Fusion method and system of artificial intelligence and database in intelligent image search

Abstract

The invention relates to the technical field of computers, in particular to a fusion method and system of artificial intelligence and a database in intelligent image search, and the method comprises the following steps: constructing an intelligent image search system, processing an input image through an image preprocessing module, carrying out the feature extraction of the preprocessed image, and carrying out the feature extraction of a database; storing the image, the feature vector and the element number in a database, extracting the feature vector of an example image uploaded by a user, performing similarity matching on the feature vector and the image in the database, and visually displaying a search result to the user; the method has the beneficial effects that the image is subjected to feature extraction and classification through the deep learning model, the image content can be more accurately understood, and the correlation of search results is improved. And efficient image retrieval is realized in combination with a database technology, and rapid query of large-scale image data is supported.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a method and system for integrating artificial intelligence and databases in intelligent image search. Background Technology

[0002] Limitations of traditional image search: Traditional image search mainly relies on keyword matching or simple image feature extraction, which makes it difficult to accurately understand the image content, resulting in low accuracy and relevance of search results.

[0003] Applications of Artificial Intelligence in Image Recognition: In recent years, artificial intelligence technology (especially deep learning) has made significant progress in image recognition and classification, enabling the extraction of richer features from images and the understanding of image content.

[0004] The development of database technology: Database technology has a mature technical system in data storage, retrieval and management, but traditional databases have shortcomings in processing unstructured data such as images. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for integrating artificial intelligence and databases in intelligent image search, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for integrating artificial intelligence and databases in intelligent image search, comprising the following steps:

[0007] Construct an intelligent image search system, which includes an image preprocessing module, a feature extraction module, a database storage module, a query processing module, and a result display module;

[0008] The image preprocessing module performs format conversion, size adjustment, noise reduction, and image enhancement on the input image. It supports multiple image format conversions and employs Gaussian filtering, median filtering, and contrast and brightness adjustment image enhancement techniques.

[0009] The feature extraction module employs a pre-trained convolutional neural network model and utilizes fine-tuning techniques to adapt it to a specific image dataset. This process extracts features from the pre-processed images, generating high-dimensional feature vectors that represent the image content and semantic information.

[0010] Using a database storage module, select a relational or non-relational database according to the requirements, design a database structure suitable for storing image feature vectors, optimize database performance using indexing technology, support distributed storage, and store images and their feature vectors and primitives in the database;

[0011] With the help of the query processing module, users can search in multiple ways, including by keywords and image content. For keyword search, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content search, the feature vectors of user-uploaded example images are extracted and matched with images in the database for similarity.

[0012] Finally, the results display module features a user-friendly interface that supports image zooming and rotation, provides search results sorting based on relevance and time, and offers pagination functionality, presenting search results intuitively to users.

[0013] Preferably, the image feature extraction and classification steps include: selecting a suitable deep learning model for image feature extraction, choosing from pre-trained models such as ResNet, VGG, and Inception in convolutional neural networks, and adjusting the learning rate, optimizer, and loss function hyperparameters through fine-tuning techniques to adapt the model to a specific image dataset; feeding the preprocessed input image into the deep learning model, and progressively extracting image features through convolutional layers, pooling layers, and fully connected layers to generate high-dimensional feature vectors; classifying the image using the trained classification model, employing a multi-label classification method to identify objects and scenes in the image, and storing the classification results as image labels in the database to assist in subsequent searching and retrieval.

[0014] Preferably, the database design and optimization steps include: designing the database structure, storing the feature vector of each image as a record, and storing the image's metadata; the database table structure includes fields such as image ID, feature vector, file path, upload time, and category label; optimizing database performance using indexing techniques, employing inverted indexes and hash indexes for the feature vector field; using inverted indexes for content-based retrieval to quickly find similar images, and hash indexes for quickly finding images with specific feature vectors; and supporting distributed storage by using a distributed database system to distribute data across multiple nodes, improving system scalability and reliability to meet the storage needs of large-scale image data.

[0015] Preferably, the intelligent search function includes the following steps: Keyword search: The user inputs keywords, and the system uses word embedding technology in natural language processing to convert the keywords into corresponding image feature vectors. It then performs a similarity search in the database, finds the image most similar to the keyword feature vector, and returns the results sorted by similarity. Image content search: The user uploads a sample image, and the system extracts the feature vector of the sample image. Using cosine similarity and Euclidean distance similarity metrics, it performs similarity matching in the database, calculates the similarity between the image feature vector in the database and the feature vector of the sample image, and returns the image results sorted by similarity. Search result display: The search results are displayed to the user in an intuitive way, supporting image zooming and rotation, providing sorting functions based on relevance and time dimension, and pagination functionality to facilitate users viewing image details and browsing a large number of results.

[0016] Preferably, the system optimization and scalability steps include: performance optimization: using caching technology to cache hot data, reducing the number of database accesses and improving system response speed; optimizing deep learning models and using lightweight models to improve model inference speed; scalability design: the system architecture adopts a modular design to facilitate subsequent function expansion and maintenance, and new feature extraction models or search algorithms can be easily added; and it supports distributed deployment by adding nodes to expand the system's processing and storage capabilities.

[0017] A method and system for integrating artificial intelligence and databases in intelligent image search, comprising:

[0018] The image preprocessing module is used to perform format conversion, size adjustment, denoising and image enhancement on input images of various formats. The denoising uses Gaussian filtering and median filtering algorithms, and the image enhancement uses contrast adjustment and brightness adjustment techniques to improve image quality to ensure the efficiency and accuracy of subsequent processing.

[0019] The feature extraction module uses a pre-trained model in a convolutional neural network to fine-tune the learning rate, optimizer, and loss function hyperparameters to adapt to a specific image dataset. It then extracts features from the pre-processed image to generate high-dimensional feature vectors that represent the image content and semantic information.

[0020] The database storage module selects relational or non-relational databases according to requirements, designs a database structure suitable for storing image feature vectors, stores image feature vectors and image metadata in the database, and uses inverted index and hash index technology to optimize database performance, and supports distributed storage to meet the needs of large-scale image data storage.

[0021] The query processing module supports users to search in multiple ways, including by keywords and image content. For keyword searches, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content searches, the feature vectors of user-uploaded example images are extracted and matched with images in the database for similarity.

[0022] The results display module features a user-friendly interface that supports image zooming and rotation. It also provides search results sorting based on relevance and time, as well as pagination, presenting the search results intuitively to the user.

[0023] Preferably, the feature extraction module also includes an image classification submodule. This submodule uses a trained classification model to classify images, identify objects and scenes in the images, adopts a multi-label classification method, allows an image to belong to multiple categories at the same time, and stores the classification results as image labels in the database to assist in subsequent search and retrieval.

[0024] Preferably, in the database storage module, the database table structure design includes fields such as image ID, feature vector, file path, upload time, and category tag. For the feature vector field, the inverted index is suitable for content-based retrieval to quickly find images similar to the query feature vector, and the hash index is suitable for quickly finding images with specific feature vectors. A distributed database system is used to distribute the data across multiple nodes, improving the system's scalability and reliability.

[0025] Preferably, when performing a keyword search, the query processing module allows the user to input keywords, and the system converts the keywords into feature vectors using word embedding technology. It then performs a similarity search in the database, finding the image most similar to the keyword feature vector and returning the results sorted by similarity. When performing an image content search, the user uploads a sample image, and the system extracts the feature vector from the sample image. Using cosine similarity and Euclidean distance similarity metrics, the system calculates the similarity between the feature vector of the image in the database and the feature vector of the sample image, and returns the image results sorted by similarity.

[0026] Preferably, the system also includes a system optimization and expansion module, which comprises: a performance optimization submodule, which uses caching technology to cache hot data, reduce the number of database accesses, and improve system response speed; optimizes deep learning models, using lightweight models to improve model inference speed; an extensibility design submodule, in which the system architecture adopts a modular design, facilitating subsequent functional expansion and maintenance, and allowing easy addition of new feature extraction models or search algorithms; and supports distributed deployment, expanding the system's processing and storage capabilities by adding nodes.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] This invention proposes a method and system for integrating artificial intelligence and databases in intelligent image search. Through deep learning models, it extracts and classifies image features, enabling a more accurate understanding of image content and improving the relevance of search results. Combined with database technology, it achieves efficient image retrieval, supporting rapid queries of large-scale image data. It allows users to search using keywords, image content, and other methods, meeting the needs of diverse users. The system architecture is rationally designed, easy to expand and maintain, and adaptable to application scenarios of varying scales. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the present invention clear and complete, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only some, not all, embodiments of the present invention, and are merely illustrative of the embodiments of the present invention. They are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Example 1, please refer to Figure 1 This invention provides a technical solution: a method for integrating artificial intelligence and databases in intelligent image search, comprising the following steps:

[0032] Construct an intelligent image search system, which includes an image preprocessing module, a feature extraction module, a database storage module, a query processing module, and a result display module;

[0033] The image preprocessing module performs format conversion, size adjustment, noise reduction, and image enhancement on the input image. It supports multiple image format conversions and employs Gaussian filtering, median filtering, and contrast and brightness adjustment image enhancement techniques.

[0034] The feature extraction module employs a pre-trained convolutional neural network model and utilizes fine-tuning techniques to adapt it to a specific image dataset. This process extracts features from the pre-processed images, generating high-dimensional feature vectors that represent the image content and semantic information. The image feature extraction and classification steps include: selecting a suitable deep learning model for image feature extraction from pre-trained models such as ResNet, VGG, and Inception within the convolutional neural network family; adjusting the learning rate, optimizer, and loss function hyperparameters using fine-tuning techniques to adapt the model to the specific image dataset; feeding the pre-processed input image into the deep learning model, which progressively extracts image features through convolutional layers, pooling layers, and fully connected layers to generate high-dimensional feature vectors; and using the trained classification model to classify the image using a multi-label classification approach to identify objects and scenes within the image. The classification results are then stored as image labels in a database to aid in subsequent searches and retrievals.

[0035] Using a database storage module, select either a relational or non-relational database based on requirements, design a database structure suitable for storing image feature vectors, optimize database performance using indexing techniques, and support distributed storage. Store images, their feature vectors, and metadata in the database. The database design and optimization steps include: designing the database structure, storing each image's feature vector as a record, and simultaneously storing image metadata; the database table structure includes fields for image ID, feature vector, file path, upload time, and category label; optimizing database performance using indexing techniques, employing inverted indexes and hash indexes for the feature vector fields. Inverted indexes are used for content-based retrieval to quickly find similar images, while hash indexes are used to quickly find images with specific feature vectors; supporting distributed storage involves using a distributed database system to distribute data across multiple nodes, improving system scalability and reliability to meet the storage needs of large-scale image data.

[0036] With the help of the query processing module, users can search in multiple ways, including by keywords and image content. For keyword search, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content search, the feature vector of the user-uploaded example image is extracted and matched with images in the database for similarity. The intelligent search function steps include: Keyword search: The user enters keywords, and the system uses word embedding technology in natural language processing to convert the keywords into corresponding image feature vectors. A similarity search is performed in the database to find the image most similar to the keyword feature vector, and the results are returned in order of similarity. Image content search: The user uploads an example image, and the system extracts the feature vector of the example image. Cosine similarity and Euclidean distance similarity measurement methods are used to perform similarity matching in the database. The similarity between the feature vector of the image in the database and the feature vector of the example image is calculated, and the image results are returned in order of similarity. Search result display: The search results are displayed to the user in an intuitive way, supporting image zooming and rotation operations. The search results are sorted by relevance and time dimension, and pagination is provided to facilitate users to view image details and browse a large number of results.

[0037] Finally, the results display module features a user-friendly interface that supports image zooming and rotation, provides search results sorting based on relevance and time, and offers pagination functionality, presenting search results intuitively to users.

[0038] System optimization and scalability steps include: Performance optimization: using caching technology to cache hot data, reducing database access frequency and improving system response speed; optimizing deep learning models, using lightweight models to improve model inference speed; Scalability design: the system architecture adopts a modular design, facilitating subsequent functional expansion and maintenance, and easily adding new feature extraction models or search algorithms; supporting distributed deployment, expanding the system's processing and storage capabilities by adding nodes.

[0039] Example 2, based on Example 1, proposes a method and system for integrating artificial intelligence and databases in intelligent image search, including:

[0040] The image preprocessing module is used to perform format conversion, size adjustment, denoising and image enhancement on input images of various formats. The denoising uses Gaussian filtering and median filtering algorithms, and the image enhancement uses contrast adjustment and brightness adjustment techniques to improve image quality to ensure the efficiency and accuracy of subsequent processing.

[0041] The feature extraction module employs a pre-trained model from a convolutional neural network, using fine-tuning techniques to adjust the learning rate, optimizer, and loss function hyperparameters to adapt to a specific image dataset. It then extracts features from the pre-processed images, generating high-dimensional feature vectors that represent image content and semantic information. The feature extraction module also includes an image classification submodule. This submodule uses a trained classification model to classify images, identifying objects and scenes within them. It employs a multi-label classification method, allowing an image to belong to multiple categories simultaneously, and stores the classification results as image labels in a database to aid in subsequent searches and retrievals.

[0042] The database storage module employs a database structure suitable for storing image feature vectors, selected from both relational and non-relational databases based on requirements. Image feature vectors and image metadata are stored in the database, and inverted indexes and hash indexes are used to optimize database performance. Distributed storage is supported to handle large-scale image data storage needs. Within the database storage module, the database table structure includes fields for image ID, feature vector, file path, upload time, and category tag. For the feature vector field, inverted indexes are suitable for content-based retrieval to quickly find images similar to the queried feature vector, while hash indexes are suitable for quickly finding images with specific feature vectors. A distributed database system is used to distribute data across multiple nodes, improving system scalability and reliability.

[0043] The query processing module supports users to search using keywords and image content. For keyword searches, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content searches, the feature vectors of user-uploaded example images are extracted and matched with images in the database for similarity. When performing a keyword search, the user enters keywords, and the system uses word embedding technology to convert the keywords into feature vectors. A similarity search is performed in the database to find the image most similar to the keyword feature vector and returns the results in order of similarity. When performing an image content search, the user uploads an example image, and the system extracts the feature vector of the example image. Cosine similarity and Euclidean distance similarity measures are used to calculate the similarity between the feature vectors of images in the database and the feature vectors of the example image, and the image results are returned in order of similarity.

[0044] The results display module features a user-friendly interface that supports image zooming and rotation. It also provides search results sorting based on relevance and time, as well as pagination, presenting the search results intuitively to the user.

[0045] It also includes a system optimization and expansion module, which includes: a performance optimization submodule, which uses caching technology to cache hot data, reduce the number of database accesses, and improve system response speed; optimization of deep learning models, using lightweight models to improve model inference speed; an extensibility design submodule, in which the system architecture adopts a modular design to facilitate subsequent function expansion and maintenance, and new feature extraction models or search algorithms can be easily added; and support for distributed deployment, which expands the system's processing and storage capabilities by adding nodes.

[0046] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for integrating artificial intelligence and databases in intelligent image search, characterized in that: Includes the following steps: Construct an intelligent image search system, which includes an image preprocessing module, a feature extraction module, a database storage module, a query processing module, and a result display module; The image preprocessing module performs format conversion, size adjustment, noise reduction, and image enhancement on the input image. It supports multiple image format conversions and employs Gaussian filtering, median filtering noise reduction algorithms, and contrast and brightness adjustment image enhancement techniques. The feature extraction module employs a pre-trained convolutional neural network model and utilizes fine-tuning techniques to adapt it to a specific image dataset. This process extracts features from the pre-processed images, generating high-dimensional feature vectors that represent the image content and semantic information. Using a database storage module, select a relational or non-relational database according to the requirements, design a database structure suitable for storing image feature vectors, optimize database performance using indexing technology, support distributed storage, and store images and their feature vectors and primitives in the database; With the help of the query processing module, users can search in multiple ways, including by keywords and image content. For keyword search, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content search, the feature vectors of user-uploaded example images are extracted and matched with images in the database for similarity. Finally, the results display module features a user-friendly interface that supports image zooming and rotation, provides search results sorting based on relevance and time, and offers pagination functionality, presenting search results intuitively to users.

2. The method for integrating artificial intelligence and database in intelligent image search according to claim 1, characterized in that: The image feature extraction and classification steps include: selecting a suitable deep learning model for image feature extraction, choosing from pre-trained models such as ResNet, VGG, and Inception in convolutional neural networks, and adjusting the learning rate, optimizer, and loss function hyperparameters through fine-tuning techniques to adapt the model to a specific image dataset; feeding the preprocessed input image into the deep learning model, and progressively extracting image features through convolutional layers, pooling layers, and fully connected layers to generate high-dimensional feature vectors; using the trained classification model to classify the image, employing a multi-label classification method to identify objects and scenes in the image, and storing the classification results as image labels in the database to assist in subsequent searches and retrievals.

3. The method for integrating artificial intelligence and databases in intelligent image search according to claim 2, characterized in that: The database design and optimization steps include: designing the database structure, storing the feature vector of each image as a record, and storing the image's metadata; the database table structure includes fields for image ID, feature vector, file path, upload time, and category label; optimizing database performance using indexing techniques, employing inverted indexes and hash indexes for the feature vector fields; using inverted indexes for content-based retrieval to quickly find similar images, and hash indexes for quickly finding images with specific feature vectors; and supporting distributed storage by using a distributed database system to distribute data across multiple nodes, improving system scalability and reliability to meet the storage needs of large-scale image data.

4. The method for integrating artificial intelligence and database in intelligent image search according to claim 3, characterized in that: The intelligent search function includes the following steps: Keyword search: The user inputs keywords, and the system uses word embedding technology in natural language processing to convert the keywords into corresponding image feature vectors. It then performs a similarity search in the database, finding the image most similar to the keyword feature vector and returning the results sorted by similarity. Image content search: The user uploads a sample image, and the system extracts the feature vector of the sample image. Using cosine similarity and Euclidean distance similarity metrics, it performs similarity matching in the database, calculating the similarity between the feature vectors of images in the database and the feature vectors of the sample image, and returning the image results sorted by similarity. Search result display: The search results are displayed to the user in an intuitive way, supporting image zooming and rotation. It provides search results sorting based on relevance and time dimension, as well as pagination, facilitating users to view image details and browse a large number of results.

5. The method for integrating artificial intelligence and database in intelligent image search according to claim 4, characterized in that: The system optimization and scalability steps include: performance optimization: using caching technology to cache hot data, reducing the number of database accesses and improving system response speed; optimizing deep learning models and using lightweight models to improve model inference speed; scalability design: the system architecture adopts a modular design to facilitate subsequent function expansion and maintenance, and new feature extraction models or search algorithms can be easily added; and it supports distributed deployment by adding nodes to expand the system's processing and storage capabilities.

6. A method and system for fusing artificial intelligence and database in intelligent image search according to claim 5, characterized in that: include: The image preprocessing module is used to perform format conversion, size adjustment, denoising and image enhancement on input images of various formats. The denoising uses Gaussian filtering and median filtering algorithms, and the image enhancement uses contrast adjustment and brightness adjustment techniques to improve image quality, so as to ensure the efficiency and accuracy of subsequent processing. The feature extraction module uses a pre-trained model in a convolutional neural network to fine-tune the learning rate, optimizer, and loss function hyperparameters to adapt to a specific image dataset. It then extracts features from the pre-processed image to generate high-dimensional feature vectors that represent the image content and semantic information. The database storage module selects relational or non-relational databases according to requirements, designs a database structure suitable for storing image feature vectors, stores image feature vectors and image metadata in the database, and uses inverted index and hash index technology to optimize database performance, and supports distributed storage to meet the needs of large-scale image data storage. The query processing module supports users to search in multiple ways, including by keywords and image content. For keyword searches, natural language processing technology is used to convert keywords into corresponding image feature vectors. For image content searches, the feature vectors of user-uploaded example images are extracted and matched with images in the database for similarity. The results display module features a user-friendly interface that supports image zooming and rotation. It also provides search results sorting based on relevance and time, as well as pagination, presenting the search results intuitively to the user.

7. The system according to claim 6, characterized in that: The feature extraction module also includes an image classification submodule. This submodule uses a trained classification model to classify images, identify objects and scenes in the images, and adopts a multi-label classification method, allowing an image to belong to multiple categories at the same time. The classification results are stored in the database as image labels to assist in subsequent search and retrieval.

8. The system according to claim 7, characterized in that: In the database storage module, the database table structure design includes fields such as image ID, feature vector, file path, upload time, and category tag. For the feature vector field, the inverted index is suitable for content-based retrieval to quickly find images similar to the query feature vector, while the hash index is suitable for quickly finding images with specific feature vectors. A distributed database system is used to distribute the data across multiple nodes, improving the system's scalability and reliability.

9. A system according to claim 8, characterized in that: When performing a keyword search, the query processing module allows the user to input keywords. The system then uses word embedding technology to convert the keywords into feature vectors, performs a similarity search in the database, finds the image most similar to the keyword feature vector, and returns the results sorted by similarity. When performing an image content search, the user uploads a sample image. The system extracts the feature vector of the sample image, uses cosine similarity and Euclidean distance similarity measurement methods to calculate the similarity between the feature vector of the image in the database and the feature vector of the sample image, and returns the image results sorted by similarity.

10. A system according to claim 9, characterized in that: It also includes a system optimization and expansion module, which includes: a performance optimization submodule, which uses caching technology to cache hot data, reduce the number of database accesses, and improve system response speed; optimization of deep learning models, using lightweight models to improve model inference speed; an extensibility design submodule, in which the system architecture adopts a modular design to facilitate subsequent function expansion and maintenance, and new feature extraction models or search algorithms can be easily added; and support for distributed deployment, which expands the system's processing and storage capabilities by adding nodes.

Images