Appearance design patent infringement preliminary judgment system based on artificial intelligence large model

Abstract

The invention provides an appearance design patent infringement preliminary judgment system based on an artificial intelligence large model. The appearance design patent infringement preliminary judgment system comprises an appearance patent retrieval module, a feature extraction and similarity calculation module, an infringement risk assessment module and a system optimization and feedback module. The appearance patent retrieval module is used for retrieving patents possibly involving infringement and matching appearance design drawings of the patents; the feature extraction and similarity calculation module is used for extracting shape, pattern and color features of products and patents and calculating similarity; the infringement risk assessment module dynamically adjusts the weight according to the product property, the similarity calculation result and historical data, and gives an infringement risk score; and the system optimization and feedback module is used for training a model by using a machine learning algorithm through collecting actual infringement cases, optimizing weight distribution, and continuously updating the weight and calculation mode of the similarity calculation and infringement risk assessment module according to user feedback.

Description

Technical Field

[0001] The present invention relates to the field of patent infringement judgment, and particularly to a preliminary judgment system for design patent infringement based on an artificial intelligence large model. Background Art

[0002] The specific steps for judging design patent infringement include: Determine the scope of protection: According to the provisions of the Patent Law, the scope of protection of a design patent right shall be determined by the design patent product shown in the pictures or photos, including the front view, top view, side view, etc., among which the front view is the most important.

[0003] Judge the product category: Determine whether the design patent product and the infringing product belong to the same or similar goods. Usually, the function and use of the product are used as the criteria, and the International Design Classification is also referred to.

[0004] Conduct a comparison and judgment: Compare the design patent with the design of the accused infringing product to determine whether they are the same or similar. This judgment is usually made from the perspective of an ordinary consumer.

[0005] Currently, for many online sellers, it is very difficult for them to determine whether the products they sell involve patent infringement, especially design patent infringement. There are mainly two aspects. On the one hand, patent retrieval is a job with relatively high professional requirements. On the other hand, after finding similar patents, judging whether the appearance of the product is infringing is a job with even higher professionalism. Therefore, the situation of rampant design patent infringement in online sales has occurred. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the present invention provides a preliminary judgment system for design patent infringement based on an artificial intelligence large model, which gives an infringement risk score according to product properties, similarity calculation results, historical data, etc., so as to give sellers a preliminary judgment and early warning.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: A preliminary judgment system for design patent infringement based on an artificial intelligence large model, comprising: A design patent retrieval module, which is used to retrieve patents that may be involved in infringement and match the design drawings of the patents; A feature extraction and similarity calculation module, which is used to extract the shape, pattern, and color features of the product and the patent and calculate the similarity; An infringement risk assessment module, which dynamically adjusts the weights according to product properties, similarity calculation results, and historical data and gives an infringement risk score.

[0008] Preferably, the feature extraction and similarity calculation module further includes a shape feature extraction unit, which extracts the object contour using the Canny edge detection algorithm and calculates the global shape invariance using Hu moments. The calculation formula is as follows: Extract the object contour using the Canny edge detection algorithm, and its calculation formula is:

[0009] where I is the image, and x, y represent coordinate values; G(x, y) is the edge gradient; Calculate the global shape invariance (including rotation, scaling, and flipping) using Hu moments. The calculation formula is as follows:

[0010] where Mpg represents the p-th and q-th moments, indicating the invariance of the shape feature; x, y represent coordinate values, and is the centroid coordinate of the shape.

[0011] Preferably, the shape feature extraction unit also extracts local shape key points using the SIFT algorithm. The calculation formula is as follows:

[0012] where L(x, y, σ) is the Gaussian blur value at different scales, and the scale is σ; k is the scale factor, which adjusts the intensity of the Gaussian blur.

[0013] Preferably, the feature extraction and similarity calculation module further includes a pattern feature extraction unit, which extracts global texture features using Gabor filters and extracts local edge and texture features using the HOG algorithm. The calculation formulas are as follows: The Gabor filter response function is:

[0014] where x′, y′ are the coordinates after rotation and scaling; λ is the wavelength; θ is the direction; ψ is the phase offset; σ is the standard deviation of the Gaussian function; γ is the aspect ratio; HOG describes the local shape and texture by calculating the histogram of image gradient directions. The formula is:

[0015] where Gradient Magnitude is the magnitude of the image gradient, indicating the edge intensity; Angle Bin represents the grouping of gradient directions, describing the texture direction; HOG(x, y) represents a value of the direction gradient histogram generated at the pixel (x, y), reflecting the edge intensity of the pixel in a certain direction.

[0016] Preferably, the feature extraction and similarity calculation module further includes a color feature extraction unit. If the patent protects the color, a color histogram is used, and the Bhattacharyya coefficient is used to calculate the color similarity. The calculation formula is as follows:

[0017] where H 1 and H 2 are the color histograms of two images; are the normalized values of the two histograms.

[0018] Preferably, the similarity calculation module calculates the global shape similarity, local shape similarity, global pattern similarity, local pattern similarity, and color similarity respectively. The calculation formulas are as follows: The global shape similarity uses the Euclidean distance of Hu moments to calculate the shape similarity:

[0019] where H 1 and H 2 are the Hu moment features of the product and the patent; The local shape similarity uses SIFT key point matching and calculates the local shape similarity through the nearest neighbor distance ratio:

[0020] where Matched Keypoints represents the number of successfully matched key points; Total Keypoints represents the total number of key points; The global pattern similarity calculates the texture similarity through the response of the Gabor filter. The formula is:

[0021] where R(i) is the response value of the Gabor filter; The local pattern similarity uses HOG features and is compared based on the cosine similarity of the histogram:

[0022] where H1(i) and H2(i) are the histogram values of the HOG features; For color similarity calculation, the Bhattacharyya coefficient is used to calculate the color similarity:

[0023] where is the color difference metric calculated by the Bhattacharyya coefficient.

[0024] Preferably, the infringement risk assessment module automatically adjusts the weights of shapes and patterns according to the product type, and calculates the final infringement risk score based on the extracted features, similarity scores, and historical data. The calculation formula is as follows: R infingement =W shape ×(0.4×S shape-global +0.2×S shape-detail )+W pattern ×(0.2×S pattern-global +0.15×S pattern-detail )+0.05×S color 。

[0025] Preferably, for textile or cultural and creative products, the weight of the pattern is larger; for industrial products, the weight of the shape is larger; for hybrid products, the weights are dynamically adjusted according to actual needs and patent types.

[0026] Preferably, it further includes a system optimization and feedback module, which collects actual infringement cases, trains a model using machine learning algorithms, optimizes the weight allocation, and continuously updates the weights and calculation methods of the similarity calculation and infringement risk assessment modules according to user feedback.

[0027] Preferably, the machine learning algorithms include, but are not limited to, random forest or XGBoost algorithms, which are used to optimize the accuracy of infringement judgment.

[0028] Compared with the prior art, the present invention has the following beneficial effects: The present invention provides a preliminary judgment system for infringement of design patents based on an artificial intelligence large model, which retrieves patents that may be involved in infringement and matches the design drawings of the patents. By extracting features and calculating similarities from the design drawings, dynamically adjusts weights according to product properties, similarity calculation results, and historical data, and gives an infringement risk score to provide a preliminary judgment and warning for sellers. Detailed implementation manners

[0029] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious deformations. The basic principles defined in the following description can be used in other implementation schemes, deformation schemes, improvement schemes, equivalent schemes, and other technical schemes that do not deviate from the spirit and scope of the present invention.

[0030] The present invention provides a preliminary judgment system for infringement of design patents based on an artificial intelligence large model, including: An appearance patent retrieval module for retrieving patents that may be involved in infringement and matching the design drawings of the patents; The feature extraction and similarity calculation module is used to extract the shape, pattern, and color features of products and patents, and calculate the similarity; The infringement risk assessment module dynamically adjusts the weights according to the product nature, similarity calculation results, and historical data, and gives an infringement risk score; The system optimization and feedback module collects actual infringement cases, trains the model using machine learning algorithms, optimizes the weight allocation, and continuously updates the weights and calculation methods of the similarity calculation and infringement risk assessment modules according to user feedback.

[0031] It should be noted that the implementation method of this application is as follows: By inputting the design drawings or descriptions of products that may be involved in infringement, the system automatically searches in the patent database to find the design patents similar to the input; Using image processing and machine learning technologies, key features such as shape contours, pattern elements, and color distributions are extracted from the design drawings and patent drawings, and then the similarity between these features is calculated; Considering factors such as the market positioning, sales area, and target audience of the product, combining the similarity calculation results and historical infringement case data, using algorithms to dynamically adjust the weights, and finally giving a comprehensive infringement risk score; Through the established feedback mechanism, collect user feedback and new infringement case data in actual use, and use this data to train and optimize the machine learning model to improve the accuracy and reliability of the system.

[0032] Specifically, In the feature extraction and similarity calculation module of this application, feature extraction includes shape feature extraction unit, pattern feature extraction, and color feature extraction.

[0033] Furthermore, shape features are mainly extracted by two methods: global shape features and local shape features; The global shape features use the Canny edge detection algorithm to extract the object contour, and its calculation formula is:

[0034] Among them, I is the image, x, y represent coordinate values; G(x, y) is the edge gradient; The global shape features use Hu moments to calculate the global shape invariance (including rotation, scaling, and flipping), and the calculation formula is as follows:

[0035] Among them, Mpg represents the p-th and q-th moments, representing the invariance of shape features; x, y represent coordinate values, which are the centroid coordinates of the shape.

[0036] The local shape features use the SIFT (Scale-Invariant Feature Transform) algorithm to extract local shape key points. The SIFT descriptor formula is as follows:

[0037] where L(x, y, σ) is the Gaussian blur value at different scales, and the scale is σ; k is the scale factor that adjusts the intensity of the Gaussian blur.

[0038] Furthermore, the pattern features are divided into two categories: global pattern features and local pattern features; The global pattern features use the Gabor filter to extract texture features. The Gabor filter response function is as follows:

[0039] where x′, y′ are the coordinates after rotation and scaling; λ is the wavelength; θ is the direction; ψ is the phase offset; σ is the standard deviation of the Gaussian function; γ is the aspect ratio.

[0040] The local pattern features use HOG (Histogram of Oriented Gradients) to extract edge and texture features. HOG describes the local shape and texture by calculating the histogram of image gradient directions:

[0041] where Gradient Magnitude represents the magnitude of the image gradient, indicating the edge intensity; Angle Bin represents the grouping of gradient directions, describing the texture direction; HOG(x, y) represents a certain value of the histogram of oriented gradients generated at the pixel (x, y), reflecting the edge intensity of the pixel in a certain direction.

[0042] Furthermore, for color feature extraction (if the patent protects colors), a color histogram is used. When calculating the color similarity, the Bhattacharyya coefficient is used:

[0043] where H 1 and H 2 are the color histograms of two images; are the normalized values of the two histograms.

[0044] In the feature extraction and similarity calculation module of the present application, the similarity calculation is for the product and the patent image after feature extraction, and the similarity of various features needs to be calculated. The specific calculation process includes: shape similarity calculation, pattern similarity calculation, and color similarity calculation.

[0045] Furthermore, the shape similarity calculation includes global shape similarity and local shape similarity; The global shape similarity calculates the shape similarity using the Euclidean distance of Hu moments:

[0046] where H 1 and H 2 are the Hu moment features of the product and the patent.

[0047] The local shape similarity uses SIFT key point matching and calculates the local shape similarity through the nearest neighbor distance ratio:

[0048] where Matched Keypoints represents the number of successfully matched key points; Total Keypoints represents the total number of key points.

[0049] Furthermore, the pattern similarity calculation includes global pattern similarity and local pattern similarity; The global pattern similarity calculates the texture similarity through the response of the Gabor filter. The formula is:

[0050] where R(i) is the response value of the Gabor filter.

[0051] The local pattern similarity uses HOG features and is compared based on the cosine similarity of the histogram:

[0052] where H1(i) and H2(i) are the histogram values of the HOG features.

[0053] Furthermore, for color similarity calculation (if color is the protected object), the Bhattacharyya coefficient is used to calculate the color similarity:

[0054] where is the color difference metric calculated by the Bhattacharyya coefficient.

[0055] In this application, the infringement risk assessment module automatically adjusts the weights of the shape and pattern according to the product type, and calculates the final infringement risk score based on the extracted features and similarity scores, combined with historical data.

[0056] The specific adjustments are as follows: Textile / cultural and creative products: The pattern weight is greater. The recommended weight: Wpattern = 0.6; Wshape = 0.4 Industrial product category: Shape has a greater weight. Suggested weight: Wshape = 0.7, Wpattern = 0.3 Hybrid products: Based on actual needs and patent types, the system dynamically adjusts the weights according to historical data.

[0057] That is, for textile products or cultural and creative products, the weight of the pattern is greater; for industrial products, the weight of the shape is greater; for hybrid products, the weights are dynamically adjusted according to actual needs and patent types.

[0058] Furthermore, the infringement risk assessment module calculates the final infringement risk score based on the extracted features and similarity scores, combined with historical data. The scoring formula is: R infingement = W shape ×(0.4×S shape-global + 0.2×S shape-detail ) + W pattern ×(0.2×S pattern-global + 0.15×S pattern-detail ) + 0.05×S color This weight will be dynamically adjusted according to the collected data and infringement cases. The system optimizes the accuracy of infringement judgment through machine learning algorithms.

[0059] In the system optimization and feedback module of this application, by collecting actual infringement cases, using machine learning algorithms to train the model, optimizing the weight distribution, and continuously updating the weights and calculation methods of the similarity calculation and infringement risk assessment modules according to user feedback.

[0060] Among them, the machine learning algorithms include but are not limited to random forest or XGBoost algorithms, which are used to optimize the accuracy of infringement judgment.

[0061] Those skilled in the art should understand that the embodiments of the present invention shown in the above description are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been shown and described in the embodiments. Without departing from the said principles, the embodiments of the present invention can have any deformation or modification.

Claims

1. A preliminary judgment system for infringement of design patents based on artificial intelligence big model, characterized by: include: The appearance patent search module is used to search for patents that may involve infringement and match the appearance design drawings of the patents; Feature extraction and similarity calculation module, used to extract the shape, pattern, and color features of products and patents, and calculate the similarity; The infringement risk assessment module dynamically adjusts weights based on product properties, similarity calculation results and historical data, and gives an infringement risk score.

2. According to claim 1, a preliminary judgment system for infringement of design patents based on artificial intelligence big model is characterized by: The feature extraction and similarity calculation module further includes a shape feature extraction unit, which uses the Canny edge detection algorithm to extract the object contour and uses the Hu moment to calculate the global shape invariance. The calculation formula is as follows: The Canny edge detection algorithm is used to extract the object contour, and its calculation formula is: ; Where I is the image, x, y represent the coordinate values; G(x, y) is the edge gradient; The Hu moment is used to calculate the global shape invariance (including rotation, scaling, and flipping). The calculation formula is as follows: ; Among them, Mpg represents the p-order and q-order moments, indicating the invariance of shape features; x, y represent coordinate values, which are the centroid coordinates of the shape.

3. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 1 is characterized in that: The shape feature extraction unit also uses the SIFT algorithm to extract local shape key points, and the calculation formula is as follows: ; Where L(x,y,σ) is the Gaussian blur value at different scales, with the scale being σ; k is the scale factor that adjusts the intensity of the Gaussian blur.

4. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 1 is characterized in that: The feature extraction and similarity calculation module further includes a pattern feature extraction unit, which uses a Gabor filter to extract global texture features and uses a HOG algorithm to extract local edge and texture features. The calculation formulas are as follows: The Gabor filter response function is: ; where x′, y′ are the rotated and scaled coordinates; λ is the wavelength; θ is the direction; ψ is the phase shift; σ is the standard deviation of the Gaussian function; γ is the aspect ratio; HOG describes the local shape and texture by calculating the image gradient direction histogram formula: ; Among them, Gradient Magnitude is the size of the image gradient, indicating the edge strength; Angle Bin represents the grouping of gradient directions, describing the texture direction; HOG(x, y) represents a value of the directional gradient histogram generated at the pixel (x, y), reflecting the edge strength of the pixel in a certain direction.

5. According to claim 1, a preliminary judgment system for infringement of design patents based on artificial intelligence big model is characterized by: The feature extraction and similarity calculation module further includes a color feature extraction unit. If the patent protects the color, a color histogram is used and the Bhattacharyya coefficient is used to calculate the color similarity. The calculation formula is as follows: ; Where H1 and H2 are the color histograms of the two images; are the normalized values ​​of the two histograms.

6. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 1 is characterized in that: The similarity calculation module calculates global shape similarity, local shape similarity, global pattern similarity, local pattern similarity and color similarity respectively, and the calculation formulas are as follows: Global shape similarity uses the Euclidean distance of the Hu moment to calculate shape similarity: ; Among them, H1 and H2 are the Hu moment characteristics of products and patents; The local shape similarity uses SIFT key point matching and calculates the local shape similarity through the nearest neighbor distance ratio: ; Among them, Matched Keypoints indicates the number of key points that are successfully matched; TotalKeypoints indicates the total number of key points; The global pattern similarity calculates the texture similarity through the response of the Gabor filter, and the formula is: ; Where R(i) is the response value of the Gabor filter; The local pattern similarity uses HOG features and is compared based on the cosine similarity of the histogram: ; Where H1(i), H2(i) are the histogram values ​​of HOG features; Color similarity calculation, using Bhattacharyya coefficient to calculate color similarity: ;in, is a color difference measure calculated by the Bhattacharyya coefficient.

7. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 1 is characterized in that: The infringement risk assessment module automatically adjusts the weights of shape and pattern according to product type, and calculates the final infringement risk score based on the extracted features and similarity scores combined with historical data. The calculation formula is as follows: R infingement =W shape ×(0.4×S shape-global +0.2×S shape-detail )+W pattern ×(0.2×S pattern-global +0.15×S pattern-detail )+0.05×S color 。 8. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 7 is characterized in that: For textiles or cultural and creative products, the pattern has a greater weight; for industrial products, the shape has a greater weight; for mixed products, the weight is dynamically adjusted according to actual needs and patent type.

9. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 1 is characterized in that: It also includes a system optimization and feedback module, which collects actual infringement cases, uses machine learning algorithms to train models, optimizes weight distribution, and continuously updates the weights and calculation methods of the similarity calculation and infringement risk assessment modules based on user feedback.

10. The system for preliminary judgment of infringement of design patents based on artificial intelligence big model according to claim 9 is characterized in that: The machine learning algorithm includes but is not limited to a random forest or XGBoost algorithm, which is used to optimize the accuracy of infringement judgment.

Images