Luxury picture recognition clustering method
By extracting the features of luxury goods image by using pre-trained CNN models and combining with the DBSCAN clustering algorithm for secondary classification, the problems of inefficient and insufficient accuracy of image classification in the luxury goods industry are solved, and efficient and accurate classification and clustering effects are achieved.
Patent Information
- Application Number
- CN202411800617.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-09
- Publication Date
- 2025-05-13
AI Technical Summary
In the luxury industry, traditional image classification methods are inefficient and poorly accurate, especially when processing large amounts of unlabeled pictures and noisy data, existing deep learning methods are difficult to effectively classify and generalize.
The pre-trained convolutional neural network (CNN) model is used for feature extraction, and the unclassified images are quadratic in combination with the DBSCAN clustering algorithm to dynamically generate luxury goods categories in the current market.
It significantly improves the accuracy and efficiency of luxury product picture classification, can effectively handle unlabeled data and noise points, and improves the accuracy and adaptability of classification.
Smart Images
Figure CN119992153A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of computer technology, and in particular relates to a method for identifying and clustering luxury goods images. Background Art
[0002] In today's e-commerce and online retail industry, the display and sale of luxury goods has gradually become an important part of the market. With the popularization of the Internet, luxury brands and major e-commerce platforms have published a large number of luxury pictures on their websites and applications to attract consumers and promote purchases. However, the diversity of luxury categories and the differences in the same type of luxury goods in different angles, styles, colors, etc. make it particularly difficult to effectively classify and manage a large number of luxury pictures.
[0003] Traditional image classification methods usually rely on manual annotation or simple image processing techniques. However, this method is inefficient and inaccurate when faced with large and diverse luxury goods images. Manual annotation is not only time-consuming and costly, but also prone to errors when processing a large number of images. In addition, traditional rule-based image processing methods cannot fully tap into the deep features in the image, resulting in classification results that cannot meet actual needs.
[0004] With the development of artificial intelligence and machine learning technologies, especially breakthroughs in the field of image recognition, image classification methods based on deep learning have gradually been applied. Deep learning algorithms can greatly improve the accuracy of classification by automatically learning high-level features in images. However, due to the complexity of the luxury industry and the diversity of similar products, existing deep learning methods still face some challenges, especially how to deal with a large number of unlabeled images and noisy data, and how to improve the generalization ability of the model in diverse luxury categories.
[0005] In the process of image classification, how to deal with the similarities and differences between different categories, especially in the absence of adequately labeled data, has become a difficult problem that needs to be solved. Traditional supervised learning methods rely on a large amount of labeled training data, but in reality, especially in the luxury industry, many product images lack accurate labels. Therefore, how to effectively classify these unlabeled images and improve the accuracy and efficiency of classification in a variety of product categories has become a problem that needs to be solved in current technology.
[0006] In view of the above problems, the present invention proposes an innovative luxury goods image recognition clustering method, which aims to improve the accuracy and robustness of luxury goods image classification through more intelligent means, especially when dealing with unlabeled data and noise points, to achieve more efficient and accurate classification effects. Summary of the invention
[0007] The purpose of the present invention is to provide a method for identifying and clustering luxury goods images in order to overcome the defects of the above-mentioned prior art.
[0008] The purpose of the present invention can be achieved by the following technical solutions:
[0009] The present invention provides a method for identifying and clustering luxury goods images, comprising the following steps:
[0010] Step S1: Obtain luxury goods picture sets of various categories;
[0011] Step S2: labeling the luxury features of the luxury image sets of each category;
[0012] Step S3: Input the labeled luxury goods pictures into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector set of each category;
[0013] Step S4: Obtain luxury goods pictures from various e-commerce platforms through a third-party platform, and use the obtained luxury goods pictures as a set of luxury goods pictures to be classified;
[0014] Step S5: classify the luxury picture set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network (CNN) model, and output a classified luxury picture set and an unclassified luxury picture data set, wherein the unclassified luxury picture data set includes unclassified luxury pictures and their corresponding feature vectors;
[0015] Step S6: performing secondary classification on the unclassified luxury goods image data set through a clustering algorithm to obtain a secondary classified luxury goods image set;
[0016] Step S7: merging the classified luxury goods picture set with the secondary classified luxury goods picture set to obtain luxury goods picture sets of various categories of luxury goods in the current market.
[0017] Furthermore, the luxury goods picture collection of each category includes pictures of the luxury goods of each category from various perspectives, and the luxury goods are classified by brand, style, and model.
[0018] Furthermore, the luxury goods characteristics include the luxury goods’ brand logo, color, shape, metal accessories, patterns and designs, and iconic decorations.
[0019] Furthermore, the convolutional neural network (CNN) model training process includes:
[0020] Collect a data set of luxury goods images. Each image is accompanied by a corresponding label, including the brand, style, and model of the luxury bag. Preprocess all images, resize them to 224x224 pixels, and normalize the pixel values to the [0,1] interval. Divide the image data set into training, validation, and test sets.
[0021] Select a convolutional neural network (CNN) model that has been pre-trained on a large-scale dataset, freeze the convolutional layers of the model, delete the original fully connected layers, and design a new fully connected layer for the classification task. The number of units in the output layer of the fully connected layer matches the number of luxury categories in the image dataset.
[0022] Initialize the weights of the fully connected layer and train the model using the training set. During the training process, adjust the weights of the fully connected layer using the back propagation algorithm and optimize the fully connected layer using the label information of the training set.
[0023] Unfreeze the convolutional layers of the model, train the entire network using the training set, and optimize the model using the loss function. During training, regularly evaluate the performance of the model on the validation set and adjust the hyperparameters based on the validation results.
[0024] After the training is completed, the test set is used to evaluate the performance of the model, calculate the accuracy, recall and F1 score indicators, and evaluate the classification effect of the model in each category. If the indicators reach the preset values, the training is completed.
[0025] Furthermore, the loss function is:
[0026]
[0027] Among them, L is the loss function of the convolutional neural network CNN model, C is the total number of categories, indicating the number of different categories in the classification task, and y c is the true label, p c is the probability of belonging to category c predicted by the model.
[0028] Furthermore, the feature vector set for each category includes feature vectors of multiple viewing angles of pictures of luxury goods in each category.
[0029] Furthermore, the method of classifying the luxury image set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network CNN, and outputting the classified luxury image set and the unclassified luxury image data set includes the following steps:
[0030] Input the set of luxury goods images to be classified into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector of each luxury goods image to be classified;
[0031] Calculate the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification;
[0032] The cosine similarity with the largest cosine similarity in the feature vector set of each category is taken as the similarity between the luxury goods image to be classified and the category;
[0033] Compare the maximum similarity of each category with a first preset value. If the maximum similarity is greater than or equal to the first preset value, classify the luxury goods image to be classified into the category. If the maximum similarity is less than the first preset value, place the luxury goods image to be classified into the unclassified luxury goods image data set.
[0034] Classify the luxury pictures to be classified, and output a set of classified luxury pictures and a dataset of unclassified luxury pictures.
[0035] Furthermore, the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification is calculated by the following formula:
[0036]
[0037] Among them, S AB is the cosine similarity between feature vector A and feature vector B, A and B are feature vectors, ||A|| and ||B|| are the moduli of feature vector A and feature vector B respectively, a i is the i-th element of eigenvector A, and n is the number of elements in eigenvector A.
[0038] Further, the step S6 includes the following steps:
[0039] Calculate the Euclidean distance between each image feature vector and other image feature vectors in the unclassified luxury image dataset, and use the Euclidean distance as the distance between images;
[0040] Set the initial parameters of the DBSCAN clustering algorithm, including the neighborhood radius and the minimum number of neighborhood points;
[0041] The distance between the unclassified luxury goods image dataset and the images is input into the DBSCAN clustering algorithm for secondary classification, and each cluster and the luxury goods images and noise points included in it are output;
[0042] The luxury pictures included in each cluster are regarded as a category, and the luxury pictures corresponding to each noise point are regarded as a category;
[0043] Output secondary classification luxury goods image set.
[0044] Furthermore, the Euclidean distance calculation formula is:
[0045]
[0046] Among them, d(A,B) is the Euclidean distance between eigenvector A and eigenvector B, a i With b i are the i-th elements of eigenvector A and eigenvector B respectively, and n is the number of elements in the eigenvector.
[0047] Compared with the prior art, the present invention has the following advantages:
[0048] (1) The present invention introduces a pre-trained convolutional neural network (CNN) model to automatically extract high-dimensional feature vectors of luxury goods images, reducing the workload of manual labeling and improving the accuracy and efficiency of image classification. This technology can accurately identify the brand, style, model and other features of luxury goods through the deep learning feature extraction capability of the convolutional layer, significantly improving the quality of classification.
[0049] (2) The present invention uses the DBSCAN clustering algorithm to perform secondary classification on unclassified images. By calculating the Euclidean distance between unclassified luxury images, the DBSCAN algorithm can effectively identify different luxury categories and classify noise points separately, which avoids the classification ambiguity problem that is prone to occur in traditional clustering algorithms. Therefore, the present invention can more accurately process complex market data and identify the potential classification of luxury images.
[0050] (3) The technical solution of the present invention can dynamically generate multiple categories of luxury goods in the current market by combining the classified images with the secondary classification results, timely adapt to the changes in the types of luxury goods in the market, and ensure the timeliness and accuracy of the classification results. This feature is of great significance for real-time updating of luxury goods image collections on e-commerce platforms.
[0051] (4) The clustering method of the present invention fully considers the diversity and variability of luxury goods images. By using the DBSCAN clustering algorithm to process images with different features, it can better solve problems such as high feature dimension and unclear categories, and improve the stability and robustness of the clustering process.
[0052] (5) The present invention further improves the refinement of classification by treating each noise point as an independent classification, effectively avoiding the influence of noise points on other categories, and ensuring the purity of classification and the accuracy of clustering.
[0053] (6) In practical applications, this method can combine the actual data of the e-commerce platform and automatically obtain the latest luxury goods pictures through the third-party platform. It has strong adaptive capabilities and can respond to dynamic changes in the luxury goods market in real time, thereby enhancing the adaptability and scalability of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0054] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION
[0055] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0056] Embodiment 1:
[0057] The present invention provides a method for identifying and clustering luxury goods images. Figure 1 As shown, the following steps are included:
[0058] Step S1: Obtain luxury goods picture sets of various categories;
[0059] Step S2: labeling the luxury features of the luxury image sets of each category;
[0060] Step S3: Input the labeled luxury goods pictures into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector set of each category;
[0061] Step S4: Obtain luxury goods pictures from various e-commerce platforms through a third-party platform, and use the obtained luxury goods pictures as a set of luxury goods pictures to be classified;
[0062] Step S5: classify the luxury image set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network (CNN) model, and output a classified luxury image set and an unclassified luxury image data set, where the unclassified luxury image data set includes unclassified luxury images and their corresponding feature vectors;
[0063] Step S6: performing secondary classification on the unclassified luxury goods image data set through a clustering algorithm to obtain a secondary classified luxury goods image set;
[0064] Step S7: merging the classified luxury goods picture set with the secondary classified luxury goods picture set to obtain luxury goods picture sets of various categories of luxury goods in the current market.
[0065] Among them, the luxury goods picture collections of each category include pictures of the luxury goods of each category from various angles, and the luxury goods are classified by the brand, style and model of the luxury goods.
[0066] Among them, luxury characteristics include luxury brand logos, colors, shapes, metal accessories, patterns and designs, and iconic decorations.
[0067] Among them, the convolutional neural network CNN model training process includes:
[0068] Collect a data set of luxury goods images. Each image is accompanied by a corresponding label, including the brand, style, and model of the luxury bag. Preprocess all images, resize them to 224x224 pixels, and normalize the pixel values to the [0,1] interval. Divide the image data set into training, validation, and test sets.
[0069] Select a convolutional neural network (CNN) model that has been pre-trained on a large-scale dataset, freeze the convolutional layers of the model, delete the original fully connected layers, and design a new fully connected layer for the classification task. The number of units in the output layer of the fully connected layer matches the number of luxury categories in the image dataset.
[0070] Initialize the weights of the fully connected layer and train the model using the training set. During the training process, adjust the weights of the fully connected layer using the back propagation algorithm and optimize the fully connected layer using the label information of the training set.
[0071] Unfreeze the convolutional layers of the model, train the entire network using the training set, and optimize the model using the loss function. During training, regularly evaluate the performance of the model on the validation set and adjust the hyperparameters based on the validation results.
[0072] After the training is completed, the test set is used to evaluate the performance of the model, calculate the accuracy, recall and F1 score indicators, and evaluate the classification effect of the model in each category. If the indicators reach the preset values, the training is completed.
[0073] Among them, the loss function is:
[0074]
[0075] Among them, L is the loss function of the convolutional neural network CNN model, C is the total number of categories, indicating the number of different categories in the classification task, and y c is the true label, p c is the probability of belonging to category c predicted by the model.
[0076] The feature vector set of each category includes feature vectors of multiple viewing angles of luxury goods of each category.
[0077] The process of classifying the luxury image set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network CNN, and outputting the classified luxury image set and the unclassified luxury image data set includes the following steps:
[0078] Input the set of luxury goods images to be classified into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector of each luxury goods image to be classified;
[0079] Calculate the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification;
[0080] The cosine similarity with the largest cosine similarity in the feature vector set of each category is taken as the similarity between the luxury goods image to be classified and the category;
[0081] Compare the maximum similarity of each category with a first preset value. If the maximum similarity is greater than or equal to the first preset value, classify the luxury goods image to be classified into the category. If the maximum similarity is less than the first preset value, place the luxury goods image to be classified into the unclassified luxury goods image data set.
[0082] Through the above process, the luxury pictures to be classified are classified, and a classified luxury picture set and an unclassified luxury picture data set are output.
[0083] Among them, the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification is calculated, and the calculation formula is:
[0084]
[0085] Among them, S AB is the cosine similarity between feature vector A and feature vector B, A and B are feature vectors, ||A|| and ||B|| are the moduli of feature vector A and feature vector B respectively, a i is the i-th element of eigenvector A, and n is the number of elements in eigenvector A.
[0086] Wherein, step S6 comprises the following steps:
[0087] Calculate the Euclidean distance between each image feature vector and other image feature vectors in the unclassified luxury image dataset, and use the Euclidean distance as the distance between images;
[0088] Set the initial parameters of the DBSCAN clustering algorithm, including the neighborhood radius and the minimum number of neighborhood points;
[0089] The distance between the unclassified luxury goods image dataset and the images is input into the DBSCAN clustering algorithm for secondary classification, and each cluster and the luxury goods images and noise points included in it are output;
[0090] The luxury pictures included in each cluster are regarded as a category, and the luxury pictures corresponding to each noise point are regarded as a category;
[0091] Output secondary classification luxury goods image set.
[0092] Among them, the Euclidean distance calculation formula is:
[0093]
[0094] Among them, d(A,B) is the Euclidean distance between eigenvector A and eigenvector B, a i With b i are the i-th elements of eigenvector A and eigenvector B respectively, and n is the number of elements in the eigenvector.
[0095] Embodiment 2:
[0096] The parts not mentioned in this embodiment are the same as those in Embodiment 1.
[0097] This embodiment provides a method for identifying and clustering luxury goods images, aiming to efficiently classify a large number of luxury goods images by using deep learning and clustering technology, especially when faced with unlabeled images and noisy data, to achieve accurate and efficient classification and clustering.
[0098] Step S1: Obtain luxury pictures of each category. This step collects luxury pictures of different categories to ensure that each category of luxury has sufficient sample data. By classifying luxury goods of different brands, styles, and models, the comprehensiveness and diversity of the data set are guaranteed, thereby providing rich training data for subsequent feature extraction and classification tasks.
[0099] Step S2: Label the luxury features of each luxury image set in each category. In this step, the brand, style, model and other features of each image are labeled manually or semi-automatically to provide label information for subsequent feature extraction. The advantage of this technical feature is that high-quality labeled data can improve the accuracy and reliability of model training and provide effective supervision signals for the classification process.
[0100] Step S3: Input the labeled luxury goods pictures into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector set of each category. By using the pre-trained convolutional neural network (CNN) model, representative high-level features can be automatically extracted from the picture. The introduction of this technical feature not only reduces the dependence on manual feature engineering, but also can extract more comprehensive and accurate image features through the training of the CNN model on a large-scale data set. The technical effect of this step is to significantly improve the classification accuracy, especially when processing complex luxury goods images, it can capture subtle visual differences such as brand and style.
[0101] Step S4: Obtain luxury goods images from various e-commerce platforms through a third-party platform, and use the obtained luxury goods images as a set of luxury goods images to be classified. By using a third-party platform, the latest and real market data can be automatically obtained from major e-commerce platforms, avoiding the high cost and time delay caused by manual collection. The advantage of this step is that it greatly improves the efficiency of data collection, ensures the real-time and richness of the image data set to be classified, and provides sufficient input data for subsequent classification.
[0102] Step S5: Classify the luxury picture set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network CNN model, and output the classified luxury picture set and the unclassified luxury picture data set, wherein the unclassified luxury picture data set includes unclassified luxury pictures and their corresponding feature vectors. This step performs preliminary classification by calculating the similarity between the feature vectors of the pictures to be classified and the labeled pictures. For pictures with high similarity, the classifier assigns them to the corresponding category. If the similarity is low, it is classified as an unclassified picture. This technical feature effectively matches the pictures to be classified with the existing categories, avoids the high time consumption and high error of all-manual classification, and ensures the efficiency and accuracy of classification.
[0103] Step S6: The unclassified luxury goods image dataset is reclassified by a clustering algorithm to obtain a reclassified luxury goods image dataset. By applying a clustering algorithm, such as DBSCAN (density-based spatial clustering algorithm), to the unclassified images, the unclassified images can be automatically reclassified. DBSCAN can identify dense areas in the data, automatically form different clusters, and identify noise points. The advantage of this step is that it can efficiently process unlabeled image data, especially the processing of noise data is more flexible, avoiding overfitting of the entire data set.
[0104] Step S7: Merge the classified luxury goods picture set with the secondary classified luxury goods picture set to obtain luxury goods picture sets of various categories of luxury goods in the current market. This step integrates the results of the preliminary classification and the secondary classification to form the final classification result. This technical feature can ensure the comprehensiveness of the classification results, and finally generate a high-quality luxury goods picture classification set by combining the stable classification structure obtained by the preliminary classification with the refined classification obtained by the secondary classification. This step not only improves the accuracy of the classification, but also ensures that the classification results can reflect the diversity of luxury goods on the market in real time, and improves the adaptability of the system.
[0105] This embodiment uses a combination of convolutional neural network (CNN) and clustering algorithm to process luxury goods image classification tasks, making full use of the automatic feature extraction capability of deep learning and the adaptive classification capability of clustering algorithm, significantly improving the accuracy and efficiency of luxury goods image classification. Through the automated data collection and classification process, the present invention can greatly reduce manual intervention, improve work efficiency, and maintain efficient and accurate processing capabilities when facing massive data. In addition, the secondary classification step effectively solves the noise problem that is difficult to handle in traditional classification methods through clustering algorithms, further improving the quality and stability of classification results. In general, the embodiments of the present invention provide an efficient, accurate and adaptable luxury goods image classification solution with broad application prospects and market value.
[0106] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.
[0107] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A method for identifying and clustering luxury goods images, characterized in that: The following steps are involved: Step S1: Obtain luxury goods picture sets of various categories; Step S2: labeling the luxury features of the luxury image sets of each category; Step S3: Input the labeled luxury goods pictures into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector set of each category; Step S4: Obtain luxury goods pictures from various e-commerce platforms through a third-party platform, and use the obtained luxury goods pictures as a set of luxury goods pictures to be classified; Step S5: classify the luxury picture set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network (CNN) model, and output a classified luxury picture set and an unclassified luxury picture data set, wherein the unclassified luxury picture data set includes unclassified luxury pictures and their corresponding feature vectors; Step S6: performing secondary classification on the unclassified luxury goods image data set through a clustering algorithm to obtain a secondary classified luxury goods image set; Step S7: merging the classified luxury goods picture set with the secondary classified luxury goods picture set to obtain luxury goods picture sets of various categories of luxury goods in the current market.
2. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The luxury goods picture collection of each category includes pictures of the luxury goods of each category from various perspectives, and the luxury goods are classified according to the brand, style, and model of the luxury goods.
3. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The luxury goods characteristics include the luxury goods’ brand logo, color, shape, metal accessories, patterns and designs, and iconic decorations.
4. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The convolutional neural network (CNN) model training process includes: Collect a data set of luxury goods images. Each image is accompanied by a corresponding label, including the brand, style, and model of the luxury bag. Preprocess all images, resize them to 224x224 pixels, and normalize the pixel values to the [0,1] interval. Divide the image data set into training, validation, and test sets. Select a convolutional neural network (CNN) model that has been pre-trained on a large-scale dataset, freeze the convolutional layers of the model, delete the original fully connected layers, and design a new fully connected layer for the classification task. The number of units in the output layer of the fully connected layer matches the number of luxury categories in the image dataset. Initialize the weights of the fully connected layer and train the model using the training set. During the training process, adjust the weights of the fully connected layer using the back propagation algorithm and optimize the fully connected layer using the label information of the training set. Unfreeze the convolutional layers of the model, train the entire network using the training set, and optimize the model using the loss function. During training, regularly evaluate the performance of the model on the validation set and adjust the hyperparameters based on the validation results. After the training is completed, the test set is used to evaluate the performance of the model, calculate the accuracy, recall and F1 score indicators, and evaluate the classification effect of the model in each category. If the indicators reach the preset values, the training is completed.
5. The method for identifying and clustering luxury goods images according to claim 4, characterized in that: The loss function is: Among them, L is the loss function of the convolutional neural network CNN model, C is the total number of categories, indicating the number of different categories in the classification task, and y c is the true label, p c is the probability of belonging to category c predicted by the model.
6. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The feature vector set for each category includes feature vectors of multiple viewing angles of luxury goods for each category.
7. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The method of classifying the luxury picture set to be classified according to the feature vector set of each classification and the pre-trained convolutional neural network CNN, and outputting the classified luxury picture set and the unclassified luxury picture data set includes the following steps: Input the set of luxury goods images to be classified into the pre-trained convolutional neural network (CNN) model for feature extraction, and output the feature vector of each luxury goods image to be classified; Calculate the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification; The cosine similarity with the largest cosine similarity in the feature vector set of each category is taken as the similarity between the luxury goods image to be classified and the category; Compare the maximum similarity of each category with a first preset value. If the maximum similarity is greater than or equal to the first preset value, classify the luxury goods image to be classified into the category. If the maximum similarity is less than the first preset value, place the luxury goods image to be classified into the unclassified luxury goods image data set. Classify the luxury pictures to be classified, and output a set of classified luxury pictures and a dataset of unclassified luxury pictures.
8. The method for identifying and clustering luxury goods images according to claim 7, characterized in that: The calculation formula for calculating the cosine similarity between the feature vector of the luxury goods image to be classified and each feature vector in the feature vector set of each classification is: Among them, S AB is the cosine similarity between feature vector A and feature vector B, A and B are feature vectors, ∥A∥ and ∥B∥ are the moduli of feature vector A and feature vector B respectively, a i is the i-th element of eigenvector A, and n is the number of elements in eigenvector A.
9. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The step S6 comprises the following steps: Calculate the Euclidean distance between each image feature vector and other image feature vectors in the unclassified luxury image dataset, and use the Euclidean distance as the distance between images; Set the initial parameters of the DBSCAN clustering algorithm, including the neighborhood radius and the minimum number of neighborhood points; The distance between the unclassified luxury goods image dataset and the images is input into the DBSCAN clustering algorithm for secondary classification, and each cluster and the luxury goods images and noise points included in it are output; The luxury pictures included in each cluster are regarded as a category, and the luxury pictures corresponding to each noise point are regarded as a category; Output secondary classification luxury goods image set.
10. The method for identifying and clustering luxury goods images according to claim 1, characterized in that: The Euclidean distance calculation formula is: Among them, d(A,B) is the Euclidean distance between eigenvector A and eigenvector B, a i With b i are the i-th elements of eigenvector A and eigenvector B respectively, and n is the number of elements in the eigenvector.