An intelligent identification system for infringing counterfeit commodities based on dynamic spectrum map topology
By generating dynamic spectral sequences and constructing pixel association maps, and calculating graph topological features, the problem of low accuracy in identifying counterfeit and infringing goods in existing technologies is solved, achieving more efficient identification of counterfeit and infringing goods.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INST OF TECH
- Filing Date
- 2026-01-30
- Publication Date
- 2026-06-09
AI Technical Summary
Existing computer vision-based automatic identification systems for counterfeit and infringing goods have low accuracy when dealing with high-quality counterfeits, making it difficult to effectively identify such products.
By generating dynamic spectral sequences, extracting temporal features from dynamic spectral data, constructing pixel association graphs and calculating graph topological features, and using the graph topological features of genuine products for comparison to identify counterfeit and infringing products.
It significantly improves the accuracy of identifying counterfeit and infringing goods, avoids visual deception by high-quality imitations, and achieves a more efficient identification effect.
Smart Images
Figure CN122176356A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology, belonging to the field of image visual recognition technology. Background Technology
[0002] Counterfeit goods refer to products that, without the permission of the intellectual property rights holder, forge and imitate the trademarks of others, or imitate genuine products in terms of materials, craftsmanship, function, and labeling, misleading consumers into believing they are legitimate products. Counterfeit goods lack compliant testing and safety guarantees, which can easily affect consumers' use of the products and severely damage brand reputation. Therefore, identifying and avoiding counterfeit goods is crucial.
[0003] Existing technologies, such as Chinese patent application (202410951992.X), disclose an automatic identification system and method for counterfeit and infringing goods based on computer vision. This system includes a product image collection module for acquiring images of target products; a product image feature encoding module for encoding features in the images of target products to obtain a target product feature map; a product identification result generation module for determining whether a target product is counterfeit or infringing based on the target product feature map; the product identification result generation module includes a target product feature map dimensionality reduction unit for expanding the target product feature map to obtain a target product feature vector; a target product feature vector optimization unit for reconstructing the target product feature vector in a high-dimensional feature space based on self-correlation dimension to obtain an optimized target product feature vector; and a feature vector classification unit for passing the optimized target product feature vector through a product classifier to obtain a classification result, which indicates whether the target product is counterfeit or infringing.
[0004] Existing computer vision-based automatic identification systems and methods for counterfeit and infringing goods rely on computer vision to acquire product images for automatic identification. However, counterfeit and infringing methods are constantly being updated, and images can be so realistic that they are almost indistinguishable from genuine products, greatly reducing the accuracy of automatic identification of counterfeit and infringing goods based on product images. Summary of the Invention
[0005] The purpose of this invention is to address the problems and shortcomings of existing technologies and to creatively provide an intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology. This system identifies counterfeit and infringing goods by using the temporal patterns of physical characteristics formed in the production chain, rather than by directly comparing appearances. This avoids visual deception by high-quality counterfeits and thus significantly improves the accuracy of identifying counterfeit and infringing goods.
[0006] To achieve the above objectives, the present invention provides the following technical solution.
[0007] A smart identification system for counterfeit and infringing goods based on dynamic spectral topology, comprising:
[0008] Data acquisition module: Utilizes an adjustable-intensity LED array to switch bands according to a preset period and sequence. It continuously captures product images using a multi-band spectral camera, generating a dynamic spectral sequence. Each dynamic spectral data point in the sequence includes pixel coordinates, spectral frequency band, time point, and reflected light intensity.
[0009] Data preprocessing module: responsible for preprocessing dynamic spectral sequences.
[0010] Temporal feature extraction module: responsible for extracting the temporal features of each dynamic spectral data in the preprocessed dynamic spectral sequence, including decay half-life, spectral offset and fluctuation entropy.
[0011] Pixel Association Graph Construction Module: This module is responsible for using the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence as nodes, using temporal features as node features, calculating the edge weights between nodes, and constructing a pixel association graph based on the node edge weights.
[0012] Graph Topology Feature Calculation Module: Responsible for calculating the topology features of the constructed pixel association graph, including the features of tight node connections and the features of node information propagation.
[0013] Counterfeit Goods Identification Module: This module is responsible for comparing the topological features of the pixel association map of the original product with the preset original product. If the value is below the preset deviation threshold, the product is identified as an original product; otherwise, it is identified as a counterfeit product.
[0014] The data acquisition module's output is connected to the data preprocessing module's input. The data preprocessing module's output is connected to the temporal feature extraction module's input. The temporal feature extraction module's output is connected to the pixel association graph construction module's input. The pixel association graph construction module's output is connected to the graph topology feature calculation module's input. The graph topology feature calculation module's output is connected to the counterfeit and infringing goods identification module's input.
[0015] Furthermore, the data preprocessing module includes the following processing steps:
[0016] Calculate the mean and standard deviation of the reflected light intensity of all dynamic spectral data in the dynamic spectral sequence. If the reflected light intensity of a certain dynamic spectral data point deviates from the mean of the reflected light intensity of all dynamic spectral data by more than three times the standard deviation, it is marked as abnormal dynamic spectral data and removed.
[0017] Furthermore, the steps for extracting the decay half-life in the time-series feature extraction module include:
[0018] First, obtain the initial values of reflected light intensity of each coordinate pixel in the dynamic spectral data of the dynamic spectral sequence under different spectral frequency bands, as well as the time point when the reflected light intensity decays to half of the initial value. Then, calculate the difference between two time points under the same spectral frequency band, which is the decay half-life; calculate the average of the decay half-life under different spectral frequency bands, as the final decay half-life.
[0019] Furthermore, the steps for extracting the spectral shift in the temporal feature extraction module include:
[0020] First, obtain the peak reflection wavelength of the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence under different spectral bands. Then, calculate the difference between the peak reflection wavelengths under the same spectral band, which is the spectral offset; calculate the mean of the spectral offsets under different spectral bands, which is the final spectral offset.
[0021] Furthermore, the steps for extracting fluctuation entropy in the temporal feature extraction module include:
[0022] First, obtain the coordinate pixels of each dynamic spectral data point in the dynamic spectral sequence, along with the reflected light intensity and time point in different spectral bands. Then, calculate the proportion of reflected light intensity at each time point in the same spectral band to the total reflected light intensity, and substitute this proportion into the Shannon entropy formula to obtain the fluctuation entropy. Calculate the mean of the fluctuation entropy in different spectral bands as the final fluctuation entropy.
[0023] Furthermore, the calculation steps for the edge weights between nodes in the pixel association graph construction module include:
[0024] First, the reflected light intensity and time points of two adjacent nodes in different spectral frequency bands are obtained, and the curve of reflected light intensity changing with time is plotted to characterize the attenuation law of reflected light. Then, the two curves of reflected light intensity changing with time are input into the DTW algorithm to obtain the DTW distance between them, and the DTW distance between them is mapped inversely to the edge weight.
[0025] Furthermore, the calculation steps for the tightly connected node feature in the graph topology feature calculation module include:
[0026] First, count the number of neighbors for each node in the pixel association graph, and then count the actual number of edges between neighbors. Next, calculate the local clustering coefficient of each node; calculate the mean of the local clustering coefficients of all nodes to obtain the average clustering coefficient, which characterizes the tightness of the connections between neighboring nodes.
[0027] Furthermore, the calculation steps for the node information propagation features in the graph topology feature calculation module include:
[0028] First, the shortest path distance between each pair of different nodes in the pixel association graph is calculated, and the information transmission speed of each pair of nodes is calculated based on the shortest path distance. Then, the average information transmission speed of all node pairs is calculated to obtain the global node information transmission speed, which characterizes the node information propagation feature. The steps for obtaining the topological features of the pixel association graph of the pre-set genuine products in the counterfeit goods identification module are the same as this process.
[0029] Beneficial effects
[0030] Compared with the prior art, the present invention has the following advantages:
[0031] This invention acquires images of a product in different wavelength bands, generates a dynamic spectral sequence, and extracts the temporal features of each dynamic spectral data point in the sequence. A pixel association graph is constructed using these temporal features, and graph topological features are calculated from this graph. By comparing the graph topological structure with that of genuine products, counterfeit or infringing products can be identified.
[0032] Compared with existing technologies, this invention identifies counterfeit and infringing goods by using the temporal patterns of physical characteristics formed in the production chain, rather than by directly comparing appearances. This avoids visual deception by high-quality imitations and significantly improves the accuracy of identifying counterfeit and infringing goods. Attached Figure Description
[0033] Figure 1 This is a system framework diagram of the present invention. Detailed Implementation
[0034] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and 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.
[0035] like Figure 1 As shown, an intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology includes:
[0036] Data Acquisition Module 1: Utilizes an adjustable-intensity LED array to switch bands according to a preset period and sequence. A multi-band spectral camera continuously captures product images, generating a dynamic spectral sequence. Each dynamic spectral data point in the sequence includes coordinate pixels, spectral frequency band, time point, and reflected light intensity.
[0037] Data preprocessing module 2: Preprocesses dynamic spectral sequences.
[0038] First, calculate the mean and standard deviation of the reflected light intensity of all dynamic spectral data in the dynamic spectral sequence.
[0039] Among them, the mean The expression is:
[0040]
[0041] Standard deviation The expression is:
[0042]
[0043] in, This represents the number of reflected light intensities from all dynamic spectral data in the dynamic spectral sequence. Indicates the first spectral sequence in the dynamic spectral sequence The reflected light intensity of each dynamic spectral data point.
[0044] If the reflected light intensity of a certain dynamic spectral data in a dynamic spectral sequence deviates from the mean reflected light intensity of all dynamic spectral data by more than three times the standard deviation, it is marked as abnormal dynamic spectral data and removed.
[0045] Temporal feature extraction module 3: responsible for extracting the temporal features of each dynamic spectral data in the preprocessed dynamic spectral sequence, including decay half-life, spectral offset and fluctuation entropy.
[0046] The decay half-life characterizes the time required for the reflected light intensity of a coordinate pixel to decrease from its initial value to half. Genuine products use higher quality inks, resulting in a slower decay of reflected light intensity, while counterfeit products use lower quality inks, also resulting in a slower decay of reflected light intensity. This can help distinguish between genuine and counterfeit products.
[0047] Spectral offset characterizes the change over time of the point where a coordinate pixel is best at reflecting light. Genuine products use higher quality inks with stable materials and less variation, while counterfeit products use lower quality inks with unstable materials and greater variation. This can distinguish genuine products from counterfeit products.
[0048] Fluctuation entropy characterizes the degree of disorder in the change of reflected light intensity of coordinate pixels over time. Genuine products use higher quality inks, and the reflected light intensity generally fluctuates regularly or weakens steadily. In contrast, counterfeit products use lower quality inks, and the reflected light intensity generally fluctuates wildly without any regularity. This can distinguish between genuine and counterfeit products.
[0049] Specifically, the steps for extracting the decay half-life in the time-series feature extraction module 3 include:
[0050] First, obtain the initial value of the reflected light intensity of the coordinate pixel of each dynamic spectral data in the dynamic spectral sequence under different spectral frequency bands, and the time point when the reflected light intensity decays to half of the initial value.
[0051] Then calculate the difference between two time points in the same spectral frequency band. This is the decay half-life, expressed as:
[0052]
[0053] in, Represents the initial time coordinates in pixels. In the spectral band The intensity of reflected light below ; Indicates time Time coordinate pixels In the spectral band The intensity of reflected light below .
[0054] Then, the average decay half-life under different spectral frequency bands is calculated as the final decay half-life.
[0055] Specifically, the specific steps for extracting the spectral shift in the time-series feature extraction module 3 include:
[0056] First, obtain the peak reflection wavelength of the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence under different spectral frequency bands.
[0057] Then, calculate the difference between the peak reflection wavelengths in the same spectral band. This is the spectral shift, expressed as:
[0058]
[0059] in, Represents the initial time coordinates in pixels. The peak reflection wavelength; Indicates time Time coordinate pixels The peak reflection wavelength.
[0060] Then, the mean of the spectral shifts in different spectral frequency bands is calculated as the final spectral shift.
[0061] Specifically, the specific steps for extracting the fluctuation entropy in the time-series feature extraction module 3 include:
[0062] First, obtain the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence, along with the reflected light intensity and time point in different spectral frequency bands.
[0063] Then, calculate the proportion of reflected light intensity to total reflected light intensity at each time point within the same spectral frequency band. Substituting this into the Shannon entropy formula, the result is the fluctuation entropy. The proportional conversion expression is as follows:
[0064]
[0065] in, Indicates time Time coordinate pixels In the spectral band The intensity of reflected light below ; Represents all moments coordinate pixels In the spectral band Total reflected light intensity .
[0066] Shannon entropy The formula is:
[0067]
[0068] Measure the contribution of uncertainty at a single moment.
[0069] Then, the mean value of the wave entropy under different spectral frequency bands is calculated as the final wave entropy.
[0070] Pixel Association Graph Construction Module 4: It is responsible for taking the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence as nodes, the temporal features as node features, calculating the edge weights between nodes, and constructing a pixel association graph based on the node edge weights.
[0071] Specifically, the steps for calculating the edge weights between nodes in pixel association graph construction module 4 include:
[0072] First, obtain the reflected light intensity and time points of two adjacent nodes in different spectral frequency bands, and plot the curve of reflected light intensity changing with time to characterize the attenuation law of reflected light.
[0073] Then, the curves showing the change of the intensity of the two reflected lights over time are input into the DTW algorithm to obtain the DTW distance between them, and the DTW distance between them is mapped inversely to the edge weight.
[0074] Graph Topology Feature Calculation Module 5: Responsible for calculating the topology features of the constructed pixel association graph, including the features of tight node connections and the features of node information propagation.
[0075] The tightness of node connections indicates whether the material is continuous. Genuine products use higher quality inks, resulting in continuous material and tighter pixel relationships. In contrast, counterfeit products use lower quality inks, leading to looser material relationships due to splicing, and thus looser pixel relationships. This feature can distinguish between genuine and counterfeit products.
[0076] The characteristics of node information transmission indicate whether the material is continuous. Genuine products use higher quality inks, resulting in continuous material, tighter pixel relationships, and faster information transmission. In contrast, counterfeit products use lower quality inks, resulting in looser material relationships due to splicing, looser pixel relationships, and slower information transmission. This feature can distinguish between genuine and counterfeit products.
[0077] Specifically, the calculation steps for the tightly connected node feature in the graph topology feature calculation module 5 include:
[0078] First, count the number of neighbors for each node in the pixel association graph, and count the actual number of edges between the neighbors;
[0079] Then, the local clustering coefficients of the nodes are calculated. The expression is:
[0080]
[0081] in, This indicates the actual number of edges between neighbors; Indicates the number of neighbors.
[0082] Next, the mean of the local clustering coefficients of all nodes is calculated to obtain the average clustering coefficient. The expression representing the tight connectivity of neighboring nodes is:
[0083]
[0084] in, Indicates the number of nodes; This indicates the actual number of edges between neighbors; Indicates the number of neighbors.
[0085] Specifically, the calculation steps for the node information propagation features in the graph topology feature calculation module 5 include:
[0086] First, calculate the shortest path distance between each pair of different nodes in the pixel association graph.
[0087] Then, the information transmission speed of each pair of nodes is calculated based on the shortest path distance between each pair of nodes. The expression is:
[0088]
[0089] in, Represents a node To the node The shortest path distance;
[0090] Next, the average information transmission speed of all node pairs is calculated to obtain the global node information transmission speed. The expression representing the characteristics of node information propagation is:
[0091]
[0092] in, Indicates the number of nodes; Represents a node To the node The shortest path distance.
[0093] Counterfeit Goods Identification Module 6: Compares the topological features of the pixel association map of the preset genuine goods with the preset deviation threshold. If the deviation threshold is lower than the deviation threshold, it is identified as a genuine goods; if it is higher than or equal to the deviation threshold, it is identified as a counterfeit goods.
[0094] The steps for obtaining the topological features of the pixel association map of the genuine product in the counterfeit and infringing product identification module 6 are the same as those described above.
[0095] The output of data acquisition module 1 is connected to the input of data preprocessing module 2. The output of data preprocessing module 2 is connected to the input of temporal feature extraction module 3. The output of temporal feature extraction module 3 is connected to the input of pixel association graph construction module 4. The output of pixel association graph construction module 4 is connected to the input of graph topology feature calculation module 5. The output of graph topology feature calculation module 5 is connected to the input of counterfeit and infringing goods identification module 6.
Claims
1. A smart identification system for counterfeit and infringing goods based on dynamic spectral topology, characterized in that, include: Data acquisition module: Utilizes an adjustable LED array to switch bands according to a preset cycle and sequence; continuously captures product images using a multi-band spectral camera to generate a dynamic spectral sequence; each dynamic spectral data in the dynamic spectral sequence includes coordinate pixels, spectral frequency band, time point, and reflected light intensity; Data preprocessing module: responsible for preprocessing dynamic spectral sequences; The temporal feature extraction module is responsible for extracting the temporal features of each dynamic spectral data in the preprocessed dynamic spectral sequence, including decay half-life, spectral offset, and fluctuation entropy. Pixel association graph construction module: It is responsible for taking the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence as nodes, taking the temporal features as node features, calculating the edge weights between nodes, and constructing a pixel association graph based on the node edge weights. Graph topology feature calculation module: responsible for calculating the topology features of the constructed pixel association graph, including the features of tight node connections and the features of node information propagation; Counterfeit Goods Identification Module: This module is responsible for comparing the topological features of the pixel association map of the genuine product with the preset genuine product. If the value is below the preset deviation threshold, the product is identified as genuine; otherwise, it is identified as counterfeit or infringing. Specifically, the output of the data acquisition module is connected to the input of the data preprocessing module; the output of the data preprocessing module is connected to the input of the temporal feature extraction module; the output of the temporal feature extraction module is connected to the input of the pixel association graph construction module; the output of the pixel association graph construction module is connected to the input of the graph topology feature calculation module; and the output of the graph topology feature calculation module is connected to the input of the counterfeit and infringing goods identification module.
2. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The data preprocessing module includes the following steps: Calculate the mean and standard deviation of the reflected light intensity of all dynamic spectral data in the dynamic spectral sequence, where the mean is... The expression is: Standard deviation The expression is: in, This represents the number of reflected light intensities from all dynamic spectral data in the dynamic spectral sequence. Indicates the first spectral sequence in the dynamic spectral sequence The reflected light intensity of each dynamic spectral data; If the reflected light intensity of a certain dynamic spectral data in a dynamic spectral sequence deviates from the mean reflected light intensity of all dynamic spectral data by more than three times the standard deviation, it is marked as abnormal dynamic spectral data and removed.
3. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The steps for extracting the decay half-life in the time-series feature extraction module include: First, obtain the initial value of the reflected light intensity of the coordinate pixel of each dynamic spectral data in the dynamic spectral sequence under different spectral frequency bands, and the time point when the reflected light intensity decays to half of the initial value; Then calculate the difference between two time points in the same spectral frequency band. This is the decay half-life, expressed as: in, Represents the initial time coordinates in pixels. In the spectral band The intensity of reflected light below ; Indicates time Time coordinate pixels In the spectral band The intensity of reflected light below ; Then, the average decay half-life under different spectral frequency bands is calculated as the final decay half-life.
4. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The steps for extracting the spectral shift in the time-series feature extraction module include: First, obtain the peak reflection wavelength of the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence under different spectral bands; Then, calculate the difference between the peak reflection wavelengths in the same spectral band. This is the spectral shift, expressed as: in, Represents the initial time coordinates in pixels. The peak reflection wavelength; Indicates time Time coordinate pixels The peak reflection wavelength; Then, the mean of the spectral shifts in different spectral frequency bands is calculated as the final spectral shift.
5. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The steps for extracting fluctuation entropy in the temporal feature extraction module include: First, obtain the coordinate pixels of each dynamic spectral data in the dynamic spectral sequence, along with the reflected light intensity and time points in different spectral frequency bands; Then, calculate the proportion of reflected light intensity to total reflected light intensity at each time point within the same spectral frequency band. The result obtained by substituting into the Shannon entropy formula is the fluctuation entropy; where the proportional conversion expression is: in, Indicates time Time coordinate pixels In the spectral band The intensity of reflected light below ; Represents all moments coordinate pixels In the spectral band Total reflected light intensity ; Shannon entropy The formula is: Measure the contribution of uncertainty at a single moment; Then, the mean value of the wave entropy under different spectral frequency bands is calculated as the final wave entropy.
6. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The steps for calculating the edge weights between nodes in the pixel association graph construction module include: First, obtain the reflected light intensity and time points of two adjacent nodes in different spectral frequency bands, and plot the curve of reflected light intensity changing with time to characterize the reflection attenuation law; then, input the two curves of reflected light intensity changing with time into the DTW algorithm to obtain the DTW distance between them, and map the DTW distance between them in reverse to the edge weight.
7. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The calculation steps for the tightly connected node feature in the graph topology feature calculation module include: First, count the number of neighbors for each node in the pixel association graph, and count the actual number of edges between the neighbors; Then, the local clustering coefficients of the nodes are calculated. The expression is: in, This indicates the actual number of edges between neighbors; Indicates the number of neighbors; Next, the mean of the local clustering coefficients of all nodes is calculated to obtain the average clustering coefficient. The expression representing the tight connectivity of neighboring nodes is: in, Indicates the number of nodes; This indicates the actual number of edges between neighbors; Indicates the number of neighbors.
8. The intelligent identification system for counterfeit and infringing goods based on dynamic spectral topology as described in claim 1, characterized in that, The calculation steps for node information propagation features in the graph topology feature calculation module include: First, calculate the shortest path distance between each pair of different nodes in the pixel association graph; Then, the information transmission speed of each pair of nodes is calculated based on the shortest path distance between each pair of nodes. The expression is: in, Represents a node To the node The shortest path distance; Next, the average information transmission speed of all node pairs is calculated to obtain the global node information transmission speed. The expression representing the characteristics of node information propagation is: in, Indicates the number of nodes; Represents a node To the node The shortest path distance.