An image retrieval method and device based on AI big data analysis

Through the image retrieval method based on AI big data analysis, key information in text descriptions are extracted and corresponding images are generated, which solves the problems of incomplete search and poor accuracy in the prior art, and achieves a more comprehensive and accurate image of the technical solution. match.

CN118656510BActive Publication Date: 2024-11-22SICHUAN GAZELLE TENGGUANG TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411139580.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-22
Estimated Expiration
2044-08-20

AI Technical Summary

Technical Problem

The existing image retrieval technology is difficult to match target images similar to technical solutions during searching, resulting in incomplete search and poor accuracy.

Method used

Using an image search method based on AI big data analysis, by obtaining the text description information to be retrieved, key information is extracted and the mapping relationship between image features and key information is established, and several images to be retrieved are generated to match and similar to the images in the large database. calculating the degree to achieve comprehensive and accurate retrieval of technical solutions.

Benefits of technology

The accuracy and comprehensiveness of image retrieval is improved, and the target image similar to the technical solution to be retrieved can be better matched.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118656510B_ABST
    Figure CN118656510B_ABST
Patent Text Reader

Abstract

The present invention is applicable to the field of image retrieval technology, and provides an image retrieval method and device based on AI big data analysis, the method comprising the following steps: extracting key information from text description information based on AI semantic understanding; the key information at least includes key component information, key associated information and key function information; obtaining patent drawings and product drawings in the existing big database, extracting image features of each patent drawing and product drawing, respectively establishing a mapping relationship between image features and key component information, key associated information and key function information, performing model training, and obtaining a preset text image model; based on the preset text image model, generating a number of images to be retrieved according to the key information in the text description information; matching a number of images to be retrieved with images in the big database respectively, and obtaining a number of target images. Compared with the traditional image retrieval method, the method provided by the present invention has the advantages of more comprehensive retrieval results and higher accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of image retrieval technology, and in particular relates to an image retrieval method and device based on AI big data analysis. Background Art

[0002] With the rapid development of information technology, artificial intelligence (AI) has been widely used in various fields. Among them, big data analysis and retrieval through AI can not only improve efficiency but also improve accuracy.

[0003] At present, the existing image retrieval technology is generally based on image recognition and search engine technology, which is implemented through steps such as image feature extraction, similarity calculation and index library matching. However, when searching for images such as drawings of invention or utility model patents, if the search is performed by inputting an image, generally only target images similar in image features can be matched, and it is difficult to match target images similar in technical solutions, so there are problems such as incomplete retrieval and poor accuracy. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide an image retrieval method based on AI big data analysis, aiming to solve the above-mentioned technical problems.

[0005] The embodiment of the present invention is implemented as follows: an image retrieval method based on AI big data analysis, which includes the following steps:

[0006] Get the text description information to be retrieved;

[0007] Extract key information from text description information based on AI semantic understanding; the key information includes at least key component information, key association information and key function information;

[0008] Obtain patent drawings and product drawings from the existing big database, extract image features of each patent drawing and product drawing, respectively establish mapping relationships between image features and key component information, key associated information and key functional information, perform model training, and obtain a preset text image model;

[0009] Based on the preset text-generated image model, several images to be retrieved are generated according to the key information in the text description information;

[0010] Match several images to be retrieved with images in a large database and calculate their similarities to obtain several target images with similarity labels.

[0011] Preferably, the key information also includes key position information and key quantity information.

[0012] Preferably, the image features include component structure features, component association features and component function features.

[0013] Preferably, the steps of obtaining patent drawings and product drawings in an existing large database, extracting image features of each patent drawing and product drawing, respectively establishing mapping relationships between image features and key component information, key associated information and key functional information, performing model training, and obtaining a preset text image model specifically include:

[0014] Obtain patent drawings and product drawings in the existing big database, and extract component structural features, component association features, and component functional features of each patent drawing and product drawing;

[0015] The mapping relationship between component structure features and key component information, the mapping relationship between component association features and key association information, and the mapping relationship between component function features and key function information are established respectively, and model training is performed to obtain a preset text graph model.

[0016] Another object of an embodiment of the present invention is to provide an image retrieval device based on AI big data analysis, comprising:

[0017] A search information acquisition unit, used to acquire text description information to be searched;

[0018] A key information extraction unit, used to extract key information from text description information based on AI semantic understanding; the key information includes at least key component information, key association information and key function information;

[0019] The model building unit is used to obtain patent drawings and product drawings in the existing big database, extract the image features of each patent drawing and product drawing, respectively establish the mapping relationship between the image features and the key component information, key related information and key functional information, perform model training, and obtain the preset text image model;

[0020] A text-generated image unit, used to generate a number of images to be retrieved according to key information in the text description information based on a preset text-generated image model;

[0021] The image matching unit is used to match and calculate the similarity of several images to be retrieved with the images in the large database respectively, and obtain several target images with similarity labels.

[0022] Preferably, the key information also includes key position information and key quantity information.

[0023] Preferably, the image features include component structure features, component association features and component function features.

[0024] Preferably, the model building unit comprises:

[0025] An image feature extraction module is used to obtain patent drawings and product drawings in the existing large database, and extract component structural features, component association features, and component functional features of each patent drawing and product drawing;

[0026] The mapping relationship establishment module is used to establish the mapping relationship between component structure features and key component information, the mapping relationship between component association features and key association information, and the mapping relationship between component function features and key function information, perform model training, and obtain a preset text graph model.

[0027] An image retrieval method based on AI big data analysis provided by an embodiment of the present invention can generate multiple corresponding retrieval images according to the input text description information of the technical solution to be retrieved by adopting a text-based graph model, and then match the multiple corresponding retrieval images with the existing images, so as to retrieve the target image that is relatively similar to the technical solution to be retrieved. Compared with the traditional image retrieval method, it has the advantages of more comprehensive retrieval results and higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flowchart of an image retrieval method based on AI big data analysis provided in an embodiment of the present invention.

[0029] Figure 2 This is a schematic diagram of a specific flow chart of step S300 in the image retrieval method provided in an embodiment of the present invention.

[0030] Figure 3 A schematic diagram of the composition of an image retrieval device based on AI big data analysis provided in an embodiment of the present invention.

[0031] Figure 4 A schematic diagram of the composition of a model building unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] like Figure 1 As shown, in one embodiment of the present invention, an image retrieval method based on AI big data analysis is provided, which includes the following steps:

[0034] S100, obtaining text description information to be retrieved;

[0035] S200, extracting key information from the text description information based on AI semantic understanding; the key information at least includes key component information, key association information and key function information;

[0036] S300, obtaining patent drawings and product drawings in the existing big database, extracting image features of each patent drawing and product drawing, respectively establishing mapping relationships between image features and key component information, key associated information and key functional information, performing model training, and obtaining a preset text image model;

[0037] S400, based on a preset text-generated image model, generating a plurality of images to be retrieved according to key information in the text description information;

[0038] S500 , matching and calculating similarities of a plurality of images to be retrieved with images in a large database respectively, and obtaining a plurality of target images with similarity labels.

[0039] Specifically, in step S100, the text description information to be retrieved includes technical features, which generally refers to the text description content of the technical solution to be retrieved; the text description information is not limited to Chinese descriptions, but can also be English or other foreign language descriptions; for example, when searching for the technical solution "a water cup includes a cup cover and a cup body, the top of the cup body is threadedly connected to the cup cover, a suction nozzle is provided on one side of the top of the cup cover, and a plurality of stirring blades and a driving member for driving the stirring blades to rotate are provided in the cup body" (the purpose is: to retrieve a target image similar to the technical solution from a large database such as existing patent drawings or product drawings), the text description information here refers to the description of the technical solution to be retrieved, that is, the aforementioned "a water cup includes a cup cover and a cup body, the top of the cup body is threadedly connected to the cup cover, a suction nozzle is provided on one side of the top of the cup cover, and a plurality of stirring blades and a driving member for driving the stirring blades to rotate are provided in the cup body".

[0040] In step S200, AI semantic understanding is a mature AI technology in the prior art, which can intelligently understand the semantics of text description information through training; therefore, through AI semantic understanding technology, key feature information in text description information can be extracted and invalid information can be eliminated. In practical applications, key feature information includes key component information, key associated information and key function information; wherein, key component information refers to the name of the component in the technical solution; key associated information refers to the description of the association between the components in the technical solution, such as the installation method, connection method, action relationship, and inclusion relationship; key function information refers to the functional description of the components in the technical solution. For example, in the above-mentioned technical solution to be retrieved, "a water cup includes a cup cover and a cup body, the top of the cup body is threadedly connected to the cup cover, a suction nozzle is provided on one side of the top of the cup cover, and a plurality of stirring blades and a driving member for driving the stirring blade to rotate are provided in the cup body": "water cup", "cup cover", "cup body", "suction nozzle", "stirring blade", "driving member" belong to key component information; "including", "threaded connection", "provided with" belong to key associated information; "used to drive the stirring blade to rotate" belongs to key function information.

[0041] In a preferred embodiment of the present invention, if the number and installation orientation of components are also important technical features, in order to improve the accuracy of retrieval, the above-mentioned key information may also include key orientation information and key quantity information; wherein, the key orientation information refers to the description of the relative positions of components and the installation directions in the technical scheme; the key quantity information refers to the description of the number of settings of the same components in the technical scheme; for example, in the above-mentioned technical scheme, "top", "one side of the top", and "inside" belong to key orientation information; "multiple groups" belong to key quantity information; in addition, the remaining information such as "a kind", "conduct", "and", and "as well as" is regarded as invalid information and can be eliminated to reduce interference.

[0042] In a preferred embodiment of the present invention, corresponding to the above-mentioned key information, the image features include component structural features, component association features and component functional features; wherein, component structural features refer to components in the image; component association features refer to the parts of the image that are related in installation method, connection method, action relationship, inclusion relationship, etc. between components (generally regarded as drawing methods, for example, two components are threadedly connected, and the threaded connection drawing method is adopted, and the corresponding component association feature is the threaded connection drawing method); component functional features refer to the combination of component blocks in the image that can realize certain functions. For example, the combination of components such as "driving member", "rotating shaft" and "stirring blade" in the image and the connection method therebetween can realize the function of "used to drive the stirring blade to rotate", and this combination can be regarded as component functional features.

[0043] like Figure 2As shown, in a preferred embodiment of the present invention, patent drawings and product drawings in an existing large database are obtained, image features of each patent drawing and product drawing are extracted, mapping relationships between image features and key component information, key associated information and key functional information are established respectively, model training is performed, and the steps of obtaining a preset text image model, that is, step S300 specifically include:

[0044] S301, obtaining patent drawings and product drawings in the existing big database, and extracting component structural features, component association features, and component functional features of each patent drawing and product drawing;

[0045] S302, respectively establish mapping relationships between component structural features and key component information, mapping relationships between component association features and key association information, and mapping relationships between component function features and key function information, perform model training, and obtain a preset text graph model.

[0046] Specifically, in step S301, the patent drawings refer to the specification drawings of inventions and utility model patents disclosed at home and abroad; the product drawings refer to the product introduction and description drawings disclosed in the prior art; the component structural features can be extracted by identifying the drawing annotations and local image contours on the patent drawings or product drawings; the component association features and component functional features can be extracted based on the arrangement of the component structural features in the image and the corresponding contents recorded in the specification.

[0047] In practical applications, the model training can adopt the convolutional neural network model training method, but is not limited thereto; specifically, the above mapping relationships actually refer to the corresponding relationship between images and texts, which is generally a many-to-many relationship; for example, the mapping relationship between component structure features and key component information is A i →B i ; A i is a collection of component structural features; B i is the set of key component information; i=1, 2, 3, ..., n; n is the total number of mapping relationships between component structural features and key component information; in addition, the set of component structural features related to the mixing structure diagram is denoted as A 1 , and A 1 B that forms a mapping relationship 1 is a collection of key component information with stirring function; since there are many ways to draw stirring structures and there are many shapes and specifications, the collection A 1 There are many component structural features in the set B. Similarly, there are many ways to name the components with stirring functions. 1 The set B contains information about multiple key components. For example, phrases with the same or similar meanings as “stirring blade”, “stirring paddle”, “stirring element”, “stirring rod”, and “dispersing blade” all belong to set B. 1 .

[0048] In addition, in step S400, according to the above-mentioned preset text graph model, the mapping relationship between image features and key feature information can be known. Based on this, the key feature information extracted from the text description information is used as input, and the corresponding image features can be output through the text graph model; as mentioned above, since the mapping relationship between image features and key feature information is generally a many-to-many relationship, inputting one key feature information can output a set of n related image features; wherein, inputting one key component information b can output a set of n related component structure features {a 1 , a 2 , a 3 , …, a n}; Input a key association information d to output a set of n related component association features {c 1 , c 2 , c 3 , …, c n}; Input a key function information f and output a set of n related component function features {e 1 , e 2 , e 3 ,…,e n}; Then, all elements in the set of all image features outputted are combined and arranged one by one according to the corresponding quantity and orientation (of course, all elements in the set of all image features outputted are also combined and arranged one by one according to the corresponding quantity and orientation according to the key quantity information and the key orientation information), and a number of images to be retrieved can be generated.

[0049] It should be noted that, in step S500, the method for matching the image to be retrieved with the image in the large database can refer to the existing image matching method, which will not be described in detail here; similarly, the method for calculating the similarity between the image to be retrieved and the target image can be calculated using the hash algorithm, cosine similarity, Euclidean distance and other algorithms in the existing technology, which will not be described in detail here.

[0050] like Figure 3 As shown, in another embodiment of the present invention, an image retrieval device based on AI big data analysis is also provided, which includes:

[0051] The search information acquisition unit 10 is used to acquire the text description information to be searched;

[0052] A key information extraction unit 20 is used to extract key information from the text description information based on AI semantic understanding; the key information at least includes key component information, key association information and key function information;

[0053] The model building unit 30 is used to obtain patent drawings and product drawings in the existing big database, extract image features of each patent drawing and product drawing, respectively establish mapping relationships between image features and key component information, key associated information and key functional information, perform model training, and obtain a preset text image model;

[0054] A text-generated image unit 40 is used to generate a plurality of images to be retrieved according to key information in the text description information based on a preset text-generated image model;

[0055] The image matching unit 50 is used to match and calculate similarities between a plurality of images to be retrieved and images in a large database, and obtain a plurality of target images with similarity labels.

[0056] Specifically, the text description information to be retrieved includes technical features, which generally refers to the text description content of the technical solution to be retrieved; the text description information is not limited to Chinese descriptions, and can also be foreign language descriptions such as English.

[0057] In addition, AI semantic understanding is a mature AI technology in the existing technology. Through training, it can intelligently understand the semantics of text description information; therefore, through AI semantic understanding technology, key feature information in text description information can be extracted and invalid information can be eliminated. In practical applications, key feature information includes key component information, key association information and key function information; among them, key component information refers to the name of the component in the technical solution; key association information refers to the description of the association between the components in the technical solution, such as the installation method, connection method, action relationship, and inclusion relationship; key function information refers to the functional description of the components in the technical solution.

[0058] In a preferred embodiment of the present invention, if the number and installation orientation of components are also important technical features, in order to improve the accuracy of retrieval, the above-mentioned key information may also include key orientation information and key quantity information; wherein, the key orientation information refers to the description of the relative positions of components and the installation directions in the technical scheme; the key quantity information refers to the description of the number of settings of the same components in the technical scheme; for example, in the above-mentioned technical scheme, "top", "one side of the top", and "inside" belong to key orientation information; "multiple groups" belong to key quantity information; in addition, the remaining information such as "a kind", "conduct", "and", and "as well as" is regarded as invalid information and can be eliminated to reduce interference.

[0059] In a preferred embodiment of the present invention, corresponding to the above-mentioned key information, the image features include component structural features, component association features and component functional features; wherein, component structural features refer to components in the image; component association features refer to the parts of the image that are related in installation method, connection method, action relationship, inclusion relationship, etc. between components (generally regarded as drawing methods, for example, two components are threadedly connected, and the threaded connection drawing method is adopted, and the corresponding component association feature is the threaded connection drawing method); component functional features refer to the combination of component blocks in the image that can realize certain functions. For example, the combination of components such as "driving member", "rotating shaft" and "stirring blade" in the image and the connection method therebetween can realize the function of "used to drive the stirring blade to rotate", and this combination can be regarded as component functional features.

[0060] like Figure 4 As shown, in a preferred embodiment of the present invention, the model building unit 30 includes:

[0061] An image feature extraction module 31 is used to obtain patent drawings and product drawings in the existing large database, and extract component structural features, component association features, and component functional features of each patent drawing and product drawing;

[0062] The mapping relationship establishment module 32 is used to establish the mapping relationship between component structure features and key component information, the mapping relationship between component association features and key association information, and the mapping relationship between component function features and key function information, perform model training, and obtain a preset text map model.

[0063] Specifically, patent drawings refer to the drawings of the specifications of invention and utility model patents disclosed at home and abroad; product drawings refer to the product introduction and description drawings disclosed in the prior art; component structural features can be extracted by identifying the drawing annotations and local image contours on patent drawings or product drawings; component association features and component functional features can be extracted based on the arrangement of component structural features in the image and their corresponding records in the specification.

[0064] In practical applications, the model training can adopt the convolutional neural network model training method, but is not limited thereto; specifically, the above mapping relationships actually refer to the corresponding relationship between images and texts, which is generally a many-to-many relationship; for example, the mapping relationship between component structure features and key component information is A i →B i ; A i is a collection of component structural features; B i is the set of key component information; i=1, 2, 3, ..., n; n is the total number of mapping relationships between component structural features and key component information; in addition, the set of component structural features related to the mixing structure diagram is denoted as A 1 , and A1 B that forms a mapping relationship 1 is a collection of key component information with stirring function; since there are many ways to draw stirring structures and there are many shapes and specifications, the collection A 1 There are many component structural features in the set B. Similarly, there are many ways to name components with stirring functions. 1 The set B contains information about multiple key components. For example, phrases with the same or similar meanings as “stirring blade”, “stirring paddle”, “stirring element”, “stirring rod”, and “dispersing blade” all belong to set B. 1 .

[0065] In addition, according to the above-mentioned preset text graph model, the mapping relationship between image features and key feature information can be known. Based on this, the key feature information extracted from the text description information is used as input, and the corresponding image features can be output through the text graph model; as mentioned above, since the mapping relationship between image features and key feature information is generally a many-to-many relationship, inputting one key feature information can output a set of n related image features; wherein, inputting one key component information b can output a set of n related component structure features {a 1 , a 2 , a 3 , …, a n}; Input a key association information d to output a set of n related component association features {c 1 , c 2 , c 3 , …, c n}; Input a key function information f and output a set of n related component function features {e 1 , e 2 , e 3 ,…,e n}; Then, all elements in the set of all image features outputted are combined and arranged one by one according to the corresponding quantity and orientation (of course, all elements in the set of all image features outputted are also combined and arranged one by one according to the corresponding quantity and orientation according to the key quantity information and the key orientation information), and a number of images to be retrieved can be generated.

[0066] It should be noted that the above-mentioned units and modules can be implemented in the form of a computer program, which can be run on a computer device. The memory of the computer device can store the computer programs constituting each module so that the processor executes each step of the above-mentioned image retrieval method based on AI big data analysis.

[0067] It should be understood that, although each step in the flow chart of each embodiment of the present invention is shown in sequence according to the indication of the arrow, these steps are not necessarily performed in sequence according to the order indicated by the arrow. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be performed in other orders. Moreover, at least a portion of the steps in each embodiment may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these sub-steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of other steps or sub-steps or stages of other steps.

[0068] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory.

[0069] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. An image retrieval method based on AI big data analysis, characterized in that: The following steps are involved: Obtaining text description information to be retrieved; the text description information to be retrieved includes technical features, which refers to the text description content of the technical solution to be retrieved; Extract key information from text description information based on AI semantic understanding; the key information includes at least key component information, key association information and key function information; Obtain patent drawings and product drawings from the existing big database, extract image features of each patent drawing and product drawing, respectively establish mapping relationships between image features and key component information, key associated information and key functional information, perform model training, and obtain a preset text image model; Based on the preset text-generated image model, several images to be retrieved are generated according to the key information in the text description information; Match and calculate the similarity of several images to be retrieved with the images in the large database, and obtain several target images with similarity labels; The image features include component structure features, component association features and component function features; The steps of obtaining patent drawings and product drawings in the existing big database, extracting image features of each patent drawing and product drawing, respectively establishing mapping relationships between image features and key component information, key associated information and key functional information, performing model training, and obtaining a preset text image model specifically include: Obtain patent drawings and product drawings in the existing big database, extract component structural features, component association features and component functional features of each patent drawing and product drawing; obtain component structural features by identifying the drawing annotations on the patent drawings or product drawings and extracting the local image contour; extract component association features and component functional features based on the arrangement of component structural features in the image and the corresponding contents recorded in the specification; Establish the mapping relationship between component structure features and key component information, the mapping relationship between component association features and key association information, and the mapping relationship between component function features and key function information, perform model training, and obtain a preset text graph model; Key component information refers to the component name in the technical solution; key association information refers to the description of the installation method, connection method, functional relationship, and inclusion relationship between components in the technical solution; key functional information refers to the functional description of the components in the technical solution; component structural features refer to the components in the image; component association features refer to the installation method, connection method, functional relationship, and inclusion relationship between components in the image; component functional features refer to the combination of component blocks in the image that can realize certain functions; The above mapping relationships actually refer to the correspondence between images and texts, which is a many-to-many relationship. The key feature information extracted from the text description information is used as input, and the corresponding image features can be output through the text graph model; since the mapping relationship between image features and key feature information is a many-to-many relationship, inputting a key feature information can output a set of n related image features; among them, inputting a key component information b can output a set of n related component structure features {a1, a2, a3, ..., a n }; Input a key association information d to output a set of n related component association features {c1, c2, c3, ..., c n }; Input a key function information f and output a set of n related component functional features {e1, e2, e3, ..., e n Then, all elements in the set of all output image features are combined and arranged one by one according to the corresponding quantity and orientation to generate a number of images to be retrieved.

2. The image retrieval method based on AI big data analysis according to claim 1, characterized in that: The key information also includes key position information and key quantity information.

3. An image retrieval device based on AI big data analysis, characterized in that: include: A search information acquisition unit is used to acquire text description information to be searched; the text description information to be searched includes technical features, which refers to the text description content of the technical solution to be searched; A key information extraction unit, used to extract key information from text description information based on AI semantic understanding; the key information includes at least key component information, key association information and key function information; The model building unit is used to obtain patent drawings and product drawings in the existing big database, extract the image features of each patent drawing and product drawing, respectively establish the mapping relationship between the image features and the key component information, key related information and key functional information, perform model training, and obtain the preset text image model; A text-generated image unit, used to generate a number of images to be retrieved according to key information in the text description information based on a preset text-generated image model; An image matching unit is used to match and calculate similarities between a number of images to be retrieved and images in a large database, and obtain a number of target images with similarity labels; The image features include component structure features, component association features and component function features; The model building unit comprises: The image feature extraction module is used to obtain patent drawings and product drawings in the existing large database, extract component structural features, component association features and component functional features of each patent drawing and product drawing; obtain component structural features by identifying the drawing annotations on the patent drawings or product drawings and extracting the local image contour; extract component association features and component functional features based on the arrangement of component structural features in the image and the corresponding contents recorded in the specification; A mapping relationship establishment module is used to establish mapping relationships between component structural features and key component information, mapping relationships between component association features and key association information, and mapping relationships between component functional features and key functional information, perform model training, and obtain a preset text graph model; Key component information refers to the component name in the technical solution; key association information refers to the description of the installation method, connection method, functional relationship, and inclusion relationship between components in the technical solution; key functional information refers to the functional description of the components in the technical solution; component structural features refer to the components in the image; component association features refer to the installation method, connection method, functional relationship, and inclusion relationship between components in the image; component functional features refer to the combination of component blocks in the image that can realize certain functions; The above mapping relationships actually refer to the correspondence between images and texts, which is a many-to-many relationship. The key feature information extracted from the text description information is used as input, and the corresponding image features can be output through the text graph model; since the mapping relationship between image features and key feature information is a many-to-many relationship, inputting a key feature information can output a set of n related image features; among them, inputting a key component information b can output a set of n related component structure features {a1, a2, a3, ..., a n }; Input a key association information d to output a set of n related component association features {c1, c2, c3, ..., c n }; Input a key function information f and output a set of n related component functional features {e1, e2, e3, ..., e n Then, all elements in the set of all output image features are combined and arranged one by one according to the corresponding quantity and orientation to generate a number of images to be retrieved.

4. The image retrieval device based on AI big data analysis according to claim 3, characterized in that: The key information also includes key position information and key quantity information.