Electronic contract signature verification system based on built-in intelligent auditing algorithm

Abstract

The invention discloses an electronic contract signature verification system based on a built-in intelligent auditing algorithm, relates to the technical field of artificial intelligence, and solves the problems that an existing electronic contract signature verification system is lack of an effective means to verify the identity of an owner of an electronic contract, an electronic seal cannot be prevented from being used in an unauthorized manner, and the verification efficiency is high. And the content of the electronic contract file cannot be completely ensured not to be tampered after being signed. The electronic contract signature verification system comprises an intelligent identification module, an electronic contract management module, a data storage supervision module, an electronic seal management module, an intelligent contract signature verification module and a verification test module. Identifying a seal signature image of the electronic seal and extracting key contract data of the electronic contract through a multi-modal segmentation model; performing hash operation on the electronic contract through the data storage supervision module to obtain an original hash value; the accuracy of the contract signature verification result is detected and evaluated through the evaluation test model, and the anomaly positioning algorithm is optimized according to the accuracy detection and evaluation result, so that electronic contract signature verification is completed; the authorization security of the electronic seal is improved, and the content of the electronic contract file is prevented from being tampered after being signed.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an electronic contract verification system based on a built-in intelligent audit algorithm. Background Art

[0002] With the advent of the digital age, more and more business activities need to rely on electronic contracts for management and transactions. However, the application of electronic contracts is not efficient because the traditional signing method is often cumbersome and time-consuming.

[0003] When a company purchases raw materials from its supplier, it needs to sign an electronic contract. In order to effectively ensure the validity and legality of the contract between the two parties, the company can use an electronic contract verification system based on a built-in intelligent audit algorithm.

[0004] First, enterprises and suppliers need to register accounts in the electronic contract verification system and upload the contracts to be signed to the system. Then, the system will automatically enable the built-in intelligent review algorithm to review the uploaded electronic contracts and check whether the contracts comply with laws and regulations and the internal compliance requirements of the enterprise.

[0005] If there are any problems with the contract during the review process, the built-in intelligent review algorithm will mark the problem and prompt the enterprise and supplier to make modifications. After the modification is completed, they can upload the contract again and review it again until it passes. Therefore, enterprises and suppliers can ensure that the electronic contracts they sign are in compliance with legal and internal compliance requirements.

[0006] Once the electronic contract has passed the review of the built-in intelligent audit algorithm, the system will automatically call the digital signature algorithm to digitally sign the contract and generate the corresponding certificate. Enterprises and suppliers can use this certificate to ensure that the recipient or any third party cannot tamper with or modify the contract. Moreover, after the contract is signed, the system can also store the signed electronic contract for future use as evidence or reference.

[0007] Patent No. CN2016107905297 discloses an electronic intelligent contract payment method based on blockchain technology, which belongs to the field of electronic contract technology, including a public payment blockchain layer, an intelligent contract platform layer, a contract creation layer and a contract signing layer, which realizes the effective integration of the signing link and the performance link (i.e. payment) of the electronic contract, thereby improving the intelligence and automation of electronic contract-related businesses. The above invention introduces blockchain as the underlying technology to unify the payment process and reduce its complexity, and relies on the improvement of blockchain technology itself in self-maintenance and security compared with other payment methods to improve the overall business availability and user experience of electronic contracts.

[0008] Patent No. CN2021110954529 discloses an electronic signature method based on dual authentication of blockchain and CA certificate. The hosting platform converts the contract documents in batches and sends them to the enterprise. The enterprise signs the converted contract documents, and the hosting platform verifies the signature. After the verification is correct, the hosting platform calls the main contract to create a contract signature. The main contract verifies whether the signature is valid with the signing contract, and records the signature details of the valid signature as the signature object in the signing contract to complete the multi-party signing of the contract. The hosting platform uses the blockchain CA certificate to sign the contract document, and calls the cryptographic machine after authorization by the enterprise, and uses the enterprise CA certificate to complete the contract signing in sequence. The technical solution provided by the above invention can effectively overcome the defect of the prior art that the hosting platform cannot effectively restrict the use of CA certificate signature.

[0009] However, the existing electronic contract verification system lacks effective means to verify the identity of the owner of the electronic contract, cannot prevent the unauthorized use of the seal, and cannot completely protect the content of the contract document from being tampered with after signing. Summary of the invention

[0010] The purpose of the present invention is to provide an electronic contract verification system based on a built-in intelligent audit algorithm, which can identify the seal signature image of an electronic seal and extract the key contract data of the electronic contract through a multimodal segmentation model; perform hash operations on the electronic contract through a data storage supervision module to obtain the original hash value; detect and evaluate the accuracy of the contract verification result through an evaluation test model, and optimize the anomaly positioning algorithm based on the accuracy detection and evaluation results, thereby completing the electronic contract verification; so as to improve the security of electronic seal authorization and at the same time ensure that the content of the electronic contract document is not tampered with after signing.

[0011] The present invention utilizes the following technical solutions: An electronic contract verification system based on a built-in intelligent audit algorithm includes an intelligent identification module, an electronic contract management module, a data storage supervision module, an electronic seal management module, an intelligent contract verification module and a verification test module; wherein, An electronic contract management module, used for creating, updating and deleting electronic contracts, and verifying the owner information of electronic contracts; Owner information includes name, document type, document number, contact information and contact address; Electronic seal management module, used to create, update and delete electronic seals, and verify the caller information of electronic seals; Caller information includes phone number, email, ID number, passport number, business registration number, digital certificate and / or caller name; An intelligent recognition module, used to recognize the seal signature image of the electronic seal and extract key contract data of the electronic contract by using an image segmentation algorithm or a preset multimodal segmentation model; The data storage and supervision module is used to perform hash operations on electronic contracts to obtain the original hash value; and to back up and supervise the original hash value, key contract data, seal signature image, electronic contract, electronic seal and contract document signing record; The smart contract signature verification module is used to use the anomaly location algorithm to double-verify and match the hash value of the electronic contract and the key contract data, and to perform anomaly detection on the contract content and electronic seal of the electronic contract to obtain the contract signature verification result; The verification test module is used to use the preset evaluation test model to detect and evaluate the accuracy of the contract verification results, and optimize the anomaly location algorithm based on the accuracy detection and evaluation results, thereby completing the electronic contract verification.

[0012] Preferably, the electronic contract management module uses a color scanner to scan the paper contract to obtain the electronic contract; the electronic contract management module generates an initial electronic contract based on the owner information and the contract type; the owner negotiates and modifies the initial electronic contract to generate an electronic contract to be signed; the contract types include paid contracts, gratuitous contracts, unilateral contracts, bilateral contracts, formal contracts and informal contracts; the owner signs and seals the electronic contract to be signed to generate an electronic contract; when any owner inquires about the electronic contract, the electronic contract management module verifies the owner information.

[0013] Preferably, the intelligent recognition module first retrieves the electronic contract and the electronic seal from the electronic contract management module and the electronic seal management module respectively; then, the feature information of the electronic contract and the electronic seal is extracted respectively through the feature extraction layer of the multimodal segmentation model to obtain the contract text feature matrix and the image data feature matrix; then, the information extraction branch of the iterative training layer is used to perform several iterative training on the contract text feature matrix to obtain the contract weight feature matrix; at the same time, the image segmentation branch of the iterative training layer is used to perform several iterative training on the image data feature matrix to obtain the seal weight feature matrix; and then the recognition output layer is used to optimize and update the contract weight feature matrix and the seal weight feature matrix, and segment and locate the key contract data and seal signature image.

[0014] Preferably, the feature extraction layer uses a context semantic model to convert the contract content in the electronic contract into a contract word vector table; uses a clustering algorithm to measure and cluster the size of all elements in the contract word vector table and all elements in the electronic seal, and obtains the maximum size and the minimum size of the elements respectively; the feature extraction layer uses a sliding window construction algorithm to construct a maximum sliding window according to the maximum size of the element, and at the same time constructs a minimum sliding window according to the minimum size of the element; constructs an inverted pyramid structure for the maximum sliding window according to the scaling ratio, and constructs a pyramid structure for the minimum sliding window according to the scaling ratio; overlaps the inverted pyramid structure with the pyramid structure to obtain the optimal pyramid structure; the feature extraction layer uses the sliding windows of each level in the optimal pyramid structure to extract feature information from the contract word vector table and the electronic seal, respectively, and then obtains the contract text feature matrix and the image data feature matrix.

[0015] Preferably, the information extraction branch of the iterative training layer first randomly generates a set of initial solutions as candidate contract keyword vectors based on the number of clusters in the contract word vector table and the bag-of-words model; then calculates the fitness of each candidate contract keyword vector, and evaluates the quality of each initial solution based on the objective function and the angle penalty distance; then expands the non-candidate contract keyword vectors in the contract text feature matrix into candidate contract keyword vectors according to a preset expansion ratio, and uses an adaptive adjustment strategy to calculate the convergence factor and step weight of the current number of iterations; then randomly extracts a part from the contract text feature matrix and the candidate contract keyword vector as a temporary contract keyword vector, and uses a genetic evolution mechanism to select, cross and mutate the contract text feature matrix based on the angle penalty distance, and at the same time adjusts the step weight and fitness of the contract text feature matrix. Update; compare each fitness of the contract text feature matrix with the preset fitness respectively: if a certain fitness is less than the preset fitness, the non-candidate contract keyword vector corresponding to this fitness is expanded to the candidate contract keyword vector; if a certain fitness is greater than or equal to the preset fitness, the candidate contract keyword vector with the smallest angle penalty distance difference in the temporary contract keyword vector is replaced with the non-candidate contract keyword vector corresponding to this fitness, and expanded to the candidate contract keyword vector; then use the keyword protection mechanism to maintain the candidate contract keyword vector; finally, judge the number of iterations: if the number of iterations reaches the preset number of iterations, output the step weights corresponding to all candidate contract keyword vectors as the contract weight feature matrix; if the number of iterations does not reach the preset number of iterations, continue iterative training until the preset number of iterations is reached.

[0016] Preferably, the image segmentation branch of the iterative training layer first determines the population number, number of iterations, search time, search speed, spatial upper limit and spatial lower limit of Harris hawks, and randomly generates several groups of parameter sets according to the image data feature matrix, and initializes the number of convolution kernels, convolution kernel size, learning rate and regularization factor of the U-Net model; then, the U-Net model is iteratively trained several times using each group of parameter sets, and the intersection-and-union ratio, equilibrium average and accuracy are calculated; then the fitness of each group of parameter sets is calculated according to the intersection-and-union ratio, equilibrium average and accuracy, and the parameter set is optimized according to the fitness using a chaotic sequence; at the same time, the individual position and search speed of Harris hawks are updated using an elite level strategy; then, the nonlinear escape energy of the prey is calculated and compared with the preset escape energy, and the position is updated using a preset update strategy according to the comparison result; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the current parameter set is output as the seal weight feature matrix; if the number of iterations does not reach the preset number of iterations, the iterative training continues until the preset number of iterations is reached.

[0017] Preferably, the data storage supervision module generates an N-bit random number through a hash operation model using a random number generator, and performs several hash operations with the contract word vector table of the electronic contract using a hash function to obtain a hash sequence; the hash sequence is then randomly divided into three parts to form a first group of hash sequences, a second group of hash sequences, and a third group of hash sequences, and the first group of hash sequences and the second group of hash sequences and the first group of hash sequences and the third group of hash sequences are hashed using a hash function to obtain a first hash sequence and a second hash sequence; the first hash sequence and the second hash sequence are then hashed to obtain an original hash value; finally, the original hash value of each electronic contract is stored and supervised, and key contract data, seal signature images, electronic contracts, electronic seals, and contract document signing records are backed up and supervised.

[0018] Preferably, the smart contract verification module performs a hash operation on the electronic contract again according to the hash operation model in the data storage supervision module to obtain the hash value to be verified; and uses the anomaly positioning algorithm to compare the hash value to be verified with the original hash value for consistency: if the hash value to be verified is consistent with the original hash value, the electronic contract has not been tampered with; if the hash value to be verified is inconsistent with the original hash value, the electronic contract has been tampered with; then the multimodal segmentation model is used to segment and locate the electronic contract to obtain the key contract data to be verified; and the anomaly positioning algorithm is used to verify the key contract data to be verified with the stored key contract data to obtain the verification result: if the verification is consistent, the electronic contract has not been tampered with; if the verification is inconsistent, the electronic contract has been tampered with, and the inconsistent key contract data is marked, and at the same time, the anomaly positioning algorithm is used to perform anomaly detection on the contract content of the electronic contract based on the inconsistent key contract data to obtain the contract verification result.

[0019] Preferably, the evaluation test model includes an evaluation index layer, a construction operation layer and a test analysis layer; the evaluation index layer extracts feature information of the original hash value, key contract data and seal signature image to obtain a feature information matrix; the construction operation layer uses two training network branches to perform several construction trainings on the feature information matrix to obtain a deep construction feature matrix; the first training network branch performs several construction trainings on the feature information matrix, the original hash value and the key contract data to obtain a contract construction feature matrix; the second training network branch performs several construction trainings on the feature information matrix, the original hash value and the seal signature image to obtain a seal construction feature matrix; at the same time, the contract construction feature matrix and the seal construction feature matrix are spliced ​​and fused using shuffled blocks to obtain a deep construction feature matrix; finally, the test analysis layer uses an average pooling layer, a fully connected layer, a random inactivation layer and a binary cross entropy-KL divergence loss function to optimize and update the deep construction feature matrix and the anomaly location algorithm, thereby completing the electronic contract verification.

[0020] The present invention uses a multimodal segmentation model to identify the seal signature image of an electronic seal and extract key contract data of an electronic contract; performs hash operation on the electronic contract through a data storage supervision module to obtain an original hash value; detects and evaluates the accuracy of the contract verification result through an evaluation test model, and optimizes the anomaly positioning algorithm based on the accuracy detection and evaluation result, thereby completing the electronic contract verification; improves the security of electronic seal authorization, and ensures that the content of the electronic contract file is not tampered with after signing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0022] Figure 1 This is the schematic diagram of the electronic contract verification system; Figure 2 Schematic diagram of the multimodal segmentation model. DETAILED DESCRIPTION

[0023] The present invention is described in detail below with reference to the accompanying drawings and embodiments: like Figure 1 As shown, the electronic contract signature verification system based on the built-in intelligent audit algorithm described in the present invention includes an intelligent identification module, an electronic contract management module, a data storage supervision module, an electronic seal management module, an intelligent contract signature verification module and a verification test module; wherein, An electronic contract management module, used for creating, updating and deleting electronic contracts, and verifying the owner information of electronic contracts; Owner information includes name, document type, document number, contact information and contact address; Electronic seal management module, used to create, update and delete electronic seals, and verify the caller information of electronic seals; Caller information includes phone number, email, ID number, passport number, business registration number, digital certificate and / or caller name; An intelligent recognition module, used to recognize the seal signature image of the electronic seal and extract key contract data of the electronic contract by using an image segmentation algorithm or a preset multimodal segmentation model; In this embodiment, key contract data includes owner information, contract subject matter, rights and obligations, liability for breach of contract, and dispute resolution; The data storage and supervision module is used to perform hash operations on electronic contracts to obtain the original hash value; and to back up and supervise the original hash value, key contract data, seal signature image, electronic contract, electronic seal and contract document signing record; The smart contract signature verification module is used to use the anomaly location algorithm to double-verify and match the hash value of the electronic contract and the key contract data, and to perform anomaly detection on the contract content and electronic seal of the electronic contract to obtain the contract signature verification result; The verification test module is used to use the preset evaluation test model to detect and evaluate the accuracy of the contract verification results, and optimize the anomaly location algorithm based on the accuracy detection and evaluation results, thereby completing the electronic contract verification.

[0024] In the present invention, the electronic contract management module uses a color scanner to scan the paper contract to obtain the electronic contract; the electronic contract management module generates an initial electronic contract according to the owner information and the contract type; the owner negotiates and modifies the initial electronic contract to generate an electronic contract to be signed; the contract types include paid contracts, gratuitous contracts, unilateral contracts, bilateral contracts, formal contracts and informal contracts; the owner signs and seals the electronic contract to be signed to generate an electronic contract; when any owner inquires about the electronic contract, the electronic contract management module verifies the owner information; In the present invention, the intelligent recognition module first retrieves the electronic contract and the electronic seal from the electronic contract management module and the electronic seal management module respectively; then, the feature information of the electronic contract and the electronic seal is extracted respectively through the feature extraction layer of the multimodal segmentation model to obtain the contract text feature matrix and the image data feature matrix; then, the information extraction branch of the iterative training layer is used to perform several iterative training on the contract text feature matrix to obtain the contract weight feature matrix; at the same time, the image segmentation branch of the iterative training layer is used to perform several iterative training on the image data feature matrix to obtain the seal weight feature matrix; and then, the recognition output layer is used to optimize and update the contract weight feature matrix and the seal weight feature matrix, and segment and locate the key contract data and the seal signature image; In the present invention, the feature extraction layer uses a context semantic model to convert the contract content in the electronic contract into a contract word vector table; uses a clustering algorithm to measure and cluster the size of all elements in the contract word vector table and all elements in the electronic seal, and obtains the maximum size and the minimum size of the elements respectively; the feature extraction layer uses a sliding window construction algorithm to construct a maximum sliding window according to the maximum size of the elements, and at the same time constructs a minimum sliding window according to the minimum size of the elements; constructs an inverted pyramid structure for the maximum sliding window according to the scaling ratio, and constructs a pyramid structure for the minimum sliding window according to the scaling ratio; overlaps the inverted pyramid structure with the pyramid structure to obtain an optimal pyramid structure; the feature extraction layer uses sliding windows of each level in the optimal pyramid structure to extract feature information from the contract word vector table and the electronic seal, and then obtains a contract text feature matrix and an image data feature matrix; In the present invention, the information extraction branch of the iterative training layer first randomly generates a group of initial solutions as candidate contract keyword vectors based on the number of clusters of the contract word vector table and the bag-of-words model; then the fitness of each candidate contract keyword vector is calculated, and the quality of each initial solution is evaluated based on the objective function and the angle penalty distance; then the non-candidate contract keyword vectors in the contract text feature matrix are expanded into candidate contract keyword vectors according to a preset expansion ratio, and the convergence factor and step weight of the current number of iterations are calculated using an adaptive adjustment strategy; then a part is randomly extracted from the contract text feature matrix and the candidate contract keyword vector as a temporary contract keyword vector, and the contract text feature matrix is ​​selected, crossed and mutated based on the angle penalty distance using a genetic evolution mechanism, and the step weight and fitness of the contract text feature matrix are adjusted. The fitness of the contract text feature matrix is ​​updated; each fitness of the contract text feature matrix is ​​compared with the preset fitness: if a fitness is less than the preset fitness, the non-candidate contract keyword vector corresponding to this fitness is expanded to the candidate contract keyword vector; if a fitness is greater than or equal to the preset fitness, the candidate contract keyword vector with the smallest angle penalty distance difference in the temporary contract keyword vector is replaced with the non-candidate contract keyword vector corresponding to this fitness, and expanded to the candidate contract keyword vector; the keyword protection mechanism is then used to maintain the candidate contract keyword vector; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the step weights corresponding to all candidate contract keyword vectors are output as the contract weight feature matrix; if the number of iterations does not reach the preset number of iterations, iterative training is continued until the preset number of iterations is reached; In the present invention, the image segmentation branch of the iterative training layer first determines the population number, iteration number, search time, search speed, spatial upper limit and spatial lower limit of Harris hawks, and randomly generates several groups of parameter sets according to the image data feature matrix, and initializes the number of convolution kernels, convolution kernel size, learning rate and regularization factor of the U-Net model; then, the U-Net model is iteratively trained several times using each group of parameter sets, and the intersection-and-union ratio, equilibrium average and accuracy are calculated; then, the fitness of each group of parameter sets is calculated according to the intersection-and-union ratio, equilibrium average and accuracy, and the parameter set is optimized according to the fitness using a chaotic sequence; at the same time, the individual position and search speed of Harris hawks are updated using an elite level strategy; then, the nonlinear escape energy of the prey is calculated, and compared with the preset escape energy, and the position is updated using a preset update strategy according to the comparison result; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the current parameter set is output as the seal weight feature matrix; if the number of iterations does not reach the preset number of iterations, the iterative training is continued until the preset number of iterations is reached; In the present invention, the data storage supervision module generates an N-bit random number by using a random number generator through a hash operation model, and performs several hash operations with the contract word vector table of the electronic contract using a hash function to obtain a hash sequence; then the hash sequence is randomly divided into three parts to form a first group of hash sequences, a second group of hash sequences and a third group of hash sequences, and the first group of hash sequences and the third group of hash sequences are respectively hashed using a hash function to obtain a first hash sequence and a second hash sequence; then the first hash sequence and the second hash sequence are hashed to obtain an original hash value; finally, the original hash value of each electronic contract is stored and supervised, and at the same time, key contract data, seal signature images, electronic contracts, electronic seals and contract document signing records are backed up and supervised; In the present invention, the smart contract verification module performs a hash operation on the electronic contract again according to the hash operation model in the data storage supervision module to obtain the hash value to be verified; and uses the abnormality positioning algorithm to compare the hash value to be verified with the original hash value for consistency: if the hash value to be verified is consistent with the original hash value, the electronic contract has not been tampered with; if the hash value to be verified is inconsistent with the original hash value, the electronic contract has been tampered with; then the multimodal segmentation model is used to segment and locate the electronic contract to obtain the key contract data to be verified; and the abnormality positioning algorithm is used to verify the key contract data to be verified with the stored key contract data to obtain the verification result: if the verification is consistent, the electronic contract has not been tampered with; if the verification is inconsistent, the electronic contract has been tampered with, and the inconsistent key contract data is marked, and at the same time, the abnormality positioning algorithm is used to perform abnormality detection on the contract content of the electronic contract according to the inconsistent key contract data to obtain the contract verification result; In this embodiment, the smart contract verification module is deployed on the blockchain, and key contract data is stored on the nodes of the blockchain.

[0025] In this embodiment, the evaluation test model includes an evaluation index layer, a construction operation layer, and a test analysis layer; the evaluation index layer uses three 2×2 convolution layers, four 1×1 point convolution layers, two 5×5 maximum pooling layers, and three normalization layers to extract feature information from the original hash value, key contract data, and seal signature image to obtain a feature information matrix; The construction operation layer uses two training network branches to perform several construction trainings on the feature information matrix to obtain a deep construction feature matrix; The first training network branch uses 2 5×5 convolutional layers, 2 3×3 upsampling layers, 2 3×3 downsampling layers, 4 1×1 point convolutional layers and 2 SoftPlus activation functions to perform several construction trainings on the feature information matrix, the original hash value and the key contract data to obtain the contract construction feature matrix; The second training network branch uses 2 1×1 convolutional layers, 3 5×1 upsampling layers, 3 1×5 downsampling layers, 2 CBS blocks, 4 inverted residual blocks, 3 InceptionV3 blocks, 2 SigMoid activation functions and 2 MISH activation functions to perform several construction trainings on the feature information matrix, the original hash value and the seal signature image to obtain the seal construction feature matrix; the CBS block includes 3 5×5 convolutional layers with a step size of 2, 2 3×3 convolutional layers with a step size of 1, 1 7×7 hole convolutional layer with a step size of 2 and 2 Silu activation functions; at the same time, the contract construction feature matrix and the seal construction feature matrix are spliced ​​and fused using shuffle blocks to obtain the deep construction feature matrix; Finally, the test analysis layer uses the average pooling layer, fully connected layer, random dropout layer and binary cross entropy-KL divergence loss function to optimize and update the deep construction feature matrix and anomaly location algorithm to complete the electronic contract verification.

[0026] Example: The electronic contract verification system is started, and the electronic contract management module uses a color scanner to scan the paper contract to obtain the electronic contract; the electronic contract management module generates an initial electronic contract based on the owner information and contract type; the owner information includes name, certificate type, certificate number, contact information and contact address; the owner negotiates and modifies the initial electronic contract to generate an electronic contract to be signed; the contract types include paid contracts, gratuitous contracts, unilateral contracts, bilateral contracts, formal contracts and informal contracts; the owner signs and seals the electronic contract to be signed to generate an electronic contract; when any owner inquires about the electronic contract, the electronic contract management module verifies the owner information; at the same time, the electronic seal management module creates, updates and deletes the electronic seal based on the caller information, and verifies the electronic seal's telephone number, email, ID number, passport number, corporate registration number, digital certificate and / or caller name; The intelligent recognition module first retrieves the electronic contract and the electronic seal from the electronic contract management module and the electronic seal management module respectively; then, the feature information of the electronic contract and the electronic seal is extracted through the feature extraction layer of the multimodal segmentation model to obtain the contract text feature matrix and the image data feature matrix; then, the information extraction branch of the iterative training layer is used to perform several iterative training on the contract text feature matrix to obtain the contract weight feature matrix; at the same time, the image segmentation branch of the iterative training layer is used to perform several iterative training on the image data feature matrix to obtain the seal weight feature matrix; then, the recognition output layer is used to optimize and update the contract weight feature matrix and the seal weight feature matrix, and segment and locate the key contract data and seal signature image; Among them, the feature extraction layer uses the context semantic model to convert the contract content in the electronic contract into a contract word vector table; the clustering algorithm is used to measure and cluster the size of all elements in the contract word vector table and all elements in the electronic seal, and the maximum size and the minimum size of the elements are obtained respectively; the feature extraction layer uses the sliding window construction algorithm to construct the maximum sliding window according to the maximum size of the element, and at the same time constructs the minimum sliding window according to the minimum size of the element; the inverted pyramid structure is constructed for the maximum sliding window according to the scaling ratio, and the pyramid structure is constructed for the minimum sliding window according to the scaling ratio; the inverted pyramid structure is overlapped with the pyramid structure to obtain the optimal pyramid structure; the feature extraction layer uses the sliding windows of each level in the optimal pyramid structure to extract feature information from the contract word vector table and the electronic seal respectively, and then obtain the contract text feature matrix and the image data feature matrix; The information extraction branch of the iterative training layer first randomly generates a set of initial solutions as candidate contract keyword vectors based on the number of clusters in the contract word vector table and the bag-of-words model; then calculates the fitness of each candidate contract keyword vector, and evaluates the quality of each initial solution based on the objective function and the angle penalty distance; then expands the non-candidate contract keyword vectors in the contract text feature matrix into candidate contract keyword vectors according to the preset expansion ratio, and uses the adaptive adjustment strategy to calculate the convergence factor and step weight of the current number of iterations; then randomly extracts parts from the contract text feature matrix and the candidate contract keyword vector as temporary contract keyword vectors, and uses the genetic evolution mechanism to select, crossover and mutate the contract text feature matrix based on the angle penalty distance, and at the same time adjusts the step weight and fitness of the contract text feature matrix. Update; compare each fitness of the contract text feature matrix with the preset fitness: if a fitness is less than the preset fitness, expand the non-candidate contract keyword vector corresponding to this fitness to the candidate contract keyword vector; if a fitness is greater than or equal to the preset fitness, replace the candidate contract keyword vector with the smallest angle penalty distance difference in the temporary contract keyword vector with the non-candidate contract keyword vector corresponding to this fitness, and expand it to the candidate contract keyword vector; then use the keyword protection mechanism to maintain the candidate contract keyword vector; finally, judge the number of iterations: if the number of iterations reaches the preset number of iterations, output the step weights corresponding to all candidate contract keyword vectors as the contract weight feature matrix; if the number of iterations does not reach the preset number of iterations, continue iterative training until the preset number of iterations is reached; The image segmentation branch of the iterative training layer first determines the population size, number of iterations, search time, search speed, spatial upper limit and spatial lower limit of Harris's hawk, and randomly generates several sets of parameter sets according to the image data feature matrix, and initializes the number of convolution kernels, convolution kernel size, learning rate and regularization factor of the U-Net model; then, the U-Net model is iteratively trained several times using each set of parameter sets, and the intersection-and-union ratio, equilibrium average and accuracy are calculated; then, the fitness of each set of parameter sets is calculated according to the intersection-and-union ratio, equilibrium average and accuracy, and the parameter sets are optimized according to the fitness using the chaotic sequence; at the same time, the individual position and search speed of Harris's hawk are updated using the elite level strategy; then, the nonlinear escape energy of the prey is calculated and compared with the preset escape energy, and the position is updated using the preset update strategy according to the comparison result; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the current parameter set is output as the seal weight feature matrix; if the number of iterations does not reach the preset number of iterations, iterative training continues until the preset number of iterations is reached; The data storage supervision module generates an N-bit random number through a hash operation model using a random number generator, and performs several hash operations with the contract word vector table of the electronic contract using a hash function to obtain a hash sequence; the hash sequence is then randomly divided into three parts to form a first group of hash sequences, a second group of hash sequences, and a third group of hash sequences, and the first group of hash sequences and the first group of hash sequences are hashed with the third group of hash sequences using a hash function to obtain a first hash sequence and a second hash sequence; the first hash sequence is then hashed with the second hash sequence to obtain an original hash value; finally, the original hash value of each electronic contract is stored and supervised, and key contract data, seal signature images, electronic contracts, electronic seals, and contract document signing records are backed up and supervised; The smart contract verification module performs hash operation on the electronic contract again according to the hash operation model in the data storage supervision module to obtain the hash value to be verified; and uses the anomaly positioning algorithm to compare the hash value to be verified with the original hash value for consistency: if the hash value to be verified is consistent with the original hash value, the electronic contract has not been tampered with; if the hash value to be verified is inconsistent with the original hash value, the electronic contract has been tampered with; then the multimodal segmentation model is used to segment and locate the electronic contract to obtain the key contract data to be verified; and the anomaly positioning algorithm is used to verify the key contract data to be verified with the stored key contract data to obtain the verification result: if the verification is consistent, the electronic contract has not been tampered with; if the verification is inconsistent, the electronic contract has been tampered with, and the inconsistent key contract data is marked, and at the same time, the anomaly positioning algorithm is used to perform anomaly detection on the contract content of the electronic contract based on the inconsistent key contract data to obtain the contract verification result; The verification test module uses a preset evaluation test model to detect and evaluate the accuracy of the contract verification results, and optimizes the anomaly location algorithm based on the accuracy detection and evaluation results to complete the electronic contract verification.

Claims

1. An electronic contract verification system based on a built-in intelligent audit algorithm, characterized by: It includes intelligent identification module, electronic contract management module, data storage supervision module, electronic seal management module, intelligent contract signature verification module and verification test module; among them, An electronic contract management module, used for creating, updating and deleting electronic contracts, and verifying the owner information of electronic contracts; Owner information includes name, ID type, ID number, contact information and contact address; The electronic seal management module is used to create, update and delete electronic seals, and verify the caller information of the electronic seal; Caller information includes phone number, email, ID number, passport number, business registration number, digital certificate and / or caller name; An intelligent recognition module, used to recognize the seal signature image of the electronic seal and extract key contract data of the electronic contract by using an image segmentation algorithm or a preset multimodal segmentation model; The data storage and supervision module is used to perform hash operations on electronic contracts to obtain the original hash value; and to back up and supervise the original hash value, key contract data, seal signature image, electronic contract, electronic seal and contract document signing record; The smart contract signature verification module is used to use the anomaly location algorithm to double-verify and match the hash value of the electronic contract and the key contract data, and to perform anomaly detection on the contract content and electronic seal of the electronic contract to obtain the contract signature verification result; The verification test module is used to use the preset evaluation test model to detect and evaluate the accuracy of the contract verification results, and optimize the anomaly location algorithm based on the accuracy detection and evaluation results, thereby completing the electronic contract verification.

2. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 1 is characterized in that: The electronic contract management module uses a color scanner to scan the paper contract to obtain the electronic contract; the electronic contract management module generates an initial electronic contract based on the owner information and the contract type; the owner negotiates and modifies the initial electronic contract to generate an electronic contract to be signed; the contract types include paid contracts, gratuitous contracts, unilateral contracts, bilateral contracts, formal contracts and informal contracts; the owner signs and seals the electronic contract to be signed to generate an electronic contract; when any owner inquires about the electronic contract, the electronic contract management module verifies the owner information.

3. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 1 is characterized in that: The intelligent recognition module first retrieves the electronic contract and the electronic seal from the electronic contract management module and the electronic seal management module respectively; then extracts feature information of the electronic contract and the electronic seal respectively through the feature extraction layer of the multimodal segmentation model to obtain the contract text feature matrix and the image data feature matrix; Then, the information extraction branch of the iterative training layer is used to perform several iterative training on the contract text feature matrix to obtain the contract weight feature matrix; at the same time, the image segmentation branch of the iterative training layer is used to perform several iterative training on the image data feature matrix to obtain the seal weight feature matrix; The recognition output layer is then used to optimize and update the contract weight feature matrix and the seal weight feature matrix, and to segment and locate the key contract data and seal signature image.

4. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 3 is characterized by: The feature extraction layer converts the contract content in the electronic contract into a contract word vector table by using a context semantic model; uses a clustering algorithm to measure and cluster the size of all elements in the contract word vector table and all elements in the electronic seal, and obtains the maximum size and the minimum size of the elements respectively; the feature extraction layer constructs a maximum sliding window according to the maximum size of the elements by using a sliding window construction algorithm, and constructs a minimum sliding window according to the minimum size of the elements; constructs an inverted pyramid structure for the maximum sliding window according to the scaling ratio, and constructs a pyramid structure for the minimum sliding window according to the scaling ratio; overlaps the inverted pyramid structure with the pyramid structure to obtain an optimal pyramid structure; the feature extraction layer uses sliding windows of each level in the optimal pyramid structure to extract feature information of the contract word vector table and the electronic seal respectively, and then obtains a contract text feature matrix and an image data feature matrix.

5. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 3 is characterized by: The information extraction branch of the iterative training layer first randomly generates a set of initial solutions as candidate contract keyword vectors based on the number of clusters in the contract word vector table and the bag-of-words model; then calculates the fitness of each candidate contract keyword vector, and evaluates the quality of each initial solution based on the objective function and the angle penalty distance; then expands the non-candidate contract keyword vectors in the contract text feature matrix into candidate contract keyword vectors according to a preset expansion ratio, and uses an adaptive adjustment strategy to calculate the convergence factor and step weight of the current number of iterations; Then, a part is randomly extracted from the contract text feature matrix and the candidate contract keyword vector as a temporary contract keyword vector, and the contract text feature matrix is ​​selected, crossed and mutated according to the angle penalty distance using the genetic evolution mechanism, and the step weight and fitness of the contract text feature matrix are updated at the same time; each fitness of the contract text feature matrix is ​​compared with the preset fitness: if a certain fitness is less than the preset fitness, the non-candidate contract keyword vector corresponding to this fitness is expanded to the candidate contract keyword vector; if a certain fitness is greater than or equal to the preset fitness, the candidate contract keyword vector with the smallest angle penalty distance difference in the temporary contract keyword vector is replaced with the non-candidate contract keyword vector corresponding to this fitness, and expanded to the candidate contract keyword vector; the candidate contract keyword vector is then maintained using the keyword protection mechanism; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the step weights corresponding to all candidate contract keyword vectors are output as the contract weight feature matrix; If the number of iterations does not reach the preset number of iterations, the iterative training continues until the preset number of iterations is reached.

6. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 3 is characterized by: The image segmentation branch of the iterative training layer first determines the population number, number of iterations, search time, search speed, spatial upper limit and spatial lower limit of Harris hawks, and randomly generates several groups of parameter sets according to the image data feature matrix, and initializes the number of convolution kernels, convolution kernel size, learning rate and regularization factor of the U-Net model; then, the U-Net model is iteratively trained several times using each group of parameter sets, and the intersection-over-union ratio, equilibrium average and accuracy are calculated; Then, the fitness of each parameter set is calculated according to the intersection ratio, equilibrium average and accuracy, and the parameter set is optimized according to the fitness using the chaotic sequence; at the same time, the individual position and search speed of the Harris hawk are updated using the elite level strategy; then the nonlinear escape energy of the prey is calculated and compared with the preset escape energy, and the position is updated using the preset update strategy according to the comparison result; finally, the number of iterations is judged: if the number of iterations reaches the preset number of iterations, the current parameter set is output as the seal weight feature matrix; If the number of iterations does not reach the preset number of iterations, the iterative training continues until the preset number of iterations is reached.

7. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 1 is characterized in that: The data storage supervision module generates an N-bit random number using a random number generator through a hash operation model, and performs several hash operations with the contract word vector table of the electronic contract using a hash function to obtain a hash sequence; then the hash sequence is randomly divided into three parts to form a first group of hash sequences, a second group of hash sequences and a third group of hash sequences, and the first group of hash sequences and the third group of hash sequences are hashed using a hash function to obtain a first hash sequence and a second hash sequence; The first hash sequence is then hashed with the second hash sequence to obtain the original hash value; finally, the original hash value of each electronic contract is stored and supervised, and the key contract data, seal signature image, electronic contract, electronic seal and contract document signing record are backed up and supervised.

8. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 1 is characterized in that: The smart contract verification module performs a hash operation on the electronic contract again according to the hash operation model in the data storage supervision module to obtain the hash value to be verified; and uses the abnormal location algorithm to compare the hash value to be verified with the original hash value for consistency: if the hash value to be verified is consistent with the original hash value, the electronic contract has not been tampered with; if the hash value to be verified is inconsistent with the original hash value, the electronic contract has been tampered with; and then uses the multimodal segmentation model to segment and locate the electronic contract to obtain the key contract data to be verified; The abnormal location algorithm is used to verify the key contract data to be verified with the stored key contract data to obtain the verification result: if the verification is consistent, the electronic contract has not been tampered with; If the verification is inconsistent, the electronic contract has been tampered with, and the inconsistent key contract data will be marked. At the same time, the anomaly positioning algorithm is used to perform anomaly detection on the contract content of the electronic contract based on the inconsistent key contract data to obtain the contract verification result.

9. The electronic contract verification system based on a built-in intelligent audit algorithm according to claim 1 is characterized in that: The evaluation test model includes an evaluation index layer, a construction operation layer and a test analysis layer; the evaluation index layer extracts feature information from the original hash value, key contract data and seal signature image to obtain a feature information matrix; The construction operation layer uses two training network branches to perform several construction trainings on the feature information matrix to obtain a deep construction feature matrix; The first training network branch performs several construction trainings on the feature information matrix, the original hash value, and the key contract data to obtain the contract construction feature matrix; The second training network branch performs several construction trainings on the feature information matrix, the original hash value and the seal signature image to obtain the seal construction feature matrix; at the same time, the contract construction feature matrix and the seal construction feature matrix are spliced ​​and fused using shuffle blocks to obtain the deep construction feature matrix; Finally, the test analysis layer uses the average pooling layer, fully connected layer, random dropout layer and binary cross entropy-KL divergence loss function to optimize and update the deep feature matrix and anomaly location algorithm to complete the electronic contract verification.

Images