Infringement and counterfeit commodity identification system and method based on visual perception technology

Abstract

The invention provides a visual perception technology-based infringement counterfeit commodity identification system and method, relates to the field of intelligent vision, and adopts a computer vision-based image processing and analysis technology to analyze multi-scale local features and global features of a target identification commodity image so as to identify an infringement counterfeit commodity. And whether the target identification commodity is a counterfeit commodity is determined. In this way, rapid and accurate identification of counterfeit commodities can be realized.

Description

Technical Field

[0001] The present application relates to the field of intelligent vision, and more specifically, to a system and method for identifying infringing and counterfeit goods based on visual perception technology. Background Art

[0002] In today's globalized business environment, the scale and frequency of commodity transactions have reached unprecedented heights. As the market prospers, the problem of counterfeit goods has become increasingly prominent, which will have a huge impact on consumer rights and brand reputation.

[0003] In traditional methods, the rapid identification of counterfeit goods often relies on simple manual comparison checks, but some counterfeit goods are very similar in appearance to genuine goods, and it is difficult to accurately judge by manual comparison of appearance alone. For example, some high-imitation sports shoes and clothing, ordinary consumers and even professional market regulators may not be able to distinguish the authenticity of some products from the location of the logo on the appearance. In some consumer scenarios, it is necessary to accurately judge whether the product is counterfeit, and the traditional accurate identification method requires the product to be sent to a professional identification agency for testing, which takes a long time and cannot meet the needs of accurate identification.

[0004] Therefore, there is a need for an optimized solution for identifying infringing and counterfeit goods based on visual perception technology. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiments of the present application provide a system and method for identifying infringing and counterfeit goods based on visual perception technology.

[0006] According to one aspect of the present application, a system for identifying infringing and counterfeit goods based on visual perception technology is provided, which includes:

[0007] A commodity image data collection module is used to obtain images of target identified commodities;

[0008] A commodity image local feature extraction module, used for extracting local features from the image of the target identified commodity to obtain a target identified commodity local feature map;

[0009] A product image local feature dimensionality reduction module is used to perform dimensionality reduction processing on the target identification product local feature map to obtain a target identification product local feature vector;

[0010] A commodity image global feature extraction module, used for performing long-distance dependency processing on the target identification commodity local feature map to obtain a target identification commodity global feature map;

[0011] A commodity image global feature dimension reduction module is used to perform dimension reduction processing on the target identification commodity global feature map to obtain a target identification commodity global feature vector;

[0012] A commodity image feature fusion module is used to align the local feature vector of the target commodity and the global feature vector of the target identified commodity at a low-dimensional feature granularity based on intrinsic attribute association to obtain a fused feature vector of the target identified commodity;

[0013] The commodity recognition result generation module is used to obtain the recognition result of the target identified commodity based on the fusion feature vector of the target identified commodity.

[0014] According to another aspect of the present application, a method for identifying infringing and counterfeit goods based on visual perception technology is provided, which includes:

[0015] Obtain an image of the target identification product;

[0016] Extracting local features of the image of the target identified commodity to obtain a local feature map of the target identified commodity;

[0017] Performing dimensionality reduction processing on the local feature map of the target identified commodity to obtain a local feature vector of the target identified commodity;

[0018] Performing long-distance dependency processing on the local feature map of the target identified commodity to obtain a global feature map of the target identified commodity;

[0019] Performing dimensionality reduction processing on the target identification commodity global feature map to obtain a target identification commodity global feature vector;

[0020] Performing low-dimensional feature granularity alignment on the target product local feature vector and the target identified product global feature vector based on intrinsic attribute association to obtain a target identified product fusion feature vector;

[0021] Based on the target identification commodity fusion feature vector, the identification result of the target identification commodity is obtained.

[0022] Compared with the prior art, the infringing and counterfeit goods identification system and method based on visual perception technology provided by the present application adopts image processing and analysis technology based on computer vision, and analyzes the multi-scale local features and global features of the target identification product image to determine whether the target identification product is a counterfeit product. In this way, fast and accurate identification of counterfeit goods can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0024] Figure 1 4 is a system block diagram of an infringing and counterfeit goods identification system based on visual perception technology according to an embodiment of the present application.

[0025] Figure 2 Schematic diagram of data flow of the infringing and counterfeit goods identification system based on visual perception technology according to an embodiment of the present application.

[0026] Figure 3 It is a block diagram of a commodity image local feature extraction module in an infringing and counterfeit commodity identification system based on visual perception technology according to an embodiment of the present application.

[0027] Figure 4 Flow chart of a method for identifying infringing and counterfeit goods based on visual perception technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0029] In today's globalized business environment, commodity trading is booming, with the scale and frequency of transactions reaching unprecedented heights. While this prosperous market situation has brought many opportunities, it has also spawned a series of thorny problems, among which the proliferation of counterfeit goods is particularly prominent.

[0030] The impact of counterfeit goods on consumer rights and brand reputation is extremely huge. The traditional means of dealing with counterfeit goods mainly relies on simple manual comparison checks to quickly identify counterfeit goods. However, this method has serious limitations. For example, in some common commodity fields, such as sports shoes and clothing industries, there are many high-imitation products. These high-imitation products are so similar to the originals in appearance that it is difficult for ordinary consumers or market regulators with professional knowledge to distinguish the authenticity of the goods by simply checking the appearance of the logo position and other aspects.

[0031] In addition, in many consumer scenarios, it is necessary to quickly and accurately determine whether a product is a counterfeit or inferior product. However, traditional accurate identification methods often require the product to be sent to a professional identification agency for testing. This testing process is not only cumbersome and time-consuming, but also cannot meet the current actual needs of quickly and accurately identifying products.

[0032] In recent years, deep learning and neural networks have been widely used in computer vision, natural language processing, text signal processing and other fields. In addition, deep learning and neural networks have also demonstrated a level close to or even beyond that of humans in image classification, object detection, semantic segmentation, text translation and other fields. The development of deep learning and neural networks has provided new solutions and solutions for the identification of infringing and counterfeit goods.

[0033] Figure 1 4 is a system block diagram of an infringing and counterfeit goods identification system based on visual perception technology according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the infringing and counterfeit goods identification system based on visual perception technology according to an embodiment of the present application. Figure 1 and Figure 2 As shown, in the infringing and counterfeit goods identification system 100 based on visual perception technology, it includes: a product image data collection module 110, which is used to obtain the image of the target identification product; a product image local feature extraction module 120, which is used to extract local features of the image of the target identification product to obtain a local feature map of the target identification product; a product image local feature dimension reduction module 130, which is used to reduce the dimension of the local feature map of the target identification product to obtain a local feature vector of the target identification product; a product image global feature extraction module 140, which is used to perform long-distance dependency processing on the local feature map of the target identification product to obtain a global feature map of the target identification product; a product image global feature dimension reduction module 150, which is used to reduce the dimension of the global feature map of the target identification product to obtain a global feature vector of the target identification product; a product image feature fusion module 160, which is used to perform low-dimensional feature granularity alignment based on intrinsic attribute association on the local feature vector of the target identification product and the global feature vector of the target identification product to obtain a fused feature vector of the target identification product; and a product identification result generation module 170, which is used to obtain the identification result of the target identification product based on the fused feature vector of the target identification product.

[0034] In the embodiment of the present application, the commodity image data collection module 110 is used to obtain the image of the target identification commodity. It should be understood that the shape and outline of the commodity are one of its basic characteristics. For many commodities, the slight difference in shape and outline may be an important basis for distinguishing genuine products from counterfeit products. At the same time, the image data of the commodity also contains the color and texture information of the commodity. There may be obvious differences in color and texture between genuine and counterfeit products, and these differences may be caused by different production processes and material selections. By capturing and analyzing these features, it can help identify counterfeit goods. In general, the target identification commodity image contains the shape, color and texture information of the commodity. By capturing and analyzing the information in the image, fast and accurate identification of counterfeit goods can be achieved.

[0035] In the embodiment of the present application, the commodity image local feature extraction module 120 is used to extract local features of the image of the target identification commodity to obtain a target identification commodity local feature map. Specifically, Figure 3 FIG. 1 is a block diagram of a commodity image local feature extraction module in a counterfeit commodity identification system based on visual perception technology according to an embodiment of the present application. Figure 3 As shown, the commodity image local feature extraction module 120 includes: a first-scale local feature generation unit 121, used to pass the image of the target identification commodity through a first commodity image feature encoder to obtain a first-scale target identification commodity local feature map; a second-scale local feature generation unit 122, used to pass the image of the target identification commodity through a second commodity image feature encoder to obtain a second-scale target identification commodity local feature map; a multi-scale local feature fusion unit 123, used to fuse the first-scale target identification commodity local feature map and the second-scale target identification commodity local feature map to obtain the target identification commodity local feature map.

[0036] In an embodiment of the present application, the first-scale local feature generating unit 121 is used to pass the image of the target identification commodity through the first commodity image feature encoder to obtain a first-scale target identification commodity local feature map. The essence of the first commodity image feature encoder in the present application is a convolutional neural network model using a first-scale two-dimensional convolution kernel. It should be understood that in order to extract feature information that is helpful for judging the authenticity of the commodity from the acquired image of the target identification commodity, these feature information include but are not limited to the shape, color, texture, etc. of the commodity, it is necessary to perform feature encoding processing on the image of the target identification commodity. Considering that the convolutional neural network (CNN) has excellent performance in the field of image processing, can automatically learn the features in the input image, and has high efficiency and accuracy in processing complex image tasks, based on this, in the technical solution of the present application, the image of the target identification commodity is passed through the first commodity image feature encoder to obtain a first-scale target identification commodity local feature map. Through the processing of the first commodity image feature encoder, the original pixel information in the image is converted into a high-level feature representation, which can capture the key information in the image.

[0037] In an embodiment of the present application, the second-scale local feature generation unit 122 is used to pass the image of the target identification product through a second product image feature encoder to obtain a second-scale target identification product local feature map. The essence of the second product image feature encoder in the present application is a convolutional neural network model using a second-scale two-dimensional convolution kernel, and the first scale is different from the second scale. It should be understood that the scale of the convolution kernel refers to the size of the convolution kernel, that is, the size of the area covered when the convolution kernel slides on the image. The smaller-scale convolution kernel (the first-scale convolution kernel in the present application) can capture the detailed features of the image, which are very important for identifying subtle differences in the product. In order to identify the overall structure and appearance characteristics of the product, and then to more comprehensively understand the characteristics of the product and thus improve the accuracy of recognition, the image of the target identification product is passed through the second product image feature encoder in the present application to obtain a second-scale target identification product local feature map.

[0038] In an embodiment of the present application, the multi-scale local feature fusion unit 123 is used to fuse the first-scale target recognition commodity local feature map and the second-scale target recognition commodity local feature map to obtain the target recognition commodity local feature map. It should be understood that the single-scale feature map obtained by the image feature encoder may not be able to fully reflect the complexity and diversity of the commodity image. In order to integrate the feature information of different scales, so as to more accurately identify the commodity and reduce the possibility of misjudgment and missed judgment, the first-scale target recognition commodity local feature map and the second-scale target recognition commodity local feature map are fused in the technical solution of the present application to obtain the target recognition commodity local feature map. Through fusion, the feature information of the first-scale target recognition commodity local feature map and the second-scale target recognition commodity local feature map are fully utilized, which can make the recognition algorithm more robust and stable, and thus can maintain stable recognition performance in complex and changeable actual environments.

[0039] Specifically, in an embodiment of the present application, a possible implementation method of fusing the first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map to obtain the object recognition commodity local feature map may be: fusing the first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map using the following fusion formula to obtain the object recognition commodity local feature map; wherein the fusion formula is:

[0040]

[0041] Among them, F c The local feature map of the target product is identified, F a The local feature map of the first-scale target recognition product, F b Identify the local feature map of the commodity for the second scale target, It represents the addition of the elements at corresponding positions of the first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map, and α and β are weighting parameters used to control the balance between the first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map in the object recognition commodity local feature map.

[0042] In an embodiment of the present application, the commodity image local feature dimensionality reduction module 130 is used to perform dimensionality reduction processing on the target identification commodity local feature map to obtain the target identification commodity local feature vector. Specifically, in an embodiment of the present application, the commodity image local feature dimensionality reduction module is used to: perform mean pooling of each feature matrix along the channel dimension on the target identification commodity local feature map to obtain the target identification commodity local feature vector. It should be understood that the target identification commodity local feature map obtained after fusion usually has a higher dimension, which will increase the computational complexity and storage requirements of subsequent processing. In order to simplify the subsequent processing process, the feature map can be pooled and reduced in dimension. Specifically, through mean pooling, each feature matrix on the channel dimension in the target identification commodity local feature map can be processed to obtain a feature vector of lower dimension. It is worth mentioning that mean pooling can maintain important information in each feature matrix along the channel dimension while removing redundancy and noise, which helps to extract the most discriminative commodity features, thereby improving the accuracy of the recognition algorithm.

[0043] In the embodiment of the present application, the commodity image global feature extraction module 140 is used to perform long-distance dependency processing on the target identification commodity local feature map to obtain the target identification commodity global feature map. Specifically, in the embodiment of the present application, the commodity image global feature extraction module is used to: pass the target identification commodity local feature map through a non-local neural network to obtain the target identification commodity global feature map. It should be understood that there is often a certain correlation between the local features of the commodity. In order to capture and utilize this correlation and further refine the features that are more valuable for identification, it is necessary to use the target identification commodity local feature map obtained by processing and fusion of multiple image encoders as input and process it through a non-local neural network to obtain the target identification commodity global feature map. The non-local neural network realizes the extraction of global features by capturing the relationship between any two points in the image. Specifically, the non-local neural network usually includes an input layer, a non-local module and an output layer. The non-local module is the core part of the non-local neural network, which captures global information by calculating the similarity between any two points in the image and performing weighted summation based on this similarity. By reprocessing the local feature map of the target product through a non-local neural network, the correlation between the various parts of the original local feature map can be increased, thereby more comprehensively describing the characteristics of the product.

[0044] Specifically, in an embodiment of the present application, a possible implementation method of passing the target identification commodity local feature map through a non-local neural network to obtain the target identification commodity global feature map can be: performing a first point convolution process, a second point convolution process and a third point convolution process on the target identification commodity local feature map respectively to obtain a first target identification commodity local feature map, a second target identification commodity local feature map and a third target identification commodity local feature map; calculating the position-weighted sum between the first target identification commodity local feature map and the second target identification commodity local feature map to obtain a target identification commodity local fusion feature map; inputting the target identification commodity local fusion feature map into a Softmax function to convert the local fusion feature map of the target identification commodity at each position into the local fusion feature map of the target identification commodity. The feature values ​​are mapped into the probability space to obtain a normalized target recognition product local fusion feature map; the position point multiplication between the normalized target recognition product local fusion feature map and the third target recognition product local feature map is calculated to obtain a target recognition product local re-fusion feature map; the target recognition product local re-fusion feature map is embedded in a Gaussian similarity function to obtain a target recognition product global similarity feature map; the target recognition product global similarity feature map is subjected to a fourth point convolution process to adjust the number of channels of the global similarity feature map to obtain a target recognition product channel-adjusted global similarity feature map; the position-weighted sum of the target recognition product channel-adjusted global similarity feature map and the target recognition product local feature map is calculated to obtain the target recognition product global feature map.

[0045] In an embodiment of the present application, the commodity image global feature dimension reduction module 150 is used to perform dimension reduction processing on the target identification commodity global feature map to obtain the target identification commodity global feature vector. Specifically, in an embodiment of the present application, the commodity image global feature dimension reduction module is used to: perform mean pooling of each feature matrix along the channel dimension on the target identification commodity global feature map to obtain the target identification commodity global feature vector. It should be understood that the target identification commodity global feature map contains rich information about the target identification commodity, but directly processing this information may face problems such as large amount of calculation, information redundancy, and excessive feature dimension. In order to effectively solve these problems and extract more representative features, in the technical solution of the present application, it is necessary to perform mean pooling of each feature matrix along the channel dimension on the target identification commodity global feature map to obtain the target identification commodity global feature vector. Through pooling, the computational efficiency of the model can be improved, and by extracting more representative features, more powerful support can be provided for subsequent commodity identification.

[0046] In the embodiment of the present application, the commodity image feature fusion module 160 is used to align the local feature vector of the target commodity and the global feature vector of the target identification commodity based on the low-dimensional feature granularity of the intrinsic attribute association to obtain the fused feature vector of the target identification commodity. It should be understood that the local feature vector of the target commodity usually focuses on the details in the original commodity image, while the global feature vector of the target identification commodity is more inclined to the global information in the original commodity image. In order to make full use of the respective advantages of these two feature vectors, the local feature vector of the target commodity and the global feature vector of the target identification commodity are fused in the technical solution of the present application. In particular, considering that the local feature vector of the target commodity is obtained by extracting the local feature encoding of the commodity image, it usually focuses on the details and local information of the image, and these features are often sensitive to the local area changes of the target identification commodity; while the global feature vector of the target identification commodity is extracted by a non-local neural network, which focuses on the overall information and captures the global features, and it is less sensitive to local changes; these two feature vectors represent different information dimensions in the target identification commodity image, and their distribution and structure in the feature space are completely different, that is, there is obvious feature heterogeneity. In the process of fusing the local feature vector of the target product and the global feature vector of the target identification product, due to the large heterogeneity between the two vectors, the fluidity formed in the feature space may not be smooth enough, which will affect the effect of feature fusion. At the same time, the heterogeneity will also lead to the mismatch of the feature space, further affecting the effective expression of the fused features. Therefore, in the technical solution of the present application, the local feature vector of the target product and the global feature vector of the target identification product are aligned at a low-dimensional feature granularity based on the intrinsic attribute association to obtain the fused feature vector of the target identification product.

[0047] Specifically, in an embodiment of the present application, the local feature vector of the target product and the global feature vector of the target identification product are aligned at a low-dimensional feature granularity based on intrinsic attribute association to obtain a fused feature vector of the target identification product, including: constructing a fused distance topological matrix of the target identification product between the local feature vector of the target product and the global feature vector of the target identification product; constructing a fused feature value granularity association matrix of the target identification product between the local feature vector of the target product and the global feature vector of the target identification product; performing topological association modulation on the fused feature value granularity association matrix of the target identification product based on the fused distance topological matrix of the target identification product to obtain a fused topological modulation association matrix of the target identification product; using the fused topological modulation association matrix of the target identification product as a perspective co-projection space, projecting the local feature vector of the target product and the global feature vector of the target identification product to the perspective co-projection space to obtain a perspective-modulated local feature vector of the target product and a perspective-modulated global feature vector of the target identification product; and fusing the perspective-modulated local feature vector of the target product and the perspective-modulated global feature vector of the target identification product to obtain the fused feature vector of the target identification product.

[0048] In an embodiment of the present application, another possible implementation method of performing low-dimensional feature granularity alignment based on intrinsic attribute association on the local feature vector of the target product and the global feature vector of the target identified product to obtain a fused feature vector of the target identified product may be: the local feature vector of the target product and the global feature vector of the target identified product are processed according to the following formula to obtain a fused feature vector of the target identified product; wherein the formula is:

[0049] M t =D(V1,V2)

[0050]

[0051] V c =α·V1'+β·V2'

[0052] Wherein, V1 represents the local feature vector of the target product, V2 represents the global feature vector of the target product, and D(V1, V2) is the target product fusion distance topological matrix between the local feature vector of the target product and the global feature vector of the target product, that is, d i,j =d(v 1i ,v 2j ), d i,j The eigenvalue of the position (i, j) in the distance topology matrix of the target identification commodity is fused, d(v 1i ,v 2j) represents the distance between the feature value of the i-th position of the local feature vector of the target product and the feature value of the j-th position of the global feature vector of the target identified product, and V1 and V2 are both column vectors, M t represents the target identification commodity fusion distance topology matrix, T represents the transpose of the vector, Represents matrix multiplication, M v represents the target identification commodity fusion feature value granularity correlation matrix, exp represents the natural exponential function, M c represents the target recognition product fusion topological modulation association matrix, V1' represents the local feature vector of the target product modulated by the viewing angle, V2' represents the global feature vector of the target recognition product modulated by the viewing angle, α and β represent weighted hyperparameters, V c Represents the target identification commodity fusion feature vector.

[0053] That is, in the technical solution of the present application, the local feature vector of the target product and the global feature vector of the target identification product are aligned at a low-dimensional feature granularity based on the intrinsic attribute association, and the target identification product fusion distance topological matrix between the local feature vector of the target product and the global feature vector of the target identification product is first constructed. This step utilizes the idea of ​​graph theory, taking the characteristic values ​​of each position in the local feature vector of the target product and the global feature vector of the target identification product as nodes, and using the distance metric function to determine the low-dimensional embedding expression of the edges between nodes, so as to construct a feature expression that can quantify the topological association relationship between the local feature vector of the target product and the global feature vector of the target identification product. This step is crucial to understanding the spatial distribution of data points and their mutual connections, and also provides basic structural information for subsequent steps.

[0054] Next, a target identification commodity fusion feature value granularity association matrix between the target commodity local feature vector and the target identification commodity global feature vector is constructed. This step focuses on exploring the correlation between the internal attributes of the feature vector. In a specific example, the product between the transposed vector of the target commodity local feature vector and the target identification commodity global feature vector is calculated to obtain the target identification commodity fusion feature value granularity association matrix. Furthermore, the target identification commodity fusion feature value granularity association matrix is ​​topologically associated modulated based on the target identification commodity fusion distance topological matrix to obtain the target identification commodity fusion topological modulation association matrix. That is, the feature value granularity association information of the target commodity local feature vector and the target identification commodity global feature vector is driven to wander on the distance topological map to construct a low-dimensional modulated perspective co-projection space domain for mapping the target commodity local feature vector and the target identification commodity global feature vector to a continuous high-dimensional regression space attribute.

[0055] Next, the target identification product fusion topological modulation association matrix is ​​used as the perspective co-projection space, and the target product local feature vector and the target identification product global feature vector are projected into the perspective co-projection space to obtain the perspective modulation target product local feature vector and the perspective modulation target identification product global feature vector. That is, the key to this technical solution is to find an implicit third space with modulation capability, in which as much information of the original high-dimensional data as possible is retained, so that the converted new feature vector not only reduces the dimension, but also maintains the relative position relationship between the original features.

[0056] Finally, the local feature vector of the target product modulated by the perspective and the global feature vector of the target identification product modulated by the perspective are fused to obtain the fused feature vector of the target identification product. Fusion can be achieved in various forms, ranging from simple arithmetic average to complex ensemble learning algorithms, depending on the application scenario and the goals pursued. The final fused feature vector of the target identification product combines the main features of the two sets of input data and provides a more comprehensive data representation, which is not only conducive to improving the performance of the machine learning model, but also promotes the ability of cross-modal or multi-source data analysis, thereby significantly enhancing the understanding and utilization efficiency of complex data structures.

[0057] In an embodiment of the present application, the commodity identification result generation module 170 is used to obtain the identification result of the target identification commodity based on the target identification commodity fusion feature vector. Specifically, in an embodiment of the present application, the commodity identification result generation module is used to: pass the target identification commodity fusion feature vector through a classifier-based counterfeit commodity identifier to obtain a recognition result, and the recognition result is used to indicate whether the target identification commodity is a counterfeit commodity. It should be understood that in order to make full use of the rich information contained in the target identification commodity fusion feature vector to accurately classify and identify commodities, it is necessary to pass the target identification commodity fusion feature vector through a classifier-based counterfeit commodity identifier. After training and optimization, the classifier can learn the subtle differences between counterfeit commodities and authentic products. Using a classifier to classify the target identification fusion feature vector can accurately distinguish between counterfeit commodities and authentic products, thereby improving the accuracy of recognition.

[0058] In summary, the infringing counterfeit goods identification system 100 based on visual perception technology according to the embodiment of the present application is explained, which uses computer vision-based image processing and analysis technology to analyze the multi-scale local features and global features of the target identification product image to determine whether the target identification product is a counterfeit product. In this way, fast and accurate identification of counterfeit goods can be achieved.

[0059] As described above, the infringing and counterfeit goods identification system 100 based on visual perception technology according to the embodiment of the present application can be implemented in various terminal devices, such as a server for infringing and counterfeit goods identification based on visual perception technology. In one example, the infringing and counterfeit goods identification system 100 based on visual perception technology according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the infringing and counterfeit goods identification system 100 based on visual perception technology can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the infringing and counterfeit goods identification system 100 based on visual perception technology can also be one of the many hardware modules of the terminal device.

[0060] Alternatively, in another example, the infringing and counterfeit goods identification system 100 based on visual perception technology and the terminal device may also be separate devices, and the infringing and counterfeit goods identification system 100 based on visual perception technology may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0061] Figure 4 Flow chart of the method for identifying infringing and counterfeit goods based on visual perception technology according to an embodiment of the present application. Figure 4 As shown, in the method for identifying infringing and counterfeit goods based on visual perception technology, the method includes: S110, obtaining an image of the target identified product; S120, performing local feature extraction on the image of the target identified product to obtain a local feature map of the target identified product; S130, performing dimensionality reduction processing on the local feature map of the target identified product to obtain a local feature vector of the target identified product; S140, performing long-distance dependency processing on the local feature map of the target identified product to obtain a global feature map of the target identified product; S150, performing dimensionality reduction processing on the global feature map of the target identified product to obtain a global feature vector of the target identified product; S160, performing low-dimensional feature granularity alignment based on intrinsic attribute association on the local feature vector of the target identified product and the global feature vector of the target identified product to obtain a fused feature vector of the target identified product; S170, obtaining the identification result of the target identified product based on the fused feature vector of the target identified product.

[0062] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned method for identifying infringing and counterfeit goods based on visual perception technology have been referred to above. Figures 1 to 3 It has been introduced in detail in the description of the infringing and counterfeit goods identification system based on visual perception technology, and therefore, its repeated description will be omitted.

[0063] In summary, the infringing counterfeit goods identification method based on visual perception technology in the embodiment of the present application is explained, which uses computer vision-based image processing and analysis technology to analyze the multi-scale local features and global features of the target identification product image to determine whether the target identification product is a counterfeit product. In this way, fast and accurate identification of counterfeit goods can be achieved.

[0064] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details of the above application are only for the purpose of illustration and ease of understanding, rather than limitation, and the above details do not limit the present application to the use of the above specific details to be implemented.

Claims

1. A system for identifying infringing and counterfeit goods based on visual perception technology, characterized in that: include: A commodity image data collection module is used to obtain images of target identified commodities; A commodity image local feature extraction module, used for extracting local features from the image of the target identified commodity to obtain a target identified commodity local feature map; A product image local feature dimensionality reduction module is used to perform dimensionality reduction processing on the target identification product local feature map to obtain a target identification product local feature vector; A commodity image global feature extraction module, used for performing long-distance dependency processing on the target identification commodity local feature map to obtain a target identification commodity global feature map; A commodity image global feature dimension reduction module is used to perform dimension reduction processing on the target identification commodity global feature map to obtain a target identification commodity global feature vector; A commodity image feature fusion module is used to align the local feature vector of the target commodity and the global feature vector of the target identified commodity at a low-dimensional feature granularity based on intrinsic attribute association to obtain a fused feature vector of the target identified commodity; The commodity recognition result generation module is used to obtain the recognition result of the target identified commodity based on the fusion feature vector of the target identified commodity.

2. The infringing and counterfeit goods identification system based on visual perception technology according to claim 1 is characterized in that: The commodity image local feature extraction module comprises: A first-scale local feature generating unit, configured to pass the image of the target identification commodity through a first commodity image feature encoder to obtain a first-scale target identification commodity local feature map; A second-scale local feature generating unit, configured to pass the image of the target identification commodity through a second commodity image feature encoder to obtain a second-scale target identification commodity local feature map; The multi-scale local feature fusion unit is used to fuse the first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map to obtain the object recognition commodity local feature map.

3. The infringing and counterfeit goods identification system based on visual perception technology according to claim 2 is characterized in that: The commodity image local feature dimensionality reduction module is used to: perform mean pooling of each feature matrix along the channel dimension on the target identification commodity local feature map to obtain the target identification commodity local feature vector.

4. The infringing and counterfeit goods identification system based on visual perception technology according to claim 3 is characterized in that: The commodity image global feature extraction module is used to: pass the target identification commodity local feature map through a non-local neural network to obtain the target identification commodity global feature map.

5. The infringing and counterfeit goods identification system based on visual perception technology according to claim 4 is characterized in that: The commodity image global feature dimension reduction module is used to: perform mean pooling of each feature matrix of the target identification commodity global feature map along the channel dimension to obtain the target identification commodity global feature vector.

6. The infringing and counterfeit goods identification system based on visual perception technology according to claim 5 is characterized in that: The first product image feature encoder is a convolutional neural network model using a first-scale two-dimensional convolution kernel, and the second product image feature encoder is a convolutional neural network model using a second-scale two-dimensional convolution kernel.

7. The infringing and counterfeit goods identification system based on visual perception technology according to claim 6 is characterized in that: The commodity image feature fusion module is used to: Constructing a target identification commodity fusion distance topological matrix between the target commodity local feature vector and the target identification commodity global feature vector; Constructing a target identification commodity fusion feature value granularity association matrix between the target commodity local feature vector and the target identification commodity global feature vector; Based on the target identification commodity fusion distance topological matrix, the target identification commodity fusion feature value granularity association matrix is ​​topologically modulated to obtain a target identification commodity fusion topological modulation association matrix; Taking the target identification commodity fusion topological modulation association matrix as the perspective co-projection space, projecting the target commodity local feature vector and the target identification commodity global feature vector into the perspective co-projection space to obtain the perspective modulation target commodity local feature vector and the perspective modulation target identification commodity global feature vector; The local feature vector of the target commodity modulated by viewing angle and the global feature vector of the target commodity recognized by viewing angle are fused to obtain the fused feature vector of the target commodity recognized by viewing angle.

8. The infringing and counterfeit goods identification system based on visual perception technology according to claim 7 is characterized in that: The commodity recognition result generating module is used to: pass the target identified commodity fusion feature vector through a classifier-based counterfeit commodity identifier to obtain a recognition result, and the recognition result is used to indicate whether the target identified commodity is a counterfeit commodity.

9. A method for identifying infringing and counterfeit goods based on visual perception technology, characterized in that: include: Obtain an image of the target identification product; Extracting local features of the image of the target identified commodity to obtain a local feature map of the target identified commodity; Performing dimensionality reduction processing on the local feature map of the target identified commodity to obtain a local feature vector of the target identified commodity; Performing long-distance dependency processing on the local feature map of the target identified commodity to obtain a global feature map of the target identified commodity; Performing dimensionality reduction processing on the target identification commodity global feature map to obtain a target identification commodity global feature vector; Performing low-dimensional feature granularity alignment on the target product local feature vector and the target identified product global feature vector based on intrinsic attribute association to obtain a target identified product fusion feature vector; Based on the target identification commodity fusion feature vector, the identification result of the target identification commodity is obtained.

10. The method for identifying infringing and counterfeit goods based on visual perception technology according to claim 9, characterized in that: Extracting local features of the image of the target identified commodity to obtain a local feature map of the target identified commodity includes: Passing the image of the target identification commodity through a first commodity image feature encoder to obtain a first-scale target identification commodity local feature map; Passing the image of the target identification commodity through a second commodity image feature encoder to obtain a second-scale target identification commodity local feature map; The first-scale object recognition commodity local feature map and the second-scale object recognition commodity local feature map are fused to obtain the object recognition commodity local feature map.

Images