A single classification-based luxury product authentication method and system

Abstract

The application provides a luxury product identification method and system based on single classification, and the steps of the method comprise: obtaining an original leather tag sample image, performing image enhancement preprocessing on the original leather tag sample image; inputting the preprocessed original leather tag sample image into a pre-trained independent identification point segmentation network, wherein the independent identification point segmentation network outputs a segmented independent identification point segmentation image; inputting the independent identification point segmentation image into a single classification model pre-trained by using training data only comprising positive labels, wherein the single classification model outputs a positive sample reconstruction image; calculating a difference value of the independent identification point segmentation image and the positive sample reconstruction image, and determining an identification result based on the difference value. The present scheme can train only by using training data comprising positive labels, and adopts a single classification mode, thereby solving the problem of precision decline caused by sample imbalance of a luxury product data set.

Description

Technical Field

[0001] The present invention relates to the technical field of identification and authentication, and in particular to a method and system for authenticating luxury goods based on a single classification. Background Art

[0002] In today's society, luxury goods are not only synonymous with quality and craftsmanship, but also a symbol of status and a reflection of life attitude.

[0003] First, from the perspective of consumer protection, genuine luxury goods often represent superior quality and comprehensive after-sales service. Counterfeit and substandard products, on the other hand, not only lack quality assurance but also pose potential safety risks, seriously threatening consumers' health and user experience. Second, the healthy and orderly development of the luxury goods market requires rigorous authenticity verification. The proliferation of counterfeit luxury goods disrupts market order, harms the interests of legitimate brands and distributors, and ultimately undermines the innovation and vitality of the entire industry.

[0004] However, in the process of constructing luxury goods datasets, there are often situations where there is only a single type of data. There are often many reasons for this situation. For example, there are many manufacturers of counterfeit goods, and the differences between them are reflected in various aspects. For example, the level and craftsmanship of counterfeiting are uneven, and the materials used and the equipment series of counterfeiting will lead to various differences in counterfeit goods. The ultimate consequence of these differences is that the possibility of counterfeit series is endless, which makes it impossible to collect all the samples. In the current field of luxury goods authentication, existing technologies regard the authenticity identification of luxury goods as a classification problem. Therefore, they are limited by factors such as the imbalance and small number of samples in the luxury goods dataset, resulting in limited generalization performance and accuracy of the model. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a luxury goods authentication method based on a single classification to eliminate or improve one or more defects in the prior art.

[0006] One aspect of the present invention provides a method for identifying counterfeit luxury goods based on a single classification, the method comprising the following steps:

[0007] Acquire an original leather swab sample image, and perform image enhancement preprocessing on the original leather swab sample image;

[0008] Inputting the pre-processed original leather tag sample image into a pre-trained independent authentication point segmentation network, the independent authentication point segmentation network outputting a segmented independent authentication point segmentation image;

[0009] Inputting the independent counterfeit detection point segmentation image into a single classification model pre-trained using training data including only positive labels, the single classification model outputting a reconstructed image of a positive class sample;

[0010] Calculate a difference value of the independent fake identification point segmentation image and the positive class sample reconstruction image, and determine a fake identification result based on the difference value.

[0011] By using the above scheme, the single classification model of the scheme is trained by using only training data including positive labels, so that the reconstructed positive class sample reconstruction image is close to the true image. The scheme further calculates a difference value of the independent fake identification point segmentation image and the positive class sample reconstruction image. The greater the difference between the two, the greater the difference caused by the single classification model, and the greater the probability of being fake. In summary, the scheme can be trained by using only training data including positive labels, and the single classification model can solve the problem of precision decline caused by the imbalance of luxury product data set samples.

[0012] In some embodiments of the present application, the step of performing image enhancement preprocessing on the original leather tag sample image includes performing image enhancement using a pre-set image enhancement algorithm, and the image enhancement algorithm includes gamma correction, adaptive contrast enhancement algorithm and background noise filtering.

[0013] In some embodiments of the present application, the method further comprises the steps of:

[0014] Detecting straight lines in the original leather tag sample image by using a Hough transform;

[0015] Calculating the cutting and tilting angles of each straight line, and calculating the dispersion of the data by using a quartile range method to determine whether the original leather tag sample image is qualified.

[0016] In some embodiments of the present application, the independent fake identification point segmentation network is a YOLO v7 network.

[0017] In some embodiments of the present application, in the step of inputting the independent fake identification point segmentation image into a single classification model pre-trained by using only training data including positive labels, the single classification model outputs a positive class sample reconstruction image, and the single classification model includes a fine-grained encoder and a fine-grained decoder. The independent fake identification point segmentation image is sequentially processed by the fine-grained encoder and the fine-grained decoder, and a positive class sample reconstruction image is outputted.

[0018] In some embodiments of the present application, the fine-grained encoder includes a first encoding convolutional layer, a first encoding activation function layer, a second encoding convolutional layer, a first encoding normalization layer, a third encoding convolutional layer, a second encoding normalization layer, a fourth encoding convolutional layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding pooling layer, a hyperbolic tangent activation function layer and a Flatten layer connected in sequence.

[0019] In some embodiments of the present invention, in the step of sequentially processing the independent counterfeit detection point segmentation image through a fine-grained encoder and a fine-grained decoder and outputting a reconstructed image of a positive sample, the independent counterfeit detection point segmentation image is input into the first encoding convolution layer of the fine-grained encoder, and the latent vector is output through the Flatten layer.

[0020] In some embodiments of the present invention, the fine-grained decoder includes a sequentially connected zeroth decoding convolution layer, a zeroth decoding normalization layer, a first decoding convolution layer, a first decoding normalization layer, a second decoding convolution layer, a second decoding normalization layer, a third decoding convolution layer, a third decoding normalization layer, a Dropout layer, a fourth decoding convolution layer and a hyperbolic tangent activation function layer.

[0021] In some embodiments of the present invention, in the step of sequentially processing the independent detection point segmentation image through a fine-grained encoder and a fine-grained decoder and outputting a positive sample reconstructed image, the latent vector is input into the zeroth decoding convolution layer of the fine-grained decoder, and the positive sample reconstructed image is output through the hyperbolic tangent activation function layer of the fine-grained decoder.

[0022] In some embodiments of the present invention, in the step of calculating the difference value between the independent authentication point segmented image and the reconstructed image of the positive sample and determining the authentication result based on the difference value, the structural similarity index of the independent authentication point segmented image and the reconstructed image of the positive sample is calculated, and the calculated structural similarity index is compared with a preset threshold to determine the authentication result.

[0023] The second aspect of the present invention also provides a luxury goods authentication system based on a single classification, the system comprising a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, the processor being configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps implemented by the method described above.

[0024] The third aspect of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps implemented by the aforementioned luxury goods authentication method based on a single classification.

[0025] Additional advantages, objects, and features of the present invention will be described in part in the following description and will become apparent to those skilled in the art after studying the following or may be learned by practice of the present invention. The objects and other advantages of the present invention may be particularly pointed out and attained in the description and drawings.

[0026] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention.

[0028] Figure 1 Schematic diagram of an embodiment of the luxury goods authentication method based on a single classification according to the present invention;

[0029] Figure 2 Schematic diagram of the processing architecture of the training process of the luxury goods authentication method based on a single classification according to the present invention;

[0030] Figure 3 Schematic diagram of the architecture of a single classification model of the present invention;

[0031] Figure 4 Schematic diagram of the structure of the fine-grained encoder of the single classification model of the present invention;

[0032] Figure 5 Schematic diagram of the structure of the fine-grained decoder of the single classification model of the present invention. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0034] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0035] like Figure 1 As shown, the present invention proposes a method for identifying counterfeit luxury goods based on a single classification, the method comprising the following steps:

[0036] Step S100, obtaining an original leather tag sample image, and performing image enhancement preprocessing on the original leather tag sample image;

[0037] In the implementation process, in the step of image enhancement preprocessing of the original leather tag sample image, a plurality of image enhancement algorithms are used for processing, such as gamma correction, adaptive contrast enhancement algorithm, background noise filtering, which mainly processes the texture of the image, and the background noise is suppressed through the enhancement algorithm, and the core authentication area required is highlighted.

[0038] Step S200, input the preprocessed original leather tag sample image into the pre-trained independent authentication point segmentation network, and the independent authentication point segmentation network outputs a segmented independent authentication point segmentation image;

[0039] In the implementation process, the independent authentication point segmentation network is a YOLO v7 network.

[0040] In the implementation process, for a whole leather tag, the authentication points thereon are not all the same difference, for example, the R letter in PARIS has a large difference between the genuine product and the imitation product, and if the whole leather tag is sent into the network for learning, the content to be processed is too much, which will cause the attention of the model to be scattered and difficult to concentrate on the part with larger difference. Therefore, the independent authentication point segmentation network is used to segment the independent authentication points with key differences, which is also a hard attention mechanism, so as to guide the attention of the model to the more subtle core authentication difference.

[0041] Step S300, input the independent authentication point segmentation image into a single classification model pre-trained by training data including only forward labels, and the single classification model outputs a positive sample reconstruction image;

[0042] Step S400, calculate the difference value of the independent authentication point segmentation image and the positive sample reconstruction image, and determine the authentication result based on the difference value.

[0043] In the implementation process, the existing luxury product authentication technology identifies the true and false of luxury products as a classification problem, so it is constrained by the imbalance and small number of genuine and imitation product samples, which limits the generalization and performance, precision, etc. of the model. The present application proposes a single classification algorithm, which solves the luxury product authentication as a single classification problem, greatly alleviates the requirement for sample collection, and has high application ability.

[0044] Using the above scheme, the single classification model of this scheme is trained with training data that only includes positive labels. This scheme can make the reconstructed image of the positive sample approach the image of the true label. This scheme further calculates the difference value between the independent authentication point segmentation image and the reconstructed image of the positive sample. The greater the difference between the two, the greater the difference brought about by the processing of the single classification model, which means the greater the probability of false. In summary, this scheme can be trained with only training data with positive labels, and adopts a single classification method to solve the problem of decreased accuracy caused by the imbalance of samples in the luxury data set.

[0045] In some embodiments of the present invention, the step of performing image enhancement preprocessing on the original leather tag sample image includes performing image enhancement using a preset image enhancement algorithm, wherein the image enhancement algorithm includes gamma correction, adaptive contrast enhancement algorithm, and background noise filtering.

[0046] In the specific implementation process, gamma correction edits the gamma curve of the image to detect the dark and light parts in the image signal and increase the ratio of the two, thereby improving the image contrast effect; the basic idea of ​​the adaptive contrast enhancement algorithm is to divide the image into low-frequency and high-frequency parts for processing. The low-frequency part mainly reflects the background information of the image, while the high-frequency part contains the details and edge information of the image. By enhancing the high-frequency part, the contrast of the image can be effectively improved, making the details of the image clearer.

[0047] In some embodiments of the present invention, the method further comprises:

[0048] Using Hough transform to detect straight lines in the original leather tag sample image;

[0049] The tangent angle of each straight line was calculated, and the interquartile range method was used to calculate the dispersion of the data to determine whether the original leather swab sample image was qualified.

[0050] In some embodiments of the present invention, the independent authentication point segmentation network is a YOLO v7 network.

[0051] like Figure 3 、 4 As shown in Figure 5, in some embodiments of the present invention, in the step of inputting the independent counterfeit detection point segmentation image into a single classification model that is pre-trained using training data that only includes positive labels, and the single classification model outputs a reconstructed image of the positive class samples, the single classification model includes a fine-grained encoder and a fine-grained decoder, and the independent counterfeit detection point segmentation image is sequentially processed by the fine-grained encoder and the fine-grained decoder, and the positive class sample reconstructed image is output.

[0052] In some embodiments of the present invention, the fine-grained encoder includes a first encoding convolution layer, a first encoding activation function layer, a second encoding convolution layer, a first encoding normalization layer, a third encoding convolution layer, a second encoding normalization layer, a fourth encoding convolution layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding pooling layer, a hyperbolic tangent activation function layer and a Flatten layer connected sequentially.

[0053] In some embodiments of the present invention, in the step of sequentially processing the independent counterfeit detection point segmentation image through a fine-grained encoder and a fine-grained decoder and outputting a reconstructed image of a positive sample, the independent counterfeit detection point segmentation image is input into the first encoding convolution layer of the fine-grained encoder, and the latent vector is output through the Flatten layer.

[0054] In some embodiments of the present invention, the fine-grained decoder includes a sequentially connected zeroth decoding convolution layer, a zeroth decoding normalization layer, a first decoding convolution layer, a first decoding normalization layer, a second decoding convolution layer, a second decoding normalization layer, a third decoding convolution layer, a third decoding normalization layer, a Dropout layer, a fourth decoding convolution layer and a hyperbolic tangent activation function layer.

[0055] In some embodiments of the present invention, in the step of sequentially processing the independent detection point segmentation image through a fine-grained encoder and a fine-grained decoder and outputting a positive sample reconstructed image, the latent vector is input into the zeroth decoding convolution layer of the fine-grained decoder, and the positive sample reconstructed image is output through the hyperbolic tangent activation function layer of the fine-grained decoder.

[0056] Single classification network construction and algorithm flow: First is the codec at the core of the network. After our comparison, we finally adopted the codec structure of DCGAN. This is because our algorithm does not require a particularly complex codec, but we added a variable image size processing module consisting of a global average pooling and a batch normalization layer to DCGAN. This is because the original OCGAN can only run on a set of handwritten digits, and its resolution is 18*18, while our image resolution is 448*448. We can perform pooling based on images of different resolutions to achieve the same processing size. At the same time, since we will use randomly distributed vectors in the algorithm, we also proposed an image interpolation module. When the size of the processed image is smaller than the randomly distributed vector, we will use the interpolation algorithm to fill in the missing part, so that the two tensors will not be wrong during processing.

[0057] The single classification model of this solution adopts a single classification algorithm. The core of the single classification algorithm is a reconstruction-based algorithm. Since only the positive class is learned, the model will reconstruct all images into positive samples, and then judge them through the reconstruction error. This requires the model to be able to achieve the following two points: first, it can reconstruct positive samples very well, and second, it can reconstruct negative samples into positive samples very well. Only then will the reconstruction error be large, so it is necessary to limit the sample space of the model to the positive sample space, rather than diverging to all categories.

[0058] like Figure 2 and 3 As shown in the figure, the single classification algorithm also includes a visual discriminator Dv, a latent space discriminator Dl, a classifier C, and the encoder En and decoder De mentioned above. The parameter update of each module complies with the following rules:

[0059] All parameters except the classifier C are fixed, n is random noise, the input image is x, l1 is En(x+n), which represents the encoder's processing of the noise input, and l2 is a random distribution vector satisfying (-1,1). At this time, the updated classifier C is: C(De(l2),0)+C(De(l1),1), which means that the classifier is updated to the positive class 1 after decoding the input image, and the random distribution is updated to the negative class 0.

[0060] Update the latent space discriminator Dl: Dl(l1,0)+Dl(l2,1). At the latent space level, the closer it is to random distribution, the update is towards the positive class 1, and the closer it is to the input, the update is towards the negative class 0. The purpose is to limit the recognition ability of the model; update the visual discriminator Dv: Dv(De(l2),0)+Dv(x,1), which means that the closer it is to the real input image, the update is towards the positive class 1, and the closer it is to random distribution, the update is towards the negative class 0, so as to ensure that the visual discriminator can distinguish the real image.

[0061] Negative information mining, updating the random distribution of l2, through C(De(l2),1), that is, letting the classifier look for randomly distributed samples that can be identified as positive class 1, that is, looking for some negative information to increase the training difficulty of the model and improve the robustness of the model.

[0062] Update the encoder and decoder, and the loss function is constructed as Dl loss + Dv loss + λMse loss. Specifically, it is expressed as:

[0063] loss_ae_all=10.0*loss_ssim+loss_ae_v+loss_ae_l

[0064] loss_ae_all represents the total loss, 10.0*loss_ssim represents the λMse loss, loss_ae_v represents the Dl loss, and loss_ae_l represents the Dl loss.

[0065] The above scheme is adopted, where loss_ae_v is the visual space discriminator loss, that is, the loss after the visual space discriminator (disc_v) discriminates the image after decoding the randomly generated latent vector (l2); l2 is a latent vector randomly sampled from the uniform distribution [-1,1] and then processed by the tanh function; loss_ae_l is the latent space discriminator loss, which is the loss obtained by the latent space discriminator (disc_l) discriminating the latent vector (l1) output by the encoder; l1 is the latent vector obtained by encoding the real image with noise added through the encoder (encoder); loss_ssim is the SSIM structured coefficient loss between the reconstructed image and the input image.

[0066] In the specific implementation, the functions of the visual discriminator include:

[0067] 1. Determine the authenticity of the reconstructed image and distinguish between the original image and the reconstructed image;

[0068] 2. The input image (3 channels) is gradually reduced in dimension through a multi-layer convolutional network, and a discriminant score is finally output;

[0069] 3. Help the generator network generate more realistic images and form an adversarial learning mechanism;

[0070] 4. Finally, a score is output, and the output range is between 0-1 (using sigmoid activation), where 1 represents the real image and 0 represents the reconstructed / generated image.

[0071] Classifier:

[0072] It is used to determine whether an image is a real image or an abnormal image and to perform binary classification on the input samples.

[0073] During implementation, this solution introduces a horizontal text image screening mechanism during the construction of the training dataset. For the leather signature dataset, the key authentication targets required by this solution are arranged in the form of multiple lines of text, such as LOUIS VUITTON. These letters are neatly arranged on the leather signature, but due to the limited shooting angle, their arrangement direction may differ from the horizontal angle to a certain extent. This will cause the key authentication area to be non-horizontal when fed into the network. Therefore, this solution proposes a multi-line text screening algorithm that can be run on the leather signature dataset to screen out leather signatures with small horizontal deviations.

[0074] Specifically, the steps include:

[0075] Grayscale conversion:

[0076] I_gray(x,y)=0.299R(x,y)+0.587G(x,y)+0.114B(x,y)

[0077] Gaussian Blur:

[0078] I_blur(x,y)=I_gray(x,y)*G(x,y)

[0079] Where G(x,y) is the Gaussian kernel, defined as:

[0080] G(x,y)=(1 / (2πσ^2))*e^(-(x^2+y^2) / (2σ^2))

[0081] Canny edge detection:

[0082] Use the Canny algorithm to perform edge detection on the blurred image and obtain the edge image I_edge(x,y).

[0083] Detecting lines using modified Hough transform: We use the modified probabilistic Hough transform to detect text lines. The standard Hough transform represents a line as:

[0084] ρ=x cos(θ)+y sin(θ)

[0085] Where ρ is the distance from the line to the origin, and θ is the angle between the normal of the line and the x-axis.

[0086] Instead of detecting all possible lines, the improved probabilistic Hough transform randomly samples edge points and only considers lines that pass through these points. This method is more suitable for detecting text lines because it can detect line segments of finite length.

[0087] For each detected line segment l_i, we calculate its tilt angle α_i:

[0088] α_i=arctan((y2_i-y1_i) / (x2_i-x1_i))*(180 / π)

[0089] Among them, (x1_i,y1_i) and (x2_i,y2_i) are the two endpoints of the line segment.

[0090] We only keep line segments with angles in the range [-45°, 45°], since we are mainly interested in nearly horizontal text lines:

[0091] That is, angle A={α_i|-45°≤α_i≤45°}

[0092] To remove outliers, we use the interquartile range (IQR) method:

[0093] Calculate the first quartile Q1 and the third quartile Q3 of A

[0094] Calculate the interquartile range: IQR = Q3 - Q1

[0095] Define lower and upper bounds:

[0096] lower_bound=Q1-1.5*IQR

[0097] upper_bound=Q3+1.5*IQR

[0098] Filter Angle:

[0099] A_filtered={α_i∈A|lower_bound≤α_i≤upper_bound}.

[0100] In some embodiments of the present invention, in the step of calculating the difference value between the independent authentication point segmented image and the reconstructed image of the positive sample and determining the authentication result based on the difference value, the structural similarity index of the independent authentication point segmented image and the reconstructed image of the positive sample is calculated, and the calculated structural similarity index is compared with a preset threshold to determine the authentication result.

[0101] In some embodiments of the present invention, SSIM (Structure Similarity Index Measure) is a metric used to measure the similarity between two images. It primarily considers three key image features: luminance, contrast, and structure.

[0102] In some embodiments of the present invention, the step of determining the preset threshold value includes:

[0103] The range of values ​​for the calculation threshold of AUC based on the training and test data sets; AUC (Area Under the Curve) is the area under the curve, usually referring to the area under the ROC (Receiver Operating Characteristic) curve. It is a commonly used evaluation metric used to measure the classification performance of machine learning models at different thresholds;

[0104] Uniformly select values ​​within the threshold value range, and use the test data set to test each threshold to obtain the model accuracy corresponding to each threshold. The threshold corresponding to the highest model accuracy is used as the final threshold.

[0105] During the calculation of AUC:

[0106] Drawing of ROC curve: The ROC curve is obtained by calculating the true positive rate (TPR) and false positive rate (FPR) at different thresholds and plotting them. Each threshold corresponds to a point on the ROC curve;

[0107] AUC calculation: The AUC is calculated by calculating the area under the ROC curve. This area can be calculated by integrating the ROC curve, and the accuracy of the integral is related to the accuracy of the threshold. The higher the threshold value, the higher the accuracy of the result.

[0108] In summary, this solution uses a single classification approach and solves the positive and false identification of luxury goods as a single classification problem, removing the model's restrictions on the dataset and thus improving the model's recognition accuracy and robustness.

[0109] An embodiment of the present invention also provides a luxury goods authentication system based on a single classification, the system comprising a computer device, the computer device comprising a processor and a memory, the memory storing computer instructions, the processor being configured to execute the computer instructions stored in the memory, and when the computer instructions are executed by the processor, the system implements the steps implemented in the method described above.

[0110] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program implements the steps of the aforementioned method for authenticating luxury goods based on a single classification. The computer-readable storage medium may be a tangible storage medium, such as a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0111] It should be understood by those skilled in the art that the various exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is specifically performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0112] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0113] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0114] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for identifying counterfeit luxury goods based on a single classification, characterized by: The steps of the method include: Acquire an original leather swab sample image, and perform image enhancement preprocessing on the original leather swab sample image; Inputting the pre-processed original leather tag sample image into a pre-trained independent authentication point segmentation network, the independent authentication point segmentation network outputting a segmented independent authentication point segmentation image; The independent counterfeit detection point segmentation image is input into a single classification model pre-trained with training data including only positive labels, and the single classification model outputs a positive sample reconstructed image. The single classification model includes a fine-grained encoder and a fine-grained decoder. The fine-grained encoder includes a first encoding convolution layer, a first encoding activation function layer, a second encoding convolution layer, a first encoding normalization layer, a third encoding convolution layer, a second encoding normalization layer, a fourth encoding convolution layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding convolution layer, a second encoding normalization layer, a fourth encoding convolution layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding convolution layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding convolution layer, a second encoding activation function layer, a third encoding normalization layer, a first encoding convolution layer, a first encoding activation function ... encoding pooling layer, hyperbolic tangent activation function layer and Flatten layer; the fine-grained decoder includes a sequentially connected zeroth decoding convolution layer, a zeroth decoding normalization layer, a first decoding convolution layer, a first decoding normalization layer, a second decoding convolution layer, a second decoding normalization layer, a third decoding convolution layer, a third decoding normalization layer, a Dropout layer, a fourth decoding convolution layer and a hyperbolic tangent activation function layer; the independent counterfeit detection point segmentation image is sequentially processed by the fine-grained encoder and the fine-grained decoder, and a reconstructed image of a positive sample is output; Calculate the difference between the image segmented by the independent detection points and the reconstructed image of the positive sample, and determine the detection result based on the difference. In the step of determining the detection result based on the difference, calculate the structural similarity index between the image segmented by the independent detection points and the reconstructed image of the positive sample, compare the calculated structural similarity index with a preset threshold, and determine the detection result.

2. The luxury goods authentication method based on single classification according to claim 1, characterized in that: The step of performing image enhancement preprocessing on the original leather swab sample image includes performing image enhancement using a preset image enhancement algorithm, wherein the image enhancement algorithm includes gamma correction, an adaptive contrast enhancement algorithm, and background noise filtering.

3. The luxury goods authentication method based on single classification according to claim 1, characterized in that: The method further comprises the steps of: Using Hough transform to detect straight lines in the original leather tag sample image; The tangent angle of each straight line was calculated, and the interquartile range method was used to calculate the dispersion of the data to determine whether the original leather swab sample image was qualified.

4. The luxury goods authentication method based on single classification according to claim 1, characterized in that: The independent authentication point segmentation network is a YOLO v7 network.

5. The luxury goods authentication method based on single classification according to claim 1, characterized in that: In the step of sequentially processing the independent counterfeit detection point segmentation image through a fine-grained encoder and a fine-grained decoder and outputting a reconstructed image of a positive sample, the independent counterfeit detection point segmentation image is input into the first encoding convolution layer of the fine-grained encoder, and a latent vector is output through a Flatten layer.

6. A luxury goods authentication system based on a single classification, characterized by: The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions. The processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method according to any one of claims 1 to 5.

Images