preprocessed data is input into the convolutional layer. In the convolutional layer, each channel's preprocessed data is convolved using independent convolutional kernels to extract feature data. A cross-channel shared convolutional kernel is introduced to perform joint convolution on all extracted feature data, capturing the cross-information between different feature data. This cross-information is then input into the pooling layer. In the pooling layer, feature dimensionality reduction technology is used to reduce the cross-information between feature data, resulting in dimensionality-reduced feature data, which is then input into the fully connected layer. In the fully connected layer, the dimensionality-reduced feature data is flattened and fused using feature fusion technology to obtain the comprehensive feature data of the Southern Red Agate sample.

[0013] Preferably, S2 specifically includes:

[0014] In the implementation of the dynamic topology reasoning algorithm, a knowledge graph is constructed by combining prior knowledge of Southern Red Agate, and the nodes of the knowledge graph are mapped into embedding vectors through vectorization.

[0015] Preferably, S2 specifically includes:

[0016] In the implementation of the dynamic topology reasoning algorithm, a node similarity matrix is ​​obtained by calculating the similarity between the comprehensive feature data of all nodes and the Southern Red Agate sample; the node similarity matrix is ​​updated in each round of reasoning based on the comprehensive feature data of the currently input Southern Red Agate sample.

[0017] Preferably, S2 specifically includes:

[0018] The dynamic topology reasoning algorithm dynamically adjusts the topology of the knowledge graph in each round of reasoning based on the node similarity matrix and introduces a learning rate to obtain an updated graph topology. The updating stops when the updated graph topology meets the preset convergence condition or reaches the preset maximum number of reasoning rounds.

[0019] Preferably, S2 specifically includes:

[0020] Based on the updated graph topology and the node similarity matrix, the cost of each inference path is calculated, and the inference path with the lowest cost is selected as the optimal path.

[0021] Preferably, S2 specifically includes:

[0022] Calculate the similarity between the nodes in the optimal path and the comprehensive feature data of the Southern Red Agate sample, and output the probability value of distinguishing between true and false; compare the probability value of distinguishing between true and false with a preset threshold to determine the authenticity of the Southern Red Agate sample; when the probability value of distinguishing between true and false exceeds the threshold, the Southern Red Agate sample is determined to be genuine; otherwise, the Southern Red Agate sample is determined to be fake, and the confidence score is calculated by the probability difference method.

[0023] The beneficial effects of the technical solution of the present invention are:

[0024] 1. The multidimensional feature convolutional neural network can simultaneously process raw data (such as texture image data, reflectance spectrum image data, hardness information data, density information data, color information data, gloss information data, etc.) from multiple different sources of Southern Red Agate samples. The multidimensional feature convolutional neural network introduces a multi-channel processing method to extract multidimensional features of Southern Red Agate samples more accurately, ensuring the independence between features and that no information is lost, thereby improving the accuracy and robustness of subsequent discrimination.

[0025] 2. By introducing a graph reasoning layer and utilizing the node relationships in the knowledge graph, the comprehensive feature data of the Southern Red Agate sample is combined with the prior knowledge of Southern Red Agate (such as mineral composition, origin, historical background, etc.), and reasoning is performed based on the knowledge graph, which improves the accuracy and reliability of the judgment results.

[0026] 3. The dynamic topology reasoning algorithm enables the topology structure of the graph to be dynamically adjusted according to the actual characteristics of the Southern Red Agate sample. This ensures that the reasoning scheme that best meets the actual needs can be selected under different circumstances, avoiding the problem that the static graph structure cannot cope with complex changes, and improving adaptability and judgment accuracy. Attached Figure Description

[0027] Figure 1 This is a flowchart of a knowledge graph-based intelligent method for distinguishing genuine from fake Southern Red Agate, as described in this invention. Detailed Implementation

[0028] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0030] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent method for distinguishing genuine and fake Southern Red Agate based on knowledge graphs provided by this invention.

[0031] See attached document Figure 1 The diagram illustrates a flowchart of a knowledge graph-based intelligent method for distinguishing genuine from fake Southern Red Agate, provided by an embodiment of the present invention. The method includes the following steps:

[0032] S1. Collect the raw data of the Southern Red Agate sample and preprocess the raw data to obtain the preprocessed data; use a multidimensional feature convolutional neural network to process the preprocessed data to obtain the comprehensive feature data of the Southern Red Agate sample.

[0033] Raw data of the Southern Red Agate samples are collected using measuring equipment such as imaging devices, spectrometers, infrared spectrometers, and X-ray fluorescence analyzers. This includes texture image data, reflectance spectral image data, hardness information, density information, color information, and surface gloss information. For example, high-resolution image acquisition equipment is used to obtain microscopic images of the Southern Red Agate sample surface; a spectrometer is used to measure the reflectance spectrum of the Southern Red Agate sample surface; a hardness tester is used to measure the hardness information of the Southern Red Agate sample; the density information of the Southern Red Agate sample is measured using the mass / volume method; image processing technology is used to quantify the color information of the Southern Red Agate sample; and a gloss meter is used to test the surface gloss information of the Southern Red Agate sample.

[0034] The acquired raw data is preprocessed by noise removal, standardization, normalization and other methods to obtain preprocessed data, which is used as numerical data suitable for neural network processing. The preprocessing process adopts technical means well known to those skilled in the art, and will not be described in detail here.

[0035] Furthermore, a multi-dimensional feature convolutional neural network is used to process the preprocessed data to obtain comprehensive feature data of the Southern Red Agate samples. This multi-dimensional feature convolutional neural network is designed with multiple input channels, allowing each type of data to be processed through independent convolutional kernels, thereby extracting effective information from the preprocessed data to obtain the comprehensive feature data of the Southern Red Agate samples. The specific implementation process is as follows:

[0036] In the input layer, preprocessed data is assigned to different channels, each receiving specific input data. For example, preprocessed texture image data is fed into a dedicated channel for image feature processing, preprocessed reflectance spectrum image data enters the spectral channel, and preprocessed hardness and density information data enter a dedicated channel for physical properties. The specific number and function of each channel are determined using expert experience based on the specific application scenario. This method effectively separates different types of preprocessed data, avoiding interference between different data types and ensuring the independence of their respective features. The separated preprocessed data is then fed into the convolutional layer of a multidimensional feature convolutional neural network for feature extraction.

[0037] In the convolutional layer, each channel's preprocessed data is convolved using independent convolutional kernels. Each kernel extracts feature data from the preprocessed data. The learning objective of the convolutional kernel is to identify different patterns, textures, or regularities from the input image or data. For example, for textured image data, the kernel detects different textures and edge information; for spectral data, it identifies specific spectral peaks or reflectance features. After obtaining feature data for each channel using independent convolutional kernels, a cross-channel shared convolutional kernel is introduced to capture the interdependencies between different features. That is, after feature extraction from the preprocessed data of each channel by all independent convolutional kernels, a cross-channel shared convolutional kernel is introduced to jointly convolve all extracted feature data, capturing the cross-information between different feature data from the feature data extracted from the preprocessed data of each channel. This cross-information is used as the output of the convolutional layer and simultaneously as the input of the pooling layer.

[0038] In the pooling layer, feature dimensionality reduction techniques (such as principal component analysis) are used to reduce the dimensionality of the cross information between feature data. That is, the dimensionality of the input features of the pooling layer is compressed and unimportant features are removed to obtain the dimensionality-reduced feature data. The dimensionality-reduced feature data is used as the output of the pooling layer and as the input of the fully connected layer.

[0039] The main function of the fully connected layer is to flatten the dimensionality-reduced feature data output by the pooling layer and then fuse it using existing feature fusion techniques to obtain the comprehensive feature data of the Southern Red Agate sample.

[0040] S2. A graph inference layer is introduced into the multidimensional feature convolutional neural network. Based on the comprehensive feature data of the Southern Red Agate samples, the optimal path is obtained through dynamic topological inference algorithm. Based on the optimal path, the authenticity of the Southern Red Agate samples is intelligently judged.

[0041] A graph reasoning layer is introduced into a multidimensional feature convolutional neural network. To improve the intelligence and accuracy of the graph reasoning layer, a dynamic topology reasoning algorithm is introduced to realize knowledge reasoning. The specific implementation process of the dynamic topology reasoning algorithm is as follows:

[0042] First, by combining scientific testing techniques, literature review, expert experience, and market research, we obtain existing, definitive prior knowledge about Southern Red Agate (such as mineral composition, processing techniques, and origin). Then, we construct a knowledge graph structure represented by nodes and relationships. Each node represents an entity (e.g., a specific mineral composition, a specific origin), while the edges between nodes represent the relationships between them (e.g., "contains mineral composition X" or "originates from region Y"). This process organizes the various prior knowledge points about Southern Red Agate into a structured knowledge graph.

[0043] In the graph reasoning process, the comprehensive feature data of the Southern Red Agate samples are matched with nodes in the knowledge graph. Each node is then vectorized and mapped into a low-dimensional embedding vector G = [g1, g2, ..., g m ], where g i This is the vector representation of the i-th node in the knowledge graph, where i∈[1,m], and m is the total number of nodes in the knowledge graph. Furthermore, the similarity between each node and the comprehensive feature data of the Southern Red Agate sample is calculated using cosine similarity, with the specific formula as follows:

[0044]

[0045] Among them, S i,j F is the similarity between the i-th node in the knowledge graph and the j-th comprehensive feature data of the Southern Red Agate sample, used to measure the matching degree between the node and the comprehensive feature data of the Southern Red Agate sample; j is the vector representation of the j-th comprehensive feature data of the Southern Red Agate sample; ||·||2 is the Euclidean norm;

[0046] By calculating the similarity between all nodes in the knowledge graph and the comprehensive feature data of the Southern Red Agate sample, a node similarity matrix S is obtained, which represents the relationship between the knowledge graph nodes and the Southern Red Agate sample. Based on the node similarity matrix, the comprehensive feature data of the Southern Red Agate sample is matched with the nodes in the knowledge graph, thereby providing support for subsequent reasoning. Nodes with high similarity are assigned higher weights, while nodes with low similarity have less impact on the reasoning process. The node similarity matrix provides a "relationship metric" between the Southern Red Agate sample and the knowledge graph nodes, and the node similarity matrix is ​​updated in each round of reasoning based on the comprehensive feature data of the currently input Southern Red Agate sample.

[0047] Furthermore, in each round of reasoning, the topology of the knowledge graph is dynamically adjusted based on the node similarity matrix to obtain an updated graph topology. The update formula is as follows:

[0048] T k+1 =T k +α·(ST k )

[0049] Among them, T k+1 T represents the (k+1)th layer topology of the knowledge graph, i.e., the updated graph topology; k The topology of the knowledge graph represents the k-th layer, i.e., the current graph topology; α is the learning rate, representing the magnitude of the knowledge graph update, determined by expert experience, with a value range of [0.01, 0.1], and can be 0.05; S is the node similarity matrix of the current Southern Red Agate sample. Through the above update formula, the topology of the knowledge graph can adaptively adjust according to the comprehensive feature data of the input Southern Red Agate sample, to better fit the comprehensive feature data of the current Southern Red Agate sample, so as to further select a suitable inference path;

[0050] In particular, the number of layers for updating the knowledge graph topology will dynamically change with the recursive depth of reasoning; when the topology of the knowledge graph meets the convergence conditions preset according to the expert experience method and can no longer be effectively updated, or when the preset maximum number of reasoning rounds is reached (determined according to the specific scenario), the update will stop.

[0051] Based on the updated graph topology and node similarity matrix, a path selection mechanism is designed to select the optimal path;

[0052] In the path selection mechanism, assuming the set of knowledge graph inference paths is P, the inference path with the lowest cost is selected as the optimal path by calculating the cost value of each inference path. The cost function for path selection is as follows:

[0053]

[0054] Among them, w i It is the weight of the i-th node in the knowledge graph. P best The optimal path is selected through a path selection mechanism. By choosing the inference path with the lowest cost, it is ensured that the inference path that best matches the comprehensive feature data of the Southern Red Agate sample is selected.

[0055] Based on the optimal path, the similarity between the nodes in the optimal path and the comprehensive feature data of the Southern Red Agate samples is calculated, and the probability value P for distinguishing between true and false is output. dtrn :

[0056]

[0057] in, It is the th in the optimal path The similarity between a node and the j-th comprehensive feature data of a Southern Red Agate sample; It is the th in the optimal path The weight of each node,

[0058] Finally, the probability value P of distinguishing between true and false is... dtrn The authenticity of the Southern Red Agate sample is determined by comparing it with a threshold θ determined based on expert experience; when the probability value P of distinguishing between true and false is... dtrn When the value exceeds the threshold θ, the Southern Red Agate sample is determined to be genuine; when the probability value of distinguishing between genuine and fake is P... dtrn When the value is less than or equal to the threshold θ, the Southern Red Agate sample is determined to be a counterfeit; and the confidence score is calculated by the probability difference method, which is a technical means well known to those skilled in the art and will not be described in detail here.

[0059] In summary, a knowledge graph-based intelligent method for distinguishing genuine from fake Southern Red Agate has been developed.

[0060] The order of the embodiments is for illustrative purposes only and does not represent the superiority or inferiority of the embodiments. The processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0061] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0062] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A knowledge graph-based intelligent identification method for true and false of South Red Agate, characterized in that, The method comprises the following steps: S1. Collecting original data of the south red agate sample and preprocessing the original data to obtain preprocessed data; The multi-dimensional feature convolutional neural network comprises an input layer, a convolutional layer, a pooling layer and a fully connected layer. S2. In the multi-dimensional feature convolutional neural network, a graph reasoning layer is introduced, and based on the comprehensive feature data of the south red agate sample, a dynamic topology reasoning algorithm is used for reasoning to obtain an optimal path; based on the optimal path, the authenticity of the south red agate sample is intelligently judged; in the implementation process of the dynamic topology reasoning algorithm, a knowledge graph represented by nodes and relationships is constructed by combining prior knowledge of the south red agate, and the nodes of the knowledge graph are mapped into embedding vectors through vectorization processing; the similarity between all nodes and the comprehensive feature data of the south red agate sample is calculated to obtain a node similarity matrix; in each round of reasoning, based on the node similarity matrix, the topology structure of the knowledge graph is dynamically adjusted to obtain an updated graph topology structure; based on the updated graph topology structure, the node similarity matrix is combined to design a path selection mechanism, calculate the cost value of each reasoning path, and select the reasoning path with the minimum cost value as the optimal path; the cost function of path selection is as follows: The S1 specifically comprises: wherein, is the weight of the first node of the knowledge graph; is the weight of the first node of the knowledge graph; is the similarity between the first node of the knowledge graph and the first comprehensive feature data of the South Red Agate sample; is the similarity between the first node of the knowledge graph and the first comprehensive feature data of the South Red Agate sample; is the similarity between the first node of the knowledge graph and the first comprehensive feature data of the South Red Agate sample; is the optimal path selected by the path selection mechanism.

2. The knowledge graph-based method for intelligent identification of true or false of South Red Agate according to claim 1, characterized in that, In the input layer of the multi-dimensional feature convolutional neural network, the preprocessed data is distributed to different channels, and the separated preprocessed data is input into the convolutional layer; in the convolutional layer, the preprocessed data of each channel is convolved by independent convolutional kernels to extract feature data, and a cross-channel shared convolutional kernel is introduced to jointly convolve all extracted feature data to capture the cross information between the feature data, and the cross information between the feature data is input into the pooling layer; in the pooling layer, the cross information between the feature data is processed by a feature dimension reduction technique to obtain reduced feature data, and the reduced feature data is input into the fully connected layer; in the fully connected layer, the reduced feature data is flattened and fused by a feature fusion technique to obtain the comprehensive feature data of the south red agate sample. The S2 specifically comprises:

3. The knowledge graph-based method for intelligent identification of true or false South Red Agate according to claim 1, characterized in that, In the implementation process of the dynamic topology reasoning algorithm, the node similarity matrix is updated according to the comprehensive feature data of the south red agate sample input in each round of reasoning. The S2 specifically comprises:

4. The knowledge graph-based method for intelligent identification of true or false of South Red Agate according to claim 3, characterized in that, The dynamic topology reasoning algorithm dynamically adjusts the topology structure of the knowledge graph based on the node similarity matrix in each round of reasoning to obtain an updated graph topology structure; when the updated graph topology structure meets the preset convergence condition or reaches the preset maximum number of reasoning rounds, the updating is stopped. The S2 specifically comprises:

5. The knowledge graph-based method for intelligent identification of true or false of South Red Agate according to claim 1, characterized in that, The similarity between the nodes in the optimal path and the comprehensive feature data of the south red agate sample is calculated, and a probability value of true or false judgment is output, and the formula is as follows: ​ in, It is the probability value for distinguishing between true and false; It is the th in the optimal path The node and the first of the Southern Red Agate samples Similarity between comprehensive feature data; It is the th in the optimal path The weights of each node are determined; the probability values ​​for distinguishing between genuine and fake are compared with a preset threshold to determine the authenticity of the Southern Red Agate sample; when the probability value for distinguishing between genuine and fake exceeds the threshold, the Southern Red Agate sample is determined to be genuine; otherwise, the Southern Red Agate sample is determined to be fake, and the confidence score is calculated using the probability difference method.

Images