Intelligent supervision and verification method and system for legal instruments

Abstract

The invention relates to the technical field of legal information processing, and discloses a legal instrument intelligent supervision and verification method, which comprises the following steps: receiving a legal instrument uploaded by a user, the legal instrument comprising text content, image data and audio information; extracting the text content of the legal document through an optical character recognition technology, and generating a corresponding text feature vector; the invention also discloses an intelligent supervision and verification system for the legal instruments, and the system comprises a data storage module which is used for storing the legal instruments uploaded by the user, and the extracted text feature vectors, image feature vectors and audio feature vectors. Dynamic rule optimization is realized through an intelligent self-learning check engine, and manual intervention is reduced; intelligent rule generation improves efficiency and accuracy; a multi-modal feature fusion technology is used for comprehensively analyzing legal documents and providing accurate compliance analysis; a real-time risk assessment mechanism ensures instant feedback, and legal risks are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of legal information processing, and in particular to a method and system for intelligent supervision and verification of legal documents. Background Art

[0002] In the current legal industry, with the development of information technology, the processing and review of legal documents have gradually become digital and intelligent. Traditional legal documents usually involve complex legal clauses and regulations, and their review process requires detailed manual inspection by legal professionals. Although this process is crucial in ensuring legal compliance, it also faces many challenges.

[0003] In recent years, with the frequent updates of laws and regulations and the rapid changes in the legal environment, traditional manual review methods have become increasingly inadequate. Many legal document review systems rely on static rule bases, which are usually manually set by legal experts. Although they are highly professional, they lack flexibility and adaptability. Whenever laws and regulations change, the system often needs to go through a tedious update process, resulting in slow response and easy legal risks.

[0004] In addition, existing technologies have obvious limitations in rule setting. Due to the reliance on personal experience in rule making, different legal experts may have differences in understanding and applying the rules, resulting in inconsistency and lack of systematicity in audit results. This not only affects the efficiency of the audit, but also increases potential legal risks.

[0005] As legal affairs become increasingly complex, the accuracy and timeliness of compliance audits become increasingly important. Existing technologies are usually unable to identify compliance issues in the first place, and audit results often require post-analysis, which may delay the handling of potential risks and increase the possibility of legal liability. For this reason, the field of legal document audits urgently needs an innovative technical solution to adapt to the rapidly changing legal environment and improve the efficiency and accuracy of audits. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a method and system for intelligent supervision and verification of legal documents, which solves the problems of static rules, slow response and high demand for manual intervention in traditional legal document review.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: a legal document intelligent supervision and verification method, comprising the following steps:

[0008] Receiving legal documents uploaded by users, wherein the legal documents include text content, image data and audio information;

[0009] Extracting the text content of the legal document through optical character recognition technology and generating a corresponding text feature vector;

[0010] Applying a convolutional neural network model to extract features from the image data of the legal document to generate an image feature vector;

[0011] Analyze the audio information of the legal document by using Mel frequency cepstral coefficients to generate an audio feature vector;

[0012] The text feature vector, the image feature vector and the audio feature vector are integrated to form a comprehensive feature vector;

[0013] The comprehensive feature vector is matched with a preset legal and regulatory knowledge base to output a compliance verification result.

[0014] Preferably, the text feature vector generated in the step is represented by the TF-IDF algorithm, and the formula is as follows:

[0015]

[0016] Among them, TF(t,d) is the frequency of term t in document d, N is the total number of documents, DF(t) is the number of documents containing term t, and the higher the TF-IDF value, the more important the term is to document d.

[0017] Preferably, the image feature vector generated in the step is output by a convolutional neural network model, the output layer of the model is a fully connected layer, and its calculation formula is:

[0018] y=σ(W·h+b)

[0019] Among them, y is the output result, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

[0020] Preferably, the audio feature vector generated in the step is calculated by Mel-frequency cepstral coefficients, and the formula is as follows:

[0021]

[0022] Among them, MFCC(n) represents the nth MFCC coefficient, which reflects the spectral characteristics of the audio signal, X(k) is the value of the audio signal in the frequency domain, k represents the frequency index, and K is the total number of frequency components.

[0023] A legal document intelligent supervision and verification system, comprising:

[0024] A data storage module, used to store legal documents uploaded by users, extracted text feature vectors, image feature vectors and audio feature vectors;

[0025] Feature extraction module, used to extract text features, image features and audio features of legal documents;

[0026] A feature fusion module, used for fusing the text feature vector, the image feature vector and the audio feature vector to generate a comprehensive feature vector;

[0027] A compliance verification module, used to match the comprehensive feature vector with the legal and regulatory knowledge base and output a verification result;

[0028] The user interface module is used to support users to upload legal documents, view verification results and provide feedback on the preferred ones. The feature extraction module includes:

[0029] Optical character recognition module for converting text in the form of images in legal documents into machine-readable text format;

[0030] Convolutional neural network module, used to extract features of the image part of the legal document;

[0031] Audio analysis module, used to analyze audio information in legal documents.

[0032] Preferably, the optical character recognition module comprises:

[0033] Text preprocessing unit, used to perform noise removal, format standardization and character standardization on uploaded legal documents;

[0034] The OCR recognition unit is used to apply a deep learning algorithm to perform character recognition on the preprocessed text and generate a text feature vector.

[0035] Preferably, the convolutional neural network module includes:

[0036] Convolutional layer unit, used to extract local features of the image through convolution operation;

[0037] Pooling layer units are used to reduce feature dimensions through maximum pooling or average pooling and extract the most important features;

[0038] The fully connected layer unit is used to flatten the features output by the pooling layer and generate the final image feature vector through the activation function.

[0039] Preferably, the compliance verification module includes:

[0040] The regulation matching unit is used to compare the comprehensive feature vector with the legal and regulatory knowledge base to identify potential compliance issues;

[0041] The risk assessment unit is used to assess the risk level of identified compliance issues and output risk reports.

[0042] The present invention provides a method and system for intelligent supervision and verification of legal documents, which has the following beneficial effects:

[0043] 1. The present invention introduces an intelligent self-learning verification engine, which allows the system to continuously optimize the verification rules based on user feedback. Compared with the existing technology, it solves the problem of static rules in traditional legal document review systems, which cannot adapt to the ever-changing legal environment and user needs. Through continuous learning, the system can adaptively update, improve the response speed to new regulations and actual cases, and significantly reduce the need for manual intervention.

[0044] 2. The intelligent rule generation capability of the present invention can automatically generate verification rules adapted to different scenarios based on laws, regulations and past documents, which is in sharp contrast to the prior art method of relying on experts to manually set rules. The latter is not only inefficient, but also prone to inconsistency and incompleteness of rules due to differences in personal experience. The present invention improves the efficiency and accuracy of rule setting, ensuring that the system can quickly adapt to new legal requirements and different types of documents.

[0045] 3. The present invention adopts multimodal feature fusion technology to comprehensively analyze text, image and audio features. Compared with the single feature type processing method in the prior art, it can more comprehensively understand the complex content of legal documents. Multimodal feature fusion improves the processing capabilities of legal documents, enabling the system to provide more comprehensive compliance analysis.

[0046] 4. The real-time risk assessment mechanism of the present invention allows the system to generate a risk level assessment report immediately after identifying compliance issues, which is significantly better than the post-analysis model in the prior art, which often leads to delayed processing of potential risks and increases the possibility of legal liability. Through instant feedback, users can quickly understand compliance issues in documents and take appropriate measures, thereby effectively reducing legal risks and potential losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flowchart of the method of the present invention;

[0048] Figure 2 It is a system flow chart of the present invention;

[0049] Figure 3 It is a block diagram of a feature extraction module of the present invention;

[0050] Figure 4 is a block diagram of the compliance verification module of the present invention;

[0051] Figure 5 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0055] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Please see attached Figure 1 The embodiment of the present invention provides a legal document intelligent supervision and verification method, comprising the following steps:

[0057] Receive legal documents uploaded by users, including text content, image data and audio information;

[0058] Extract the text content of legal documents through optical character recognition technology and generate corresponding text feature vectors;

[0059] Apply the convolutional neural network model to extract features from the image data of legal documents and generate image feature vectors;

[0060] Analyze the audio information of legal documents through Mel frequency cepstral coefficients to generate audio feature vectors;

[0061] The text feature vector, the image feature vector and the audio feature vector are integrated to form a comprehensive feature vector;

[0062] The comprehensive feature vector is matched with the preset legal and regulatory knowledge base to output the compliance verification results.

[0063] The text feature vector generated in the step is represented by the TF-IDF algorithm, and the formula is as follows:

[0064]

[0065] Among them, TF(t,d) is the frequency of term t in document d, N is the total number of documents, DF(t) is the number of documents containing term t, and the higher the TF-IDF value, the more important the term is to document d.

[0066] The image feature vector generated in the step is output by the convolutional neural network model. The output layer of the model is a fully connected layer, and its calculation formula is:

[0067] y=σ(W·h+b)

[0068] Among them, y is the output result, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

[0069] The audio feature vector generated in the step is calculated by the Mel frequency cepstral coefficient, and the formula is as follows:

[0070]

[0071] Among them, MFCC(n) represents the nth MFCC coefficient, which reflects the spectral characteristics of the audio signal, X(k) is the value of the audio signal in the frequency domain, k represents the frequency index, and K is the total number of frequency components.

[0072] Please refer to the attached Figure 2 , a legal document intelligent supervision and verification system, comprising:

[0073] A data storage module, used to store legal documents uploaded by users, extracted text feature vectors, image feature vectors and audio feature vectors;

[0074] Feature extraction module, used to extract text features, image features and audio features of legal documents;

[0075] A feature fusion module is used to fuse text feature vectors, image feature vectors and audio feature vectors to generate a comprehensive feature vector;

[0076] The compliance verification module is used to match the comprehensive feature vector with the legal and regulatory knowledge base and output the verification results;

[0077] The user interface module is used to support users in uploading legal documents, viewing verification results and providing feedback.

[0078] Specifically, in this embodiment, the legal document intelligent supervision and verification system includes the following modules:

[0079] In this embodiment, the data storage module is used to store the legal documents uploaded by the user and related information. This module can store the following contents:

[0080] Legal document raw data: including text, images and audio files.

[0081] Text feature vector: Text information extracted by optical character recognition (OCR) technology. The generated feature vector is represented by the TF-IDF algorithm. The formula is as follows:

[0082]

[0083] Among them, TF(t,d) is the frequency of term t in document d, N is the total number of documents, DF(t) is the number of documents containing term t, and the higher the TF-IDF value, the more important the term is to document d.

[0084] Image feature vector: Features extracted by the Convolutional Neural Network (CNN) module, calculated using the following formula:

[0085] y=σ(W·h+b)

[0086] Among them, y is the output result, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

[0087] Audio feature vector: Audio features generated by Mel-frequency cepstral coefficient (MFCC) analysis, the formula is:

[0088]

[0089] Among them, MFCC(n) represents the nth MFCC coefficient, which reflects the spectral characteristics of the audio signal, X(k) is the value of the audio signal in the frequency domain, k represents the frequency index, and K is the total number of frequency components.

[0090] In this embodiment, the feature extraction module is used to extract text features, image features and audio features of legal documents. This module includes the following submodules:

[0091] Optical character recognition module: responsible for converting the image text in legal documents into editable text, and generating text feature vectors through denoising, standardization and other processes.

[0092] Convolutional neural network module: used to analyze image data in legal documents, extract features such as seals and signatures, extract local features through a hierarchical structure and classify them.

[0093] Audio analysis module: extract features from audio information in legal documents, analyze speech content and emotions, and generate audio feature vectors.

[0094] In this embodiment, the feature fusion module fuses the text feature vector, the image feature vector and the audio feature vector to generate a comprehensive feature vector. This process uses an adaptive weighting mechanism to ensure that the impact of different modal features on the final result is adjusted according to their importance, thereby enhancing the expressiveness of the comprehensive features.

[0095] In this embodiment, the compliance verification module matches the comprehensive feature vector with the legal and regulatory knowledge base and outputs the verification result. This module uses machine learning algorithms such as logistic regression and support vector machines to evaluate compliance, generate legal risk reports, and help users identify potential compliance issues.

[0096] In this embodiment, the user interface module supports users to upload legal documents, view verification results and provide feedback. This module provides a friendly interactive interface, where users can easily upload documents, query verification status and results, and rate and provide feedback on system output, helping the system to continuously optimize.

[0097] Please see attached Figure 3 , the feature extraction module includes:

[0098] Optical character recognition module for converting text in the form of images in legal documents into machine-readable text format;

[0099] Convolutional neural network module, used to extract features of the image part of the legal document;

[0100] Audio analysis module, used to analyze audio information in legal documents.

[0101] Specifically, in the embodiment, the feature extraction module includes the following three submodules:

[0102] In this embodiment, the optical character recognition module is used to convert the image-based text in the legal document into a machine-readable text format. The workflow of this module includes:

[0103] Image preprocessing: De-noise and standardize the uploaded legal document images to improve recognition accuracy.

[0104] Text extraction: Use the OCR algorithm to identify characters in the image and generate corresponding text content. The TF-IDF algorithm used is used to represent the features of the extracted text. The formula is as follows:

[0105]

[0106] Among them, TF(t,d) is the frequency of term t in document d, N is the total number of documents, DF(t) is the number of documents containing term t, and the higher the TF-IDF value, the more important the term is to document d.

[0107] In this embodiment, the convolutional neural network module is used to extract the features of the image part in the legal document. The structure of this module includes:

[0108] Convolutional layer: Extracts local features of the input image through multiple convolution operations, thereby capturing important patterns in the image.

[0109] Pooling layer: downsamples the output of the convolutional layer to reduce the feature dimension and enhance the invariance of the feature. The calculation formula of the pooling process is:

[0110] y=σ(W·h+b)

[0111] Among them, y is the output result, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

[0112] Fully connected layer: Flattens the features after convolution and pooling, and generates the final image feature vector through activation function.

[0113] In this embodiment, the audio analysis module is used to analyze the audio information in the legal document and extract features related to the text content. The main functions of this module include:

[0114] Feature extraction: Extract the Mel frequency cepstral coefficient (MFCC) of the audio signal to reflect the spectral characteristics of the audio signal. The formula is:

[0115]

[0116] Among them, MFCC(n) represents the nth MFCC coefficient, which reflects the spectral characteristics of the audio signal, X(k) is the value of the audio signal in the frequency domain, k represents the frequency index, and K is the total number of frequency components.

[0117] Speech recognition: Converts extracted audio features into text content to provide necessary information for compliance analysis.

[0118] Sentiment analysis: Evaluate emotional information in audio to understand the context and tone of legal documents.

[0119] Please see attached Figure 3 , the optical character recognition module includes:

[0120] Text preprocessing unit, used to perform noise removal, format standardization and character standardization on uploaded legal documents;

[0121] The OCR recognition unit is used to apply a deep learning algorithm to perform character recognition on the preprocessed text and generate a text feature vector.

[0122] Specifically, in this embodiment, the optical character recognition module includes the following two sub-units:

[0123] In this embodiment, the main function of the text preprocessing unit is to perform noise removal, format standardization and character standardization on the uploaded legal documents. The specific steps include:

[0124] Denoising: Use image processing technology to remove background noise in the image to improve the clarity of characters.

[0125] Format standardization: Unify legal documents from different sources and formats into a standard format, such as unifying page size, adjusting alignment, etc. to ensure consistency.

[0126] Character standardization: Standardize the extracted characters, including font conversion, size unification, etc., so that the subsequent OCR recognition algorithm can process them effectively.

[0127] Through the above steps, the text preprocessing unit can significantly improve the accuracy of OCR recognition and ensure the quality of the generated text feature vector.

[0128] In this embodiment, the OCR recognition unit uses a deep learning algorithm to perform character recognition on the preprocessed text to generate a text feature vector. Its workflow includes:

[0129] Application of deep learning algorithm: Convolutional neural network deep learning technology is used to extract features and classify the preprocessed images. The process includes multiple layers of convolution, pooling and fully connected layers to capture the complex features of characters.

[0130] Text feature vector generation: The recognized characters are converted into feature vectors, usually represented using the TF-IDF algorithm to highlight the importance of keywords in the text.

[0131] Through the collaborative work of the text preprocessing unit and the OCR recognition unit, the optical character recognition module can efficiently and accurately convert the image text in legal documents into processable text feature vectors, providing a reliable data foundation for subsequent compliance verification.

[0132] Please refer to the attached Figure 3 , the convolutional neural network module includes:

[0133] Convolutional layer unit, used to extract local features of the image through convolution operation;

[0134] Pooling layer units are used to reduce feature dimensions through maximum pooling or average pooling and extract the most important features;

[0135] The fully connected layer unit is used to flatten the features output by the pooling layer and generate the final image feature vector through the activation function.

[0136] Specifically, in this embodiment, the convolutional neural network module includes the following three sub-units:

[0137] In this embodiment, the main function of the convolutional layer unit is to extract local features of the image through convolution operations. The working process of this unit includes:

[0138] Convolution operation: The input image is convolved with multiple convolution kernels (filters) to extract feature maps at different levels. Each convolution kernel focuses on capturing specific patterns in the image, such as edges, textures, etc.

[0139] Feature Mapping: The result of the convolution operation generates a feature map, the formula is:

[0140] S(i,j)=(I*K)(i,j)=∑ m ∑ n I(m,n)K(im,jn)

[0141] Among them, S(i,j) is the value of the feature map, I is the input image, and K is the convolution kernel.

[0142] In this embodiment, the main function of the pooling layer unit is to reduce the feature dimension through the maximum pooling or average pooling operation and extract the most important features. The specific process includes:

[0143] Max pooling: Select the largest feature value in each pooling window to retain the most significant information. For example, for a 2×2 window, the max pooling operation can be expressed as follows:

[0144] P(i,j)=max{S(i+m,j+n)|0≤m,n<2}

[0145] Among them, P(i,j) is the value of the feature map after pooling, and S is the feature map output by the convolutional layer.

[0146] Average pooling: Calculate the average value of the feature value within the pooling window to maintain the smoothness of the feature. The formula is:

[0147]

[0148] In this embodiment, the function of the fully connected layer unit is to flatten the features output by the pooling layer and generate the final image feature vector through the activation function. The specific steps include:

[0149] Flattening operation: Flatten the output feature map of the pooling layer into a one-dimensional vector so that it can be input into the fully connected layer for further processing.

[0150] Application of activation function: Use activation function in the fully connected layer to perform nonlinear transformation on the flattened feature vector to enhance the expressive power of the model. The formula is:

[0151] y=σ(W·h+b)

[0152] Among them, y is the output feature vector, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

[0153] Through the hierarchical processing of convolutional layer units, pooling layer units and fully connected layer units, the convolutional neural network module can efficiently and accurately extract the key features in legal document images, providing solid data support for compliance verification.

[0154] Please refer to the attached Figure 4 , the compliance verification module includes:

[0155] The regulation matching unit is used to compare the comprehensive feature vector with the legal and regulatory knowledge base to identify potential compliance issues;

[0156] The risk assessment unit is used to assess the risk level of identified compliance issues and output risk reports.

[0157] Specifically, in this embodiment, the compliance verification module includes the following two sub-units:

[0158] In this embodiment, the main function of the regulation matching unit is to compare the comprehensive feature vector with the legal and regulatory knowledge base to identify potential compliance issues. Its workflow includes:

[0159] Feature vector comparison: This unit will match the comprehensive feature vector obtained from the feature extraction module with the provisions stored in the legal and regulatory knowledge base. The comparison algorithm can use cosine similarity, Euclidean distance and other methods to calculate similarity. The formula is:

[0160]

[0161] Among them, A and B are the comprehensive feature vector and regulatory feature vector respectively, and ∥·∥ represents the norm of the vector.

[0162] Compliance judgment: Based on the comparison results, the parts that are inconsistent with laws and regulations are identified and marked as potential compliance issues. This process involves a preset threshold judgment mechanism to ensure the accuracy of identification.

[0163] In this embodiment, the risk assessment unit is used to assess the risk level of the identified compliance issues and output a risk report. The main tasks of this unit include:

[0164] Risk level assessment: Based on the nature and impact of compliance issues, each issue is classified into a risk level, with common levels including low, medium, and high. The assessment algorithm can combine historical data and expert rules for quantitative and qualitative analysis.

[0165] Risk report generation: Based on the assessment results, a risk report containing problem description, risk level and recommended measures is automatically generated. The report can be generated in a templated manner to ensure the integrity and consistency of the information. The report may include the following sample formats:

[0166] Compliance issues: Failure to follow the latest contract law provisions.

[0167] Risk level: High.

[0168] Recommended action: Update contract templates promptly to ensure compliance with the latest regulations.

[0169] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A legal document intelligent supervision and verification method, characterized in that: The following steps are involved: Receiving legal documents uploaded by users, wherein the legal documents include text content, image data and audio information; Extracting the text content of the legal document through optical character recognition technology and generating a corresponding text feature vector; Applying a convolutional neural network model to extract features from the image data of the legal document to generate an image feature vector; Analyze the audio information of the legal document by using Mel frequency cepstral coefficients to generate an audio feature vector; The text feature vector, the image feature vector and the audio feature vector are integrated to form a comprehensive feature vector; The comprehensive feature vector is matched with a preset legal and regulatory knowledge base to output a compliance verification result.

2. According to claim 1, a legal document intelligent supervision and verification method is characterized in that: The text feature vector generated in the step is represented by the TF-IDF algorithm, and the formula is as follows: Among them, TF(t,d) is the frequency of term t in document d, N is the total number of documents, DF(t) is the number of documents containing term t, and the higher the TF-IDF value, the more important the term is to document d.

3. According to claim 1, a legal document intelligent supervision and verification method is characterized in that: The image feature vector generated in the step is output by a convolutional neural network model, the output layer of the model is a fully connected layer, and its calculation formula is: y=σ(W·h+b) Among them, y is the output result, W is the weight matrix, h is the input feature vector, b is the bias term, and σ is the activation function.

4. According to claim 1, a legal document intelligent supervision and verification method is characterized in that: The audio feature vector generated in the step is calculated by Mel frequency cepstral coefficients, and the formula is as follows: Among them, MFCC(n) represents the nth MFCC coefficient, which reflects the spectral characteristics of the audio signal, X(k) is the value of the audio signal in the frequency domain, k represents the frequency index, and K is the total number of frequency components.

5. A legal document intelligent supervision and verification system, characterized in that: include: A data storage module, used to store legal documents uploaded by users, extracted text feature vectors, image feature vectors and audio feature vectors; Feature extraction module, used to extract text features, image features and audio features of legal documents; A feature fusion module, used for fusing the text feature vector, the image feature vector and the audio feature vector to generate a comprehensive feature vector; A compliance verification module, used to match the comprehensive feature vector with the legal and regulatory knowledge base and output a verification result; The user interface module is used to support users in uploading legal documents, viewing verification results and providing feedback.

6. The intelligent supervision and verification system for legal documents according to claim 5 is characterized in that: The feature extraction module comprises: Optical character recognition module for converting text in the form of images in legal documents into machine-readable text format; Convolutional neural network module, used to extract features of the image part of the legal document; Audio analysis module, used to analyze audio information in legal documents.

7. The intelligent supervision and verification system for legal documents according to claim 6 is characterized in that: The optical character recognition module comprises: Text preprocessing unit, used to perform noise removal, format standardization and character standardization on uploaded legal documents; The OCR recognition unit is used to apply a deep learning algorithm to perform character recognition on the preprocessed text and generate a text feature vector.

8. The intelligent supervision and verification system for legal documents according to claim 6 is characterized in that: The convolutional neural network module includes: Convolutional layer unit, used to extract local features of the image through convolution operation; Pooling layer units are used to reduce feature dimensions through maximum pooling or average pooling and extract the most important features; The fully connected layer unit is used to flatten the features output by the pooling layer and generate the final image feature vector through the activation function.

9. The intelligent supervision and verification system for legal documents according to claim 5 is characterized in that: The compliance verification module includes: The regulation matching unit is used to compare the comprehensive feature vector with the legal and regulatory knowledge base to identify potential compliance issues; The risk assessment unit is used to assess the risk level of identified compliance issues and output risk reports.

Images