Personalized insurance scheme customization method and device, equipment and storage medium

Abstract

The invention discloses a personalized insurance scheme customization method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the steps of obtaining a to-be-processed text and a to-be-processed image; capturing high-resolution details of the image to be processed; capturing low-resolution details of the image to be processed; performing feature fusion on the image high-resolution features and the image low-resolution features to obtain image fusion features; performing text semantic recognition on the to-be-processed text to obtain text semantic features; carrying out feature fusion on the image fusion features and the text semantic features to obtain a multi-modal feature sequence; and importing the multi-modal feature sequence into a pre-trained insurance scheme generation model, and outputting a personalized insurance scheme. The invention further relates to the technical field of block chains, and the to-be-processed data is stored on the block chain node. According to the method, the information complementarity between the image and the text is enhanced through multi-modal feature fusion, and the accuracy and personalized matching degree of insurance scheme generation are effectively improved.

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and specifically relates to a method, device, equipment and storage medium for customizing personalized insurance plans. Background Art

[0002] Personalized insurance plan customization refers to tailoring a unique and comprehensive insurance plan based on the specific needs, risk status, financial status, and personal or family circumstances of the insured. This customization process not only takes into account the basic information of the insured (such as age, gender, occupation, etc.), but also deeply explores their personalized needs and preferences to ensure that the insurance plan can meet the insured's protection needs to the greatest extent. Through technologies such as intelligent analysis, big data, and machine learning, insurance companies can deeply understand the personalized needs of policyholders and provide more accurate and comprehensive insurance products and services.

[0003] At present, the insurance plan customization system based on multimodal large language model is mainly implemented based on CLIP (Contradistinction Language-Image Pre-training) model. CLIP is an artificial intelligence model that can understand text and images at the same time. Through contrastive learning method, it associates the two forms of information, text and image, and realizes cross-modal understanding and matching. However, the visual encoder of CLIP usually processes input images with relatively low resolution, which has limitations for tasks that need to process high-resolution images. For the insurance plan customization system, high-resolution images may contain more detailed information about the insured's risk status, property status or health status, which is crucial for formulating accurate and personalized insurance plans. Therefore, the CLIP model may not be able to fully extract key information when processing such high-resolution images, resulting in the generated insurance plan being insufficiently accurate, the coverage is not comprehensive enough, or does not match the actual needs. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a personalized insurance plan customization method, apparatus, computer equipment and storage medium to solve the technical problems that the insurance plans generated by the existing CLIP model are insufficiently accurate, have incomplete coverage or do not match actual needs.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a personalized insurance plan customization method, which adopts the following technical solutions:

[0006] A method for customizing a personalized insurance plan, comprising:

[0007] Acquiring data to be processed, wherein the data to be processed includes text to be processed and images to be processed for generating an insurance plan;

[0008] Capturing high-resolution details of the image to be processed through a preset first visual encoder to obtain high-resolution features of the image;

[0009] Capturing low-resolution details of the image to be processed through a preset second visual encoder to obtain low-resolution features of the image;

[0010] Perform feature fusion on the high-resolution features and low-resolution features of the image to obtain image fusion features;

[0011] Based on natural language technology, text semantic recognition is performed on the text to be processed to obtain text semantic features;

[0012] Perform feature fusion on image fusion features and text semantic features to obtain a multimodal feature sequence;

[0013] Import the multimodal feature sequence into the pre-trained insurance plan generation model to output personalized insurance plans.

[0014] Furthermore, the first visual encoder includes a plurality of first convolutional layers, and the steps of capturing high-resolution details of the image to be processed through the preset first visual encoder and obtaining high-resolution features of the image specifically include:

[0015] Performing a downsampling operation on the image to be processed to obtain a plurality of first input images, wherein each first input image corresponds to an input of a first convolutional layer;

[0016] Input each first input image into the corresponding first convolution layer, obtain the local convolution features output by each first convolution layer, and obtain a plurality of first image features;

[0017] A plurality of first image features are combined to obtain high-resolution image features.

[0018] Further, the step of performing a downsampling operation on the image to be processed to obtain a plurality of first input images specifically includes:

[0019] Traversing the first visual encoder, identifying the input image specifications of each first convolutional layer;

[0020] Acquire the image to be processed and preprocess the image to be processed;

[0021] Based on the input image specifications of each first convolutional layer, the preprocessed images to be processed are sequentially downsampled to obtain a plurality of first input images, wherein the resolution of each first input image matches the input image specification of a first convolutional layer.

[0022] Furthermore, the second visual encoder includes a plurality of second convolutional layers, and the steps of capturing low-resolution details of the image to be processed through the preset second visual encoder to obtain low-resolution features of the image specifically include:

[0023] Performing an upsampling operation on the image to be processed to obtain a plurality of second input images, wherein each second input image corresponds to an input of a second convolutional layer;

[0024] Input each second input image into the corresponding second convolution layer, obtain the global convolution feature output by each second convolution layer, and obtain a plurality of second image features;

[0025] A plurality of second image features are combined to obtain low-resolution image features.

[0026] Furthermore, the step of performing an upsampling operation on the image to be processed to obtain a plurality of second input images specifically includes:

[0027] Traversing the second visual encoder, identifying the input image specifications of each second convolutional layer;

[0028] Acquire the image to be processed and preprocess the image to be processed;

[0029] Based on the input image specifications of each second convolutional layer, the preprocessed images to be processed are sequentially upsampled to obtain a plurality of second input images, wherein the resolution of each second input image matches the input image specification of a second convolutional layer.

[0030] Furthermore, the step of fusing the high-resolution image features and the low-resolution image features to obtain image fusion features specifically includes:

[0031] Configuring a first initial weight for a high-resolution feature of the image, and configuring a second initial weight for a low-resolution feature of the image;

[0032] Perform an average pooling operation on the high-resolution features of the image to obtain the channel average signal strength of the first visual encoder;

[0033] Perform an average pooling operation on the low-resolution features of the image to obtain the channel average signal strength of the second visual encoder;

[0034] Based on a preset attention mechanism, adjusting the first initial weight according to the channel average signal strength of the first visual encoder, and adjusting the second initial weight according to the channel average signal strength of the second visual encoder;

[0035] Based on the adjusted first initial weight and the adjusted second initial weight, feature fusion is performed on the image high-resolution feature and the image low-resolution feature to obtain an image fusion feature.

[0036] Furthermore, the step of fusing the image fusion features and the text semantic features to obtain a multimodal feature sequence specifically includes:

[0037] Map the image fusion features and text semantic features into the preset feature space;

[0038] Use the preset sliding window to slide gradually in the feature space to obtain several fused feature fragments;

[0039] Several fused feature fragments are combined in sliding order to obtain a multimodal feature sequence.

[0040] In order to solve the above technical problems, the embodiment of the present application further provides a personalized insurance plan customization device, which adopts the following technical solution:

[0041] A personalized insurance plan customization device, comprising:

[0042] A module for acquiring data to be processed, used for acquiring data to be processed, wherein the data to be processed includes text to be processed and images to be processed for generating an insurance plan;

[0043] A high-resolution processing module, used to capture high-resolution details of the image to be processed through a preset first visual encoder to obtain high-resolution features of the image;

[0044] A low-resolution processing module, used to capture low-resolution details of the image to be processed through a preset second visual encoder to obtain low-resolution features of the image;

[0045] An image feature fusion module is used to fuse the high-resolution image features and the low-resolution image features to obtain image fusion features;

[0046] A text semantic recognition module, which is used to perform text semantic recognition on the text to be processed based on natural language technology to obtain text semantic features;

[0047] A multimodal feature fusion module is used to fuse image fusion features and text semantic features to obtain a multimodal feature sequence;

[0048] The insurance plan generation module is used to import multimodal feature sequences into the pre-trained insurance plan generation model and output personalized insurance plans.

[0049] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0050] A computer device comprises a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, the steps of the personalized insurance plan customization method as described in any one of the above items are implemented.

[0051] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0052] A computer-readable storage medium having computer-readable instructions stored thereon, wherein the computer-readable instructions, when executed by a processor, implement the steps of any one of the above-described methods for customizing a personalized insurance plan.

[0053] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0054] The present application discloses a personalized insurance plan customization method, device, equipment and storage medium, which belongs to the field of artificial intelligence technology. The present application obtains the text to be processed and the image to be processed; captures the high-resolution details of the image to be processed to obtain the high-resolution features of the image; captures the low-resolution details of the image to be processed to obtain the low-resolution features of the image; performs feature fusion of the high-resolution features of the image and the low-resolution features of the image to obtain image fusion features; performs text semantic recognition on the text to be processed to obtain text semantic features; performs feature fusion of the image fusion features and the text semantic features to obtain a multimodal feature sequence; imports the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan. The present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further strengthens the information complementarity between image and text through multimodal feature fusion, effectively improving the accuracy of insurance plan generation and the degree of personalized matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0056] Figure 1 An exemplary system architecture diagram is shown in which the present application can be applied;

[0057] Figure 2 A flowchart showing an embodiment of a method for customizing a personalized insurance plan according to the present application is shown;

[0058] Figure 3A schematic structural diagram of an embodiment of a device for customizing a personalized insurance plan according to the present application is shown;

[0059] Figure 4 A schematic structural diagram of an embodiment of a computer device according to the present application is shown. DETAILED DESCRIPTION

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0061] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0062] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0063] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0064] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0065] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop computer 1011, the tablet computer 1012 or the mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer and a desktop computer, etc.

[0066] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0067] It should be noted that the personalized insurance plan customization method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the personalized insurance plan customization device is generally set in the server / terminal device.

[0068] It should be understood that Figure 1 The numbers of terminal devices, networks and servers in the embodiment are merely illustrative. The above system may have any number of terminal devices, networks and servers according to implementation requirements.

[0069] Continue to refer Figure 2 , shows a flow chart of an embodiment of a method for customizing a personalized insurance plan according to the present application. The personalized insurance plan customization method comprises the following steps:

[0070] S201, obtaining data to be processed, wherein the data to be processed includes text to be processed and images to be processed for generating an insurance plan.

[0071] Specifically, the process of obtaining the data to be processed involves obtaining multimodal data related to the generation of insurance plans from multiple data sources. The multimodal data includes text data and image data. The text data includes the user's basic information, demand description, financial status, health records, nature of work and other related document contents, while the image data comes from relevant pictures uploaded by the user, such as identity certificates, medical examination reports, property photos, etc. The obtained data to be processed needs to ensure the authenticity and integrity of the data, and then the system must pre-process the obtained data to be processed, such as deduplication, denoising and formatting of text, denoising and enhancement of images, etc. The acquisition of data to be processed usually relies on natural language processing (NLP) and image recognition technology, and may be combined with user interaction interfaces to achieve real-time data collection.

[0072] S202, capturing high-resolution details of the image to be processed through a preset first visual encoder to obtain high-resolution features of the image.

[0073] Specifically, the high-resolution details of an image mainly refer to the tiny information in the image or the fine features of a smaller area. These details can help the model identify the key content in the image more accurately. At this stage, the system will perform high-resolution processing on the image to be processed through the preset first visual encoder. The first visual encoder can be an image encoder built based on the U-Net network architecture, or it can be a trained deep learning model, such as a convolutional neural network (CNN). Its purpose is to identify fine features in the image, such as tissue details in medical images, minor damage in asset photos, and fine information in ID card photos. By capturing these high-resolution details, the system can obtain high-resolution feature vectors of the image. These vectors contain a detailed description of the image at the microscopic level. Some details in the high-resolution features of the image directly affect the assessment of insurance risks, such as subtle signs of a certain disease, the degree of asset damage, etc.

[0074] S203, capturing low-resolution details of the image to be processed through a preset second visual encoder to obtain low-resolution features of the image.

[0075] Specifically, low-resolution details usually involve the overall shape, layout, and large-scale information of the image, which is very important for understanding the overall structure and background environment of the image. At this stage, the system captures the low-resolution features of the image through the second visual encoder. The second visual encoder is usually built using the U-Net network architecture, or implemented using a trained deep learning model, such as a convolutional neural network (CNN), to capture the global features of the image. For example, in the scenario of real estate insurance, low-resolution features can be used to identify the overall layout of the house, the environmental location, etc., while in health insurance, the low-resolution features of medical images may reflect the approximate shape of the organ or the extent of the lesion. These low-resolution features provide a macro perspective for the generation of insurance plans and help the system perform global risk assessments, such as identifying specific external environmental risks or assessing overall health status.

[0076] S204, performing feature fusion on the high-resolution image features and the low-resolution image features to obtain image fusion features.

[0077] Specifically, high-resolution features and low-resolution features represent the microscopic information and macroscopic information of the image, respectively. In order to understand the image content more comprehensively, the system needs to fuse the two. This step uses a multi-layer neural network or a specific fusion algorithm to weightedly combine the two features to obtain the fusion features of the image. This feature fusion process can involve a variety of operations, such as feature splicing, weighted averaging, attention mechanism, etc. Image fusion features not only contain the detailed information of the image, but also take into account the global background and structural information, so that the system can fully understand the image from different levels. In the process of generating personalized insurance plans, image fusion features can effectively improve the accuracy of risk assessment. For example, in health insurance, fusion features can focus on both the details of tiny lesions and the overall condition of organs; in property insurance, it can take into account both the subtle damage and the overall condition of assets.

[0078] S205, performing text semantic recognition on the text to be processed based on natural language technology to obtain text semantic features.

[0079] Specifically, text semantic recognition is to conduct in-depth analysis of the text data to be processed and extract the core semantic information of the text. Text semantic recognition relies on natural language processing (NLP) technology, especially advanced technologies such as semantic understanding, syntactic analysis, and sentiment analysis. The system processes the text information provided by the user through a trained language model (such as BERT, GPT, etc.) to extract key information and implicit semantic features in the text. Text semantic features can reflect the user's needs, risk tendencies, and background information. For example, in the health insurance scenario, the user's medical history, treatment status, medical report and other text information can be extracted through semantic recognition technology to obtain specific health status, potential disease risks and other information; in the property insurance scenario, the user's property description, purchase record, surrounding environment description and other text information can reveal the risk level and protection needs of the property. Text semantic features provide an important basis for the personalized customization of insurance plans, which reflects the user's personal needs, background and potential risk factors.

[0080] S206, performing feature fusion on the image fusion features and the text semantic features to obtain a multimodal feature sequence.

[0081] Specifically, multimodal feature fusion is the process of unified processing of two different forms of data, image and text. Since images and text have different feature expressions, the system needs to use a specific fusion method to combine image fusion features and text semantic features to form a unified multimodal feature sequence. In the process of multimodal fusion, commonly used technologies include feature interaction mapping, cross-modal attention mechanism, multimodal neural network, etc. The purpose of this feature fusion is to achieve information complementarity and enhancement. For example, in the scenario of health insurance, images can provide specific health status, while text can explain the patient's medical history and treatment. The two complement each other and constitute a complete health risk assessment. In the scenario of property insurance, images can show the actual condition of the house, while text can provide more detailed descriptions and background information, such as the history and environment of the house. By fusing these multimodal features, the system can understand the user's needs and risks more comprehensively and accurately.

[0082] S207, importing the multimodal feature sequence into the pre-trained insurance plan generation model, and outputting a personalized insurance plan.

[0083] Specifically, at this stage, the system inputs the multimodal feature sequences obtained above into the pre-trained insurance plan generation model, and the model generates personalized insurance plans based on these feature sequences. The model can be implemented based on the pre-trained large language model, and is usually composed of multiple submodules, including risk assessment module, plan matching module, personalized recommendation module, etc. In the risk assessment module, the model analyzes the user's health status, property status, etc., and combines multimodal features to predict risks. In the plan matching module, the system will select insurance plans suitable for users from a large number of preset insurance product libraries based on risk prediction results and user needs. Finally, in the personalized recommendation module, the system further optimizes the details of the insurance plan, such as the insured amount, premium, additional terms, etc., based on the user's personalized characteristics (such as age, occupation, income, health, etc.), to ensure that the insurance plan can fully meet the user's needs. The generated personalized insurance plan is not only highly accurate, but also takes into account the user's personal preferences and risk situation, so as to provide the most suitable insurance solution.

[0084] In the above embodiments, the present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further enhances the information complementarity between images and texts through multimodal feature fusion, thereby effectively improving the accuracy and personalized matching degree of insurance plan generation.

[0085] Furthermore, the first visual encoder includes a plurality of first convolutional layers, and the steps of capturing high-resolution details of the image to be processed through the preset first visual encoder and obtaining high-resolution features of the image specifically include:

[0086] Performing a downsampling operation on the image to be processed to obtain a plurality of first input images, wherein each first input image corresponds to an input of a first convolutional layer;

[0087] Input each first input image into the corresponding first convolution layer, obtain the local convolution features output by each first convolution layer, and obtain a plurality of first image features;

[0088] A plurality of first image features are combined to obtain high-resolution image features.

[0089] In this embodiment, the image to be processed is first downsampled to generate a number of first input images of different resolutions. The purpose of downsampling is to reduce the size of the image, thereby reducing the amount of calculation, while retaining the key information of the image. Each downsampled first input image will correspond to a dedicated first convolutional layer input, and the purpose of these convolutional layers is to capture local features of different levels on images of different resolutions. In this way, the model can fully extract high-resolution detail information of each local area in the image.

[0090] Next, the system inputs each downsampled first input image into the first convolution layer of the preset convolutional neural network. The convolution layer scans the local area of ​​the image through a filter and extracts the local convolution features. The core of the convolution operation is to effectively identify details such as edges, textures, and color changes in the image, thereby generating a number of first image features. These local convolution features represent the local detail information of the image at different resolutions. Therefore, through multi-layer convolution operations, the system can better capture the high-resolution details in the image.

[0091] Finally, several first image features are combined to obtain the complete image high-resolution features. The feature combination process usually integrates features from different convolutional layers through splicing or fusion operations. These combinations will obtain detailed local features, thus forming a comprehensive high-resolution feature vector. This high-resolution feature provides sufficient detail information for subsequent processing steps, and is particularly suitable for tasks that require accurate identification of image details, such as damage assessment in insurance claims or lesion identification in health testing.

[0092] Through the above steps, the system can fully extract high-resolution local features in the image, so that the detailed information of the image can be captured more comprehensively. This solution processes the image in layers through multi-layer convolution operations and enhances the representation ability of image details through feature combination.

[0093] Further, the step of performing a downsampling operation on the image to be processed to obtain a plurality of first input images specifically includes:

[0094] Traversing the first visual encoder, identifying the input image specifications of each first convolutional layer;

[0095] Acquire the image to be processed and preprocess the image to be processed;

[0096] Based on the input image specifications of each first convolutional layer, the preprocessed images to be processed are sequentially downsampled to obtain a plurality of first input images, wherein the resolution of each first input image matches the input image specification of a first convolutional layer.

[0097] In this embodiment, the input image specification of each first convolutional layer is first identified by traversing the first visual encoder. The input specification of each convolutional layer determines the image resolution that it can process, and different convolutional layers may be good at capturing image features at different levels, so it is necessary to ensure the matching of image input specifications during the processing.

[0098] Next, the image to be processed is obtained and preprocessed. The preprocessing usually includes normalizing the image, adjusting the image size, and processing the color channels to adapt it to the requirements of the model.

[0099] Subsequently, based on the convolutional layer input specifications obtained in the previous step, the preprocessed images are downsampled in sequence. The purpose of downsampling is to gradually reduce the image resolution so that it can adapt to the input requirements of different convolutional layers. Each downsampling operation will generate a new first input image, and the resolution of each input image accurately matches a specific convolutional layer input specification. By downsampling the image in sequence, each convolutional layer can process the appropriate input resolution, thereby maximizing the feature extraction capability of each layer. This not only maintains an efficient image processing process, but also makes full use of the hierarchical feature extraction capabilities of the visual encoder to ensure that image details at all scales are preserved and processed.

[0100] Through the above steps, the system can ensure that each convolutional layer receives an image resolution that matches its input specifications, thereby extracting the best local feature information at different levels, effectively improving the efficiency and accuracy of image feature extraction.

[0101] Furthermore, the second visual encoder includes a plurality of second convolutional layers, and the steps of capturing low-resolution details of the image to be processed through the preset second visual encoder to obtain low-resolution features of the image specifically include:

[0102] Performing an upsampling operation on the image to be processed to obtain a plurality of second input images, wherein each second input image corresponds to an input of a second convolutional layer;

[0103] Input each second input image into the corresponding second convolution layer, obtain the global convolution feature output by each second convolution layer, and obtain a plurality of second image features;

[0104] A plurality of second image features are combined to obtain low-resolution image features.

[0105] In this embodiment, the image to be processed is first upsampled. Upsampling refers to increasing the resolution of the original image to a higher level than the input image to generate several second input images. Each upsampled image will correspond to a specific second convolutional layer input. Through upsampling, the system can better retain the global structural information of the image and increase the resolution of the image, so that the convolutional neural network can extract global features at high resolution without paying too much attention to details. The core goal of the upsampling operation is to amplify the global information of the image while keeping the main content of the image undistorted.

[0106] Next, each upsampled second input image is input into the corresponding second convolutional layer. Each second convolutional layer is responsible for extracting global features of the image, such as large-scale object contours, scene layout, and overall structure. When processing images at high resolution, the convolution operation tends to capture the overall pattern of the image rather than local details. The output of these convolutional layers is called global convolutional features, which represent the overall structure and layout information of the image at high resolution. These global features complement the local features and can provide rich global background information for subsequent feature fusion.

[0107] Finally, several second image features are combined to obtain low-resolution features of the image. Feature combination usually uses weighted or splicing methods to integrate the global features output by different convolutional layers. Through this feature combination, the system can obtain a comprehensive low-resolution feature representation that contains the global structure of the image and a wide range of pattern information.

[0108] Through the above steps, the system can fully extract the global features of the image and ensure the recognition of the overall pattern and structure of the image at high resolution. Through upsampling operations and extraction of global convolution features, the system can capture a wide range of image information while reducing unnecessary detail interference and enhancing the ability to understand the global content of the image.

[0109] Furthermore, the step of performing an upsampling operation on the image to be processed to obtain a plurality of second input images specifically includes:

[0110] Traversing the second visual encoder, identifying the input image specifications of each second convolutional layer;

[0111] Acquire the image to be processed and preprocess the image to be processed;

[0112] Based on the input image specifications of each second convolutional layer, the preprocessed images to be processed are sequentially upsampled to obtain a plurality of second input images, wherein the resolution of each second input image matches the input image specification of a second convolutional layer.

[0113] In this embodiment, the second visual encoder is first traversed to identify the input image specification of each second convolutional layer. The input image specification of the second convolutional layer refers to the image resolution or size that the convolutional layer expects to receive, and each convolutional layer may process images of different sizes.

[0114] Next, the image to be processed is obtained and preprocessed. The preprocessing stage usually includes image resizing, normalization, and adjustment of color channels.

[0115] The preprocessed image is then progressively upsampled based on the input specifications of each second convolutional layer identified earlier. The purpose of upsampling is to increase the image resolution to a level that is suitable for the input requirements of a particular convolutional layer. Through techniques such as interpolation, the image is enlarged to match the input specifications of each convolutional layer. Each upsampling generates a new second input image with a resolution that accurately matches the input requirements of a particular convolutional layer. For example, if a convolutional layer needs to receive a high-resolution image, the system will upsample it to a size that matches it; if a medium resolution is required, it will be adjusted to the corresponding specifications.

[0116] Through the above steps, the system ensures that each convolutional layer receives a high-resolution image that meets its input specifications, thereby extracting global features at different levels. The upsampling operation gradually increases the image resolution, allowing the second visual encoder to more accurately capture the overall structure and pattern of the image.

[0117] Furthermore, the step of fusing the high-resolution image features and the low-resolution image features to obtain image fusion features specifically includes:

[0118] Configuring a first initial weight for a high-resolution feature of the image, and configuring a second initial weight for a low-resolution feature of the image;

[0119] Perform an average pooling operation on the high-resolution features of the image to obtain the channel average signal strength of the first visual encoder;

[0120] Perform an average pooling operation on the low-resolution features of the image to obtain the channel average signal strength of the second visual encoder;

[0121] Based on a preset attention mechanism, adjusting the first initial weight according to the channel average signal strength of the first visual encoder, and adjusting the second initial weight according to the channel average signal strength of the second visual encoder;

[0122] Based on the adjusted first initial weight and the adjusted second initial weight, feature fusion is performed on the image high-resolution feature and the image low-resolution feature to obtain an image fusion feature.

[0123] In this embodiment, initial weights are first configured for the high-resolution features and low-resolution features of the image, respectively. Specifically, the system sets a set of initial weight values ​​for the detail features (high resolution) and global features (low resolution) of the image.

[0124] Next, the system performs average pooling operations on the high-resolution features and low-resolution features of the image respectively. Average pooling is used to map larger features to smaller spatial dimensions while retaining the overall information of the features. Average pooling of high-resolution features can obtain the average signal strength of the feature on different channels of the first visual encoder. This signal strength reflects the contribution of each channel in feature extraction. Similarly, low-resolution features are also subjected to average pooling operations to generate the average signal strength of the channels of the second visual encoder, which is used to describe the contribution of global features. Through the average pooling operation, the system can obtain the intensity distribution of each feature channel.

[0125] Subsequently, based on the preset attention mechanism, the system uses the channel average signal strength generated by the pooling operation to dynamically adjust the initial weights. The core function of the attention mechanism is to adjust the weights according to the importance of the features, so that more meaningful features are given higher weights during the fusion process. In this process, the channel average signal strength of the first visual encoder is used to adjust the initial weights of high-resolution features, while the channel average signal strength of the second visual encoder is used to adjust the initial weights of low-resolution features. For example, when the channel average signal strength of the first visual encoder is higher than the channel average signal strength of the second visual encoder, the first initial weight is adjusted up. The use of the attention mechanism ensures that the system can dynamically identify the key features of the image, and both local details and global information can be reasonably weighted.

[0126] Finally, based on the adjusted weights, the high-resolution features and low-resolution features of the image are fused. The purpose of feature fusion is to unify the features at different resolutions in order to generate a comprehensive feature representation. This fusion not only retains the detailed information of the image, but also integrates the global background and layout information, thereby forming a more complete and comprehensive image feature expression.

[0127] Through the above steps, the system can effectively adjust the weights of features of different resolutions through the attention mechanism, so that important details and global information can receive appropriate attention and processing during the fusion process. This feature fusion method greatly improves the feature representation capability of the system, so that the final image fusion features can fully reflect the key content and global background of the image.

[0128] Furthermore, the step of fusing the image fusion features and the text semantic features to obtain a multimodal feature sequence specifically includes:

[0129] Map the image fusion features and text semantic features into the preset feature space;

[0130] Use the preset sliding window to slide gradually in the feature space to obtain several fused feature fragments;

[0131] Several fused feature fragments are combined in sliding order to obtain a multimodal feature sequence.

[0132] In this embodiment, the image fusion features and text semantic features are first mapped into a preset feature space to uniformly represent features from different modalities (image and text) so that multimodal features can be compared and fused in the same space. Multimodal feature fusion is achieved through a shared embedding space, that is, the image fusion features and text semantic features are projected into the same feature vector space, so that they remain semantically consistent.

[0133] Next, a preset sliding window is used to perform a step-by-step sliding operation in this feature space. The sliding window technology can divide the feature space into smaller continuous segments. In the feature space, the sliding window captures the local relationship between the image fusion feature and the text semantic feature by moving step by step. In each slide, the window captures a part of the feature and saves it as a fusion feature segment.

[0134] Then, these captured fusion feature fragments are combined in a sliding order to form a complete multimodal feature sequence. This process ensures that the multimodal features of images and texts can be processed sequentially and the cross-modal contextual relationship is preserved. Through this sliding window and fragment combination strategy, the system can gradually build a hierarchical feature sequence that not only reflects the detailed information of the image and text, but also preserves the semantic association between the image and text.

[0135] Through the above steps, the system can effectively map image fusion features and text semantic features into the same feature space, and gradually extract and combine feature fragments through sliding windows to construct a multimodal feature sequence, which not only enhances the combination effect of cross-modal information, but also retains the order and contextual relationship between the two, ensuring that the generated feature sequence has rich semantic information.

[0136] In the above embodiments, the present application discloses a personalized insurance plan customization method, which belongs to the field of artificial intelligence technology. The present application obtains the text to be processed and the image to be processed; captures the high-resolution details of the image to be processed to obtain the high-resolution features of the image; captures the low-resolution details of the image to be processed to obtain the low-resolution features of the image; performs feature fusion of the high-resolution features of the image and the low-resolution features of the image to obtain image fusion features; performs text semantic recognition on the text to be processed to obtain text semantic features; performs feature fusion of the image fusion features and the text semantic features to obtain a multimodal feature sequence; imports the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan. The present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further strengthens the information complementarity between image and text through multimodal feature fusion, effectively improving the accuracy of insurance plan generation and the degree of personalized matching.

[0137] In this embodiment, the electronic device (eg, Figure 1 The server shown in the figure can receive instructions or obtain data through a wired connection or a wireless connection. It should be noted that the above wireless connection method may include but is not limited to 3G / 4G connection, Wi-Fi connection, Bluetooth connection, Wi MAX connection, Zigbee connection, UWB (Ultra Wi-Fi Band) connection, and other wireless connection methods currently known or to be developed in the future.

[0138] It should be emphasized that in order to further ensure the privacy and security of the above-mentioned data information to be processed, the above-mentioned data information to be processed can also be stored in a node of a blockchain.

[0139] The blockchain referred to in this application is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0140] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0141] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0142] Those 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 computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0143] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0144] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a device for customizing a personalized insurance plan. Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0145] like Figure 3As shown, the personalized insurance plan customization device 300 described in this embodiment includes:

[0146] The to-be-processed data acquisition module 301 is used to acquire the to-be-processed data, wherein the to-be-processed data includes to-be-processed text and to-be-processed images for generating an insurance plan;

[0147] A high-resolution processing module 302 is used to capture high-resolution details of the image to be processed through a preset first visual encoder to obtain high-resolution features of the image;

[0148] A low-resolution processing module 303 is used to capture low-resolution details of the image to be processed through a preset second visual encoder to obtain low-resolution features of the image;

[0149] An image feature fusion module 304 is used to fuse the image high-resolution features and the image low-resolution features to obtain image fusion features;

[0150] A text semantic recognition module 305 is used to perform text semantic recognition on the text to be processed based on natural language technology to obtain text semantic features;

[0151] A multimodal feature fusion module 306 is used to fuse image fusion features and text semantic features to obtain a multimodal feature sequence;

[0152] The insurance plan generation module 307 is used to import the multimodal feature sequence into the pre-trained insurance plan generation model to output a personalized insurance plan.

[0153] Further, the first visual encoder includes a plurality of first convolutional layers, and the high-resolution processing module 302 is specifically used for:

[0154] A downsampling operation is performed on the image to be processed to obtain several first input images, wherein each first input image corresponds to an input of a first convolutional layer; each first input image is input into the corresponding first convolutional layer, and a local convolution feature of each first convolutional layer output is obtained to obtain several first image features; and several first image features are combined to obtain image high-resolution features.

[0155] Furthermore, the high-resolution processing module 302 is further configured to:

[0156] Traverse the first visual encoder to identify the input image specifications of each first convolutional layer; obtain the image to be processed and preprocess the image to be processed; based on the input image specifications of each first convolutional layer, downsample the preprocessed image to be processed in sequence to obtain a plurality of first input images, wherein the resolution of each first input image matches the input image specification of a first convolutional layer.

[0157] Further, the second visual encoder includes a plurality of second convolutional layers, and the low-resolution processing module 303 is specifically used for:

[0158] An upsampling operation is performed on the image to be processed to obtain a plurality of second input images, wherein each second input image corresponds to an input of a second convolutional layer; each second input image is input into the corresponding second convolutional layer, and a global convolution feature output by each second convolutional layer is obtained to obtain a plurality of second image features; and a plurality of second image features are combined to obtain low-resolution image features.

[0159] Furthermore, the low-resolution processing module 303 is further used for:

[0160] The second visual encoder is traversed to identify the input image specifications of each second convolutional layer; an image to be processed is obtained and preprocessed; based on the input image specifications of each second convolutional layer, the preprocessed image to be processed is sequentially upsampled to obtain a plurality of second input images, wherein the resolution of each second input image matches the input image specification of a second convolutional layer.

[0161] Furthermore, the image feature fusion module 304 is specifically used for:

[0162] A first initial weight is configured for the high-resolution feature of the image, and a second initial weight is configured for the low-resolution feature of the image; an average pooling operation is performed on the high-resolution feature of the image to obtain the channel average signal strength of the first visual encoder; an average pooling operation is performed on the low-resolution feature of the image to obtain the channel average signal strength of the second visual encoder; based on a preset attention mechanism, the first initial weight is adjusted according to the channel average signal strength of the first visual encoder, and the second initial weight is adjusted according to the channel average signal strength of the second visual encoder; based on the adjusted first initial weight and the adjusted second initial weight, feature fusion is performed on the high-resolution feature of the image and the low-resolution feature of the image to obtain an image fusion feature.

[0163] Furthermore, the multimodal feature fusion module 306 is specifically used for:

[0164] Map the image fusion features and text semantic features into the preset feature space;

[0165] Use the preset sliding window to slide gradually in the feature space to obtain several fused feature fragments;

[0166] Several fused feature fragments are combined in sliding order to obtain a multimodal feature sequence.

[0167] In the above embodiment, the present application discloses a personalized insurance plan customization device, which belongs to the field of artificial intelligence technology. The present application obtains the text to be processed and the image to be processed; captures the high-resolution details of the image to be processed to obtain the high-resolution features of the image; captures the low-resolution details of the image to be processed to obtain the low-resolution features of the image; performs feature fusion of the high-resolution features of the image and the low-resolution features of the image to obtain image fusion features; performs text semantic recognition on the text to be processed to obtain text semantic features; performs feature fusion of the image fusion features and the text semantic features to obtain a multimodal feature sequence; imports the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan. The present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further strengthens the information complementarity between image and text through multimodal feature fusion, effectively improving the accuracy of insurance plan generation and the degree of personalized matching.

[0168] To solve the above technical problems, the present application also provides a computer device. Figure 4 , Figure 4 This is a basic structural block diagram of the computer device in this embodiment.

[0169] The computer device 4 includes a memory 41, a processor 42, and a network interface 43 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 4 having a memory 41, a processor 42, and a network interface 43, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to a microprocessor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital processor (DSP), an embedded device, etc.

[0170] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0171] The memory 41 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 41 can be an internal storage unit of the computer device 4, such as a hard disk or memory of the computer device 4. In other embodiments, the memory 41 can also be an external storage device of the computer device 4, such as a plug-in hard disk equipped on the computer device 4, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 41 can also include both the internal storage unit of the computer device 4 and its external storage device. In this embodiment, the memory 41 is generally used to store the operating system and various application software installed on the computer device 4, such as computer-readable instructions of a personalized insurance plan customization method, etc. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or are to be output.

[0172] The processor 42 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device 4. In this embodiment, the processor 42 is used to run computer-readable instructions or process data stored in the memory 41, such as computer-readable instructions for running the personalized insurance solution customization method.

[0173] The network interface 43 may include a wireless network interface or a wired network interface. The network interface 43 is generally used to establish a communication connection between the computer device 4 and other electronic devices.

[0174] In the above embodiment, the present application discloses a computer device, which belongs to the field of artificial intelligence technology. The present application obtains the text to be processed and the image to be processed; captures the high-resolution details of the image to be processed to obtain the high-resolution features of the image; captures the low-resolution details of the image to be processed to obtain the low-resolution features of the image; performs feature fusion of the high-resolution features of the image and the low-resolution features of the image to obtain image fusion features; performs text semantic recognition on the text to be processed to obtain text semantic features; performs feature fusion of the image fusion features and the text semantic features to obtain a multimodal feature sequence; imports the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan. The present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further strengthens the information complementarity between image and text through multimodal feature fusion, effectively improving the accuracy of insurance plan generation and the degree of personalized matching.

[0175] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the personalized insurance plan customization method as described above.

[0176] In the above embodiments, the present application discloses a computer-readable storage medium, which belongs to the field of artificial intelligence technology. The present application obtains the text to be processed and the image to be processed; captures the high-resolution details of the image to be processed to obtain the high-resolution features of the image; captures the low-resolution details of the image to be processed to obtain the low-resolution features of the image; performs feature fusion of the high-resolution features of the image and the low-resolution features of the image to obtain image fusion features; performs text semantic recognition on the text to be processed to obtain text semantic features; performs feature fusion of the image fusion features and the text semantic features to obtain a multimodal feature sequence; imports the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan. The present application not only combines high-resolution and low-resolution image features, but also uses natural language processing technology to extract text semantic information, and further strengthens the information complementarity between image and text through multimodal feature fusion, effectively improving the accuracy of insurance plan generation and the degree of personalized matching.

[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0178] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0179] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A method for customizing a personalized insurance plan, characterized in that: include: Acquiring data to be processed, wherein the data to be processed includes text to be processed and images to be processed for generating an insurance plan; Capturing high-resolution details of the image to be processed by a preset first visual encoder to obtain high-resolution features of the image; Capturing low-resolution details of the image to be processed by a preset second visual encoder to obtain low-resolution features of the image; Performing feature fusion on the high-resolution image features and the low-resolution image features to obtain image fusion features; Performing text semantic recognition on the text to be processed based on natural language technology to obtain text semantic features; Performing feature fusion on the image fusion feature and the text semantic feature to obtain a multimodal feature sequence; The multimodal feature sequence is imported into a pre-trained insurance plan generation model to output a personalized insurance plan.

2. The personalized insurance plan customization method according to claim 1, characterized in that: The first visual encoder includes a plurality of first convolutional layers. The step of capturing high-resolution details of the image to be processed by the preset first visual encoder to obtain high-resolution features of the image specifically includes: Performing a downsampling operation on the image to be processed to obtain a plurality of first input images, wherein each of the first input images corresponds to an input of the first convolutional layer; Inputting each of the first input images into the corresponding first convolutional layer, obtaining local convolutional features output by each of the first convolutional layers, and obtaining a plurality of first image features; A plurality of the first image features are combined to obtain the image high-resolution feature.

3. The personalized insurance plan customization method according to claim 2, characterized in that: The step of performing a downsampling operation on the image to be processed to obtain a plurality of first input images specifically includes: Traversing the first visual encoder to identify input image specifications of each of the first convolutional layers; Acquire the image to be processed, and preprocess the image to be processed; Based on the input image specifications of each of the first convolutional layers, the preprocessed images to be processed are sequentially downsampled to obtain a plurality of first input images, wherein the resolution of each of the first input images matches the input image specifications of one of the first convolutional layers.

4. The personalized insurance plan customization method according to claim 1, characterized in that: The second visual encoder includes a plurality of second convolutional layers. The step of capturing low-resolution details of the image to be processed by the preset second visual encoder to obtain low-resolution features of the image specifically includes: Performing an upsampling operation on the image to be processed to obtain a plurality of second input images, wherein each of the second input images corresponds to an input of the second convolutional layer; Inputting each of the second input images into the corresponding second convolutional layer, obtaining a global convolutional feature output by each of the second convolutional layers, and obtaining a plurality of second image features; A plurality of the second image features are combined to obtain the image low-resolution feature.

5. The personalized insurance plan customization method according to claim 4, characterized in that: The step of performing an upsampling operation on the image to be processed to obtain a plurality of second input images specifically includes: Traversing the second visual encoder to identify input image specifications of each of the second convolutional layers; Acquire the image to be processed, and preprocess the image to be processed; Based on the input image specifications of each of the second convolutional layers, upsampling operations are performed on the preprocessed images to be processed in sequence to obtain a plurality of second input images, wherein the resolution of each of the second input images matches the input image specifications of one of the second convolutional layers.

6. The personalized insurance plan customization method according to claim 1, characterized in that: The step of fusing the high-resolution image features and the low-resolution image features to obtain image fusion features specifically includes: Configuring a first initial weight for the high-resolution image feature, and configuring a second initial weight for the low-resolution image feature; Performing an average pooling operation on the high-resolution features of the image to obtain a channel average signal strength of the first visual encoder; Performing an average pooling operation on the low-resolution features of the image to obtain a channel average signal strength of the second visual encoder; Based on a preset attention mechanism, adjusting the first initial weight according to the channel average signal strength of the first visual encoder, and adjusting the second initial weight according to the channel average signal strength of the second visual encoder; Based on the adjusted first initial weight and the adjusted second initial weight, feature fusion is performed on the image high-resolution feature and the image low-resolution feature to obtain the image fusion feature.

7. The personalized insurance plan customization method according to claim 1, characterized in that: The step of fusing the image fusion feature and the text semantic feature to obtain a multimodal feature sequence specifically includes: Mapping the image fusion features and the text semantic features into a preset feature space; Using a preset sliding window to gradually slide in the feature space, a plurality of fused feature fragments are obtained; A plurality of the fused feature fragments are combined in a sliding order to obtain the multimodal feature sequence.

8. A personalized insurance plan customization device, characterized in that: include: A module for acquiring data to be processed, used for acquiring data to be processed, wherein the data to be processed includes text to be processed and images to be processed for generating an insurance plan; A high-resolution processing module, used to capture high-resolution details of the image to be processed through a preset first visual encoder to obtain high-resolution features of the image; A low-resolution processing module, used to capture low-resolution details of the image to be processed through a preset second visual encoder to obtain low-resolution features of the image; An image feature fusion module, used for fusing the high-resolution image features and the low-resolution image features to obtain image fusion features; A text semantic recognition module, used to perform text semantic recognition on the text to be processed based on natural language technology to obtain text semantic features; A multimodal feature fusion module, used for fusing the image fusion feature and the text semantic feature to obtain a multimodal feature sequence; The insurance plan generation module is used to import the multimodal feature sequence into a pre-trained insurance plan generation model to output a personalized insurance plan.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the personalized insurance plan customization method as claimed in any one of claims 1 to 7 when executing the computer-readable instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the personalized insurance plan customization method according to any one of claims 1 to 7.

Images