A luggage identification and evaluation system based on AI multimodal large model

Abstract

The present invention proposes a luggage identification and evaluation system based on an AI multimodal large model, and relates to the technical field of luggage identification. Aiming at the technical problem that the luggage identification system of the existing multimodal fusion model lacks a dynamic weight calculation method based on historical data during data fusion, the present invention designs a multimodal data acquisition module, a judgment module, a preprocessing module, a feature extraction module, a feature fusion module, and a multimodal analysis module, wherein the calculation method of the preset weight can accurately calculate the preset weight of each feature vector, realize the dynamic adjustment of the weight of each feature vector according to the historical identification data, can more accurately evaluate the importance of different feature vectors, and adjust the weight accordingly, thereby improving the accuracy and reliability of the identification results.

Description

Technical Field

[0001] The present invention relates to the technical field of luggage identification, and in particular to a luggage identification and evaluation system based on an AI multimodal large model. Background Art

[0002] In the current field of luggage identification, in order to overcome the subjectivity, low efficiency and low accuracy of manual identification, some luggage identification systems based on large models have been put into use, and they have made some progress in providing identification results. However, these systems often only focus on the appearance characteristics or user evaluation of luggage, ignoring other important factors such as luggage materials, design elements, brand logos, etc. This single-dimensional evaluation method limits the depth and breadth of the evaluation, making it impossible for consumers to obtain comprehensive product information, affecting consumers' accurate judgment of product authenticity and quality.

[0003] To overcome the above problems, multimodal fusion technology came into being, making the data considered more comprehensive. However, the existing technology also has shortcomings in data fusion. For example, the Chinese invention patent with the publication (announcement) number CN118280522A proposes a personalized coronary intervention treatment planning system with multimodal data fusion. Although it can adjust the fusion strategy according to the data source information and gives a specific calculation method to realize the fusion processing of data, it fails to make full use of historical data in the fusion process. This approach lacks the ability to dynamically adjust the fusion weight. As historical data accumulates, its reference value is ignored, resulting in a certain blindness in the fusion method.

[0004] In summary, the existing technology has obvious limitations in luggage identification and evaluation, and there is an urgent need for a solution that can provide more efficient, accurate and comprehensive evaluation. Summary of the invention

[0005] The present invention provides a luggage identification and evaluation system based on an AI multimodal large model, so as to solve the technical problem that the existing luggage identification system of the multimodal fusion model lacks a dynamic weight calculation method based on historical data during data fusion.

[0006] A luggage identification and evaluation system based on AI multimodal large model, specifically including the following modules:

[0007] Multimodal data collection module: used to collect bag images, user reviews and transaction history;

[0008] Judgment module: connected to the multimodal data acquisition module, used to divide the user evaluation into evaluation text and evaluation image, calculate the proportion of luggage in the evaluation image, and if the evaluation image is a real photo and the proportion exceeds a threshold, retain the evaluation image, otherwise delete the evaluation image;

[0009] Preprocessing module: connected to the multimodal data acquisition module and the judgment module, and used to preprocess the luggage image and the transaction history, and the user evaluation output by the judgment module respectively;

[0010] Feature extraction module: connected to the preprocessing module, used to obtain feature vectors, specifically including image feature extraction submodule, evaluation feature extraction submodule and transaction history feature extraction submodule, wherein:

[0011] Image feature extraction submodule: used to obtain color, texture, and shape features based on the preprocessed bag image to form an image feature vector;

[0012] Evaluation feature extraction submodule: used to obtain descriptive features and sentiment tendencies based on preprocessed user evaluations to form evaluation feature vectors;

[0013] Transaction history feature extraction submodule: used to obtain the transaction times and times in the preprocessed transaction history to form a transaction history feature vector;

[0014] Feature fusion module: connected to the feature extraction module, used to sequentially connect the image feature vector, evaluation feature vector and transaction history feature vector according to preset weights to form a fusion feature vector; the preset weights are obtained according to the cosine similarity between the image feature vector, evaluation feature vector, transaction history feature vector in the historical identification data and the identification result and the user evaluation result;

[0015] Multimodal analysis module: connected to the feature fusion module, used to input the fused feature vector into the trained multimodal deep learning model, and output identification results and user evaluation results; the identification results include authenticity results and new and old results; the user evaluation results include descriptive features and emotional tendencies.

[0016] Furthermore, the system also includes a result output module and a historical data module, wherein:

[0017] Result output module: connected to the multimodal analysis module, used to generate an identification and evaluation report according to the identification result and the user evaluation result;

[0018] Historical data module: connected to the feature fusion module and the result output module, used to store the identification data into the historical data; the identification data includes: luggage images, user evaluations, transaction history, identification results and user evaluation results.

[0019] Furthermore, in the judgment module, the proportion of bags in the image is calculated by:

[0020] Step A1, obtaining the number of pixels H1 of the image;

[0021] Step A2, obtaining the number of pixels H2 occupied by the bag in the image;

[0022] Step A3: the proportion of bags in the image is α=H2 / H1.

[0023] Furthermore, in the preprocessing module,

[0024] The preprocessing of the luggage image includes image cropping, image enhancement, image denoising and normalization;

[0025] The preprocessing of the user evaluation includes text preprocessing and image preprocessing, the text preprocessing includes cleaning; the image preprocessing includes image cropping, image enhancement, image denoising and normalization;

[0026] The preprocessing of the transaction history includes data type conversion and normalization.

[0027] Furthermore, in the feature extraction module:

[0028] The image feature extraction submodule is a convolutional neural network, and the evaluation feature extraction submodule is a natural language processing model and a convolutional neural network.

[0029] Furthermore, in the evaluation feature extraction submodule,

[0030] Descriptive features include: product attributes, user experience, performance indicators, appearance evaluation and comparative description;

[0031] Emotional tendencies include: positive emotion, neutral emotion, and negative emotion.

[0032] Furthermore, in the feature fusion module, the method for obtaining the preset weights is:

[0033] Step B1, obtaining the historical records of bag identification and evaluation, including bag images, user evaluations, transaction history, and identification results and user evaluation results;

[0034] Step B2, normalizing the identification results and user evaluation results, and forming identification and evaluation feature vectors;

[0035] Step B3, obtaining an image feature vector according to the luggage image; obtaining an evaluation feature vector according to the user evaluation; and obtaining a transaction history feature vector according to the transaction history;

[0036] Step B4, calculating the cosine similarity between each of the identification and evaluation feature vectors and the image feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the evaluation feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the transaction history feature vector;

[0037] Step B5, calculating a preset weight according to the calculated cosine similarity.

[0038] Furthermore, the calculation formulas of the preset weights ω1, ω2, and ω3 are as follows:

[0039] ;

[0040] ;

[0041] ;

[0042] in,

[0043] ω1 is the weight of the image feature vector; ω2 is the weight of the evaluation feature vector; ω3 is the weight of the transaction history feature vector; S 1i is the cosine similarity between the identification result and the user evaluation result and the image feature vector in the i-th historical identification data; S 2i is the cosine similarity between the identification result and the user evaluation result and the evaluation feature vector in the i-th historical identification data; S 3i is the cosine similarity between the identification result and user evaluation result and the transaction history feature vector in the i-th historical identification data; n is the number of items in the historical identification data.

[0044] Furthermore, the multimodal deep learning model adopts the Transformer model.

[0045] Furthermore, the parameters of the multimodal deep learning model are: model dimension 512, number of attention heads 8, number of layers 6, feedforward network dimension 2048, learning rate 0.0003, batch size 32, training rounds 30, activation function GELU, optimizer AdamW.

[0046] Compared with the prior art, the beneficial effects of the present invention are:

[0047] First, the present invention first screens the luggage images in the user evaluation, and the system can automatically identify and prioritize those images that are real and have a high luggage ratio, thereby reducing the processing time for low-quality or irrelevant images and significantly improving the overall computing efficiency;

[0048] Secondly, the preset weight calculation method of the present invention can accurately calculate the preset weight of each feature vector, where the size of the weight reflects the contribution of each data source to the identification result. By dynamically adjusting the weight of each feature vector according to historical identification data, the system can more accurately evaluate the importance of different feature vectors and adjust the weight accordingly, thereby improving the accuracy and reliability of the identification results and ensuring the flexibility and precision of the identification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 This is a structural diagram of a luggage identification and evaluation system based on AI multimodal large model;

[0051] Figure 2 This is a flow chart of a method for calculating preset weights in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. Obviously, the described embodiments are only 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 belong to the scope of protection of the present invention.

[0053] The specific implementation of the present invention is described below with reference to the accompanying drawings.

[0054] In order to solve the technical problem that the existing luggage identification system of the multimodal fusion model lacks a dynamic weight calculation method based on historical data during data fusion, the present invention designs a multimodal data acquisition module, a judgment module, a preprocessing module, a feature extraction module, a feature fusion module, and a multimodal analysis module, wherein the calculation method of the preset weights can accurately calculate the preset weights of each feature vector, dynamically adjust the weights of each feature vector according to the historical identification data, more accurately evaluate the importance of different feature vectors, and adjust the weights accordingly, thereby improving the accuracy and reliability of the identification results.

[0055] like Figure 1 As shown, the present invention proposes a luggage identification and evaluation system based on an AI multimodal large model, which specifically includes the following modules:

[0056] Multimodal data collection module: used to collect bag images, user reviews and transaction history. Comprehensively collect bag images, user reviews and transaction history data to provide a rich data source for subsequent identification and evaluation.

[0057] Judgment module: connected to the multimodal data acquisition module, used to divide the user evaluation into evaluation text and evaluation image, calculate the proportion of luggage in the evaluation image, if the evaluation image is a real photo and the proportion exceeds the threshold, retain the evaluation image, otherwise delete the evaluation image.

[0058] Calculate the proportion of bags in the image. The specific method is:

[0059] Step A1, obtaining the number of pixels H1 of the image;

[0060] Step A2, obtaining the number of pixels H2 occupied by the bag in the image;

[0061] Step A3: the proportion of bags in the image is α=H2 / H1.

[0062] User reviews usually contain text descriptions and image descriptions, but images are uploaded by users themselves and may have nothing to do with the final identification and evaluation. Therefore, the text description and image description contained in the user-submitted evaluation are first identified. For the image description part, the module calculates the area or pixel ratio of the luggage in the image. If the image is actually taken by the user and the proportion of luggage in the image exceeds the threshold set by the system, it indicates that the image content is relevant to the evaluation and has a high reference value. The system will retain these images for subsequent analysis and identification processes. On the contrary, if the proportion of luggage in the image is insufficient, it means that the image may not be clear enough or the relevance is not high. The module deletes these images from the data set to improve data quality and reduce noise and irrelevant information in subsequent processing. This screening process ensures that only the most relevant and useful data are used for multimodal analysis, thereby enhancing the accuracy and reliability of identification and evaluation results.

[0063] Preprocessing module: connected to the multimodal data acquisition module and the judgment module, and used to preprocess the luggage image and the transaction history, and the user evaluation output by the judgment module respectively.

[0064] The preprocessing of luggage images includes:

[0065] Image cropping: remove unnecessary edges or background parts of the image and focus on the bag itself to reduce the interference of irrelevant information;

[0066] Image enhancement: Improve the contrast and clarity of the image to ensure that the image features are more obvious and facilitate subsequent feature extraction;

[0067] Image denoising: Use filtering techniques (such as Gaussian filtering and median filtering) to remove random noise in images and improve image quality;

[0068] Normalization: Scaling image pixel values ​​to a uniform range (such as 0 to 1 or -1 to 1) to facilitate data processing and analysis.

[0069] The preprocessing of user reviews includes text preprocessing and image preprocessing.

[0070] Text preprocessing includes:

[0071] Cleaning: Remove useless characters from the text, such as special symbols, HTML tags, etc., and retain useful information.

[0072] Image preprocessing includes image cropping, image enhancement, image denoising and normalization; the same as the preprocessing steps for luggage images, to ensure the quality of the images uploaded by users that are relevant to the evaluation.

[0073] The preprocessing of transaction history includes:

[0074] Data type conversion: ensure that all transaction history data (such as date, time, amount, quantity, etc.) are of the correct data type to facilitate statistics and analysis;

[0075] Normalization: Normalize numerical data to eliminate the impact of different dimensions and make different features comparable in the model.

[0076] Through the above preprocessing steps, the system can process and analyze data more effectively and improve the accuracy of luggage identification and evaluation, which not only helps to improve data quality, but also speeds up model training and improves the generalization ability of the model.

[0077] Feature extraction module: connected to the preprocessing module, used to obtain feature vectors, specifically including image feature extraction submodule, evaluation feature extraction submodule and transaction history feature extraction submodule, wherein:

[0078] Image feature extraction submodule: used to obtain color, texture, and shape features based on the preprocessed bag image to form an image feature vector;

[0079] The image feature extraction submodule is a convolutional neural network (CNN), which automatically learns the hierarchical features of images, such as edges, textures, shapes, and object parts, by using convolutional layers. It can capture local features and combine these local features through the network hierarchy to identify more complex patterns. By using pooling layers, CNN can reduce the spatial dimension of features while increasing the invariance to image displacement. Finally, the fully connected layer is used to map the learned features to the final task, such as classification or feature extraction.

[0080] Evaluation feature extraction submodule: used to obtain descriptive features and sentiment tendencies based on preprocessed user evaluations to form evaluation feature vectors;

[0081] In the evaluation feature extraction submodule, the descriptive features include: product attributes, usage experience, performance indicators, appearance evaluation and comparative description; the emotional tendencies include: positive emotion, neutral emotion and negative emotion.

[0082] User reviews contain both text and images. When extracting text features, a natural language processing model (NLP) can be used to understand and process human language. Sentiment analysis is an application of NLP, which can identify and classify the emotional tendencies in text, such as positive, negative, or neutral. Using recurrent neural networks (RNN) or long short-term memory networks (LSTM) to process sequence information in text data helps understand the context of the review text. The CNN model can be used to extract image features.

[0083] In other embodiments, other models that can process text and images simultaneously, such as a multimodal deep learning model, may also be used to comprehensively consider the characteristics of text and images.

[0084] Transaction history feature extraction submodule: used to obtain the transaction number and time in the preprocessed transaction history to form a transaction history feature vector.

[0085] By combining CNN and NLP models, the feature extraction module can extract rich features from image and text data for multimodal analysis to improve the accuracy of luggage identification and evaluation. CNN's powerful visual feature extraction capabilities and NLP's in-depth text analysis capabilities enable the system to more comprehensively understand user reviews and image content, thereby providing more reliable identification results.

[0086] Through the work of the feature extraction module, the original multimodal data (images, texts, transaction records) are converted into a series of numerical feature vectors, which capture the key information of the data and provide input for subsequent data analysis, model training and prediction. This method allows different data types to be compared and integrated under the same framework, enhances the model's ability to understand data, and improves the accuracy and efficiency of luggage identification and evaluation.

[0087] Feature fusion module: connected to the feature extraction module, used to connect the image feature vector, evaluation feature vector and transaction history feature vector in sequence according to preset weights to form a fused feature vector; the preset weights are obtained based on the cosine similarity between the image feature vector, evaluation feature vector, transaction history feature vector in the historical identification data and the identification results and user evaluation results.

[0088] The feature fusion module integrates image features, user evaluation features and transaction history features into a unified feature vector, which can comprehensively consider visual information (such as color, texture and shape) and non-visual information (such as user's emotional tendencies and transaction patterns) to obtain a more comprehensive understanding of luggage characteristics, improve the accuracy and efficiency of identification, and reduce errors caused by human factors.

[0089] The preset weight is obtained based on the cosine similarity between the image feature vector, the evaluation feature vector, the transaction history feature vector in the historical identification data and the identification result and the user evaluation result, such as Figure 2 As shown, the specific method is:

[0090] Step B1, obtaining the historical records of bag identification and evaluation, including bag images, user evaluations, transaction history, and identification results and user evaluation results;

[0091] Step B2, normalizing the identification results and user evaluation results, and forming identification and evaluation feature vectors;

[0092] Step B3, obtaining an image feature vector according to the luggage image; obtaining an evaluation feature vector according to the user evaluation; and obtaining a transaction history feature vector according to the transaction history;

[0093] Step B4, calculating the cosine similarity between each of the identification and evaluation feature vectors and the image feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the evaluation feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the transaction history feature vector;

[0094] Step B5, calculating a preset weight according to the calculated cosine similarity.

[0095] Furthermore, the calculation formulas of the preset weights ω1, ω2, and ω3 are as follows:

[0096] ;

[0097] ;

[0098] ;

[0099] in,

[0100] ω1 is the weight of the image feature vector; ω2 is the weight of the evaluation feature vector; ω3 is the weight of the transaction history feature vector; S 1i is the cosine similarity between the identification result and the user evaluation result and the image feature vector in the i-th historical identification data; S 2iis the cosine similarity between the identification result and the user evaluation result and the evaluation feature vector in the i-th historical identification data; S 3i is the cosine similarity between the identification result and user evaluation result and the transaction history feature vector in the i-th historical identification data; n is the number of items in the historical identification data.

[0101] The present invention provides a method for calculating the relative importance of different data sources in luggage identification. First, the system collects and marks historical identification records, including images, evaluations, and transaction histories. Next, the corresponding feature vectors are extracted from these marked data. Using these feature vectors, the system calculates the cosine similarity between them and the identification results and the user evaluation results. This metric reflects the consistency of the feature vectors and the evaluation results in direction. Finally, based on these similarity values, the system calculates and assigns preset weights, and the size of the weights directly reflects the contribution of each data type to the identification results. The present invention can adaptively adjust the weights of each modal data, ensuring that the model can be dynamically optimized based on historical data, thereby improving the accuracy and efficiency of identification. In this way, the system can identify which features are more critical to identification, and then give these features higher weights when fusion is performed, so as to achieve more accurate identification results.

[0102] The normalized identification and evaluation feature vectors, image feature vectors, evaluation feature vectors, and transaction history feature vectors have the same data format, and the relationship between the feature vectors can be obtained by calculating the cosine similarity. The specific cosine similarity calculation process is an existing technology and will not be explained in detail.

[0103] Multimodal analysis module: connected to the feature fusion module, used to input the fused feature vector into the trained multimodal deep learning model, and output identification results and user evaluation results; the identification results include authenticity results and new and old results; the user evaluation results include descriptive features and emotional tendencies.

[0104] The identification results cover two key aspects: authenticity and new and old results. Authenticity refers to the system's determination of whether the luggage is genuine based on multimodal data analysis, based on a detailed comparison of features such as brand logos, materials, and production processes. New and old results evaluate the use of luggage, judging whether it is brand new, second-hand, or worn, etc., involving the evaluation of damage to the appearance of the luggage, traces of use, and overall condition. User evaluation results are further refined into descriptive features and emotional tendencies. Descriptive features include users' specific descriptions of luggage, such as product attributes, usage experience, performance indicators, appearance evaluation, and comparative descriptions, providing direct feedback on the characteristics of luggage. Emotional tendencies reflect users' emotional attitudes towards luggage, which can be positive, neutral, or negative, and help understand consumer satisfaction and market acceptance. Based on the above results, the system can provide users with comprehensive and detailed identification and evaluation reports, thereby enhancing users' ability to judge product authenticity and quality.

[0105] The multimodal deep learning model uses the Transformer model with the following parameters: model dimension 512, number of attention heads 8, number of layers 6, feedforward network dimension 2048, learning rate 0.0003, batch size 32, training rounds 30, activation function GELU, and optimizer AdamW.

[0106] The multimodal analysis module is the core of the system, responsible for inputting the fused feature vector generated by the feature fusion module into the deep learning model for luggage identification and evaluation. The Transformer model used in this module is an advanced neural network architecture that is particularly suitable for processing sequence data and capturing long-distance dependencies, but its architecture can also be extended to the analysis of multimodal data.

[0107] Result output module: connected to the multimodal analysis module, used to generate an identification and evaluation report based on the identification result and the user evaluation result, and to display the identification result and evaluation report to the user in a visual manner.

[0108] Historical data module: connected to the feature fusion module and the result output module, used to store the identification data into the historical data; the identification data includes: luggage images, user evaluations, transaction history, identification results and user evaluation results.

[0109] The feature fusion module of the present invention makes full use of historical identification data to dynamically adjust the weights of each feature vector through a carefully designed preset weight calculation method, ensuring that the fused feature vector can accurately reflect the key characteristics of the luggage. The in-depth analysis of historical identification data provides a reliable basis for weight allocation, enabling the system to identify which features have a greater impact on the identification results, thereby giving these features higher weights during feature fusion. This method not only improves the accuracy and reliability of the identification results, but also enhances the system's generalization ability for different types of luggage identification tasks by continuously learning and adapting to new data.

[0110] It should be noted that the terms used in the present invention are only for describing specific embodiments, rather than limiting the scope of the present invention. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of more restrictions, the elements defined by the sentence "include one..." do not exclude the presence of other identical elements in the process, method or device including the elements.

[0111] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A luggage identification and evaluation system based on AI multimodal large model, characterized by: Includes the following modules: Multimodal data acquisition module: used to collect luggage images, user reviews and transaction history; A judgment module is connected to the multimodal data acquisition module and is used to divide the user evaluation into evaluation text and evaluation image, calculate the proportion of luggage in the evaluation image, and retain the evaluation image if the evaluation image is a real photo and the proportion exceeds a threshold; otherwise, delete the evaluation image; Preprocessing module: connected to the multimodal data acquisition module and the judgment module, used to preprocess the luggage image and the transaction history, and the user evaluation output by the judgment module respectively; Feature extraction module: connected to the preprocessing module, used to obtain feature vectors, specifically including image feature extraction submodule, evaluation feature extraction submodule and transaction history feature extraction submodule, in, Image feature extraction submodule: used to obtain color, texture, and shape features based on the preprocessed luggage image to form an image feature vector; Evaluation feature extraction submodule: used to obtain descriptive features and sentiment tendencies based on pre-processed user evaluations to form evaluation feature vectors; Transaction history feature extraction submodule: used to obtain the transaction number and time in the pre-processed transaction history to form a transaction history feature vector; Feature fusion module: connected to the feature extraction module, used to sequentially connect the image feature vector, evaluation feature vector, and transaction history feature vector according to preset weights to form a fused feature vector; the preset weights are obtained based on the cosine similarity between the image feature vector, evaluation feature vector, transaction history feature vector in the historical identification data and the identification results and user evaluation results; Multimodal analysis module: connected to the feature fusion module, used to input the fused feature vector into the trained multimodal deep learning model, and output identification results and user evaluation results; the identification results include authenticity results and new and old results; the user evaluation results include descriptive features and emotional tendencies.

2. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: The system also includes a result output module and a historical data module, wherein: Result output module: connected to the multimodal analysis module, used to generate an identification and evaluation report based on the identification results and the user evaluation results; Historical data module: connected to the feature fusion module and the result output module, used to store the identification data in the historical data; the identification data includes: luggage images, user evaluation, transaction history, identification results and user evaluation results.

3. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: In the judgment module, the proportion of bags in the image is calculated by: Step A1, obtaining the number of pixels H1 of the image; Step A2, obtaining the number of pixels H2 occupied by the bag in the image; Step A3: The proportion of bags in the image is α=H2 / H1.

4. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: In the pre-processing module, The preprocessing of the luggage image includes image cropping, image enhancement, image denoising and normalization; The preprocessing of the user evaluation includes text preprocessing and image preprocessing, wherein the text preprocessing includes cleaning; the image preprocessing includes image cropping, image enhancement, image denoising and normalization; The preprocessing of the transaction history includes data type conversion and normalization.

5. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: In the feature extraction module: The image feature extraction submodule is a convolutional neural network, and the evaluation feature extraction submodule is a natural language processing model and a convolutional neural network.

6. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: In the evaluation feature extraction submodule, Descriptive features include: product attributes, user experience, performance indicators, appearance evaluation and comparative description; Emotional tendencies include: positive emotions, neutral emotions, and negative emotions.

7. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: In the feature fusion module, the method for obtaining the preset weights is: Step B1, obtaining the historical records of bag identification and evaluation, including bag images, user evaluations, transaction history, and identification results and user evaluation results; Step B2: normalizing the identification results and user evaluation results to form identification and evaluation feature vectors; Step B3, obtaining an image feature vector based on the luggage image; obtaining an evaluation feature vector based on the user evaluation; and obtaining a transaction history feature vector based on the transaction history; Step B4, calculating the cosine similarity between each of the identification and evaluation feature vectors and the image feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the evaluation feature vector; calculating the cosine similarity between each of the identification and evaluation feature vectors and the transaction history feature vector; Step B5: Calculate a preset weight based on the calculated cosine similarity.

8. The luggage identification and evaluation system based on AI multimodal large model according to claim 7 is characterized in that: The calculation formula of the preset weights ω1, ω2, and ω3 is: ； ； ； in, ω1 is the weight of the image feature vector; ω2 is the weight of the evaluation feature vector; ω3 is the weight of the transaction history feature vector; S 1i is the cosine similarity between the identification result, user evaluation result and image feature vector in the i-th historical identification data; S 2i is the cosine similarity between the identification result, user evaluation result and evaluation feature vector in the i-th historical identification data; S 3i is the cosine similarity between the identification result and user evaluation result and the transaction history feature vector in the i-th historical identification data; n is the number of items in the historical identification data.

9. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: The multimodal deep learning model adopts the Transformer model.

10. The luggage identification and evaluation system based on AI multimodal large model according to claim 1 is characterized in that: The parameters of the multimodal deep learning model are: model dimension 512, number of attention heads 8, number of layers 6, feedforward network dimension 2048, learning rate 0.0003, batch size 32, training epochs 30, activation function GELU, optimizer AdamW.

Images