Engineering contract structure diagram organization management method and device
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUATENG JIANXIN TECH CO LTD
- Filing Date
- 2026-03-26
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies lack flexibility and accuracy in identifying issues in engineering contract management, thus affecting management efficiency.
Image encoding is performed using a generative adversarial neural network to generate virtual contract images and ground truth data of cost targets. A cost target recognition model is then trained using a convolutional neural network to identify contract images uploaded by users. Based on the recognition results, project costs are updated and warnings are issued.
It improved the accuracy and adaptability of contract identification, enhanced management efficiency, and enabled accurate cost updates and timely early warnings.
Smart Images
Figure CN122243411A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of contract management technology, and in particular to a method and apparatus for organizing and managing engineering contract structure diagrams. Background Technology
[0002] Engineering contracts are a crucial component of project management, especially in construction, manufacturing, and other large-scale projects. Contract execution and related cost management directly impact the smooth progress and economic benefits of the project. As contracts become increasingly complex, traditional manual cost management models face numerous challenges, particularly when processing and parsing large volumes of contract documents. Manual analysis is not only tedious but also prone to errors. Existing AI-based solutions, including image recognition, natural language processing (NLP), and machine learning methods, often struggle to adapt to the varied circumstances of different contracts. When dealing with complex contract information, they suffer from low flexibility, poor accuracy, and technical issues that negatively impact management efficiency. Summary of the Invention
[0003] This invention addresses the technical problems of low flexibility and poor accuracy in the identification of engineering contract structure diagrams, which affect management efficiency, by providing a method and apparatus for organizing and managing these diagrams.
[0004] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for organizing and managing engineering contract structure diagrams, wherein the method includes: Image encoding is performed using a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one.
[0005] Using the ground truth data of the cost target as supervision and the virtual contract image as input to the convolutional neural network, a cost target recognition model is trained.
[0006] The cost target identification model is used to process the contract image uploaded by the user to obtain the cost target identification result.
[0007] Based on the existing cost summary values, the identification results of the cost targets are superimposed to obtain the updated project cost results.
[0008] When the updated cost result of the project is greater than or equal to the cost threshold, a cost warning signal is sent to the user terminal.
[0009] Secondly, the present invention provides an engineering contract structure diagram organization and management device, wherein the device includes: The data generation module is used to encode images through a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one.
[0010] The recognition model training module is used to train a cost target recognition model by using the ground truth data of the cost target as supervision and the virtual contract image as input to a convolutional neural network.
[0011] The cost target identification module is used to process the contract image to be identified uploaded by the user terminal according to the cost target identification model to obtain the cost target identification result.
[0012] The project cost update module is used to overlay the cost target identification results based on the existing cost summary value to obtain the project cost update result.
[0013] A cost early warning module is used to obtain a cost early warning signal and send it to the user terminal when the updated cost result of the project is greater than or equal to a cost threshold.
[0014] The beneficial effects of this invention are as follows: Image encoding is performed using a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and ground truth data of cost targets correspond one-to-one; a cost target recognition model is trained using the ground truth data of cost targets as supervision and the virtual contract images as input to a convolutional neural network; the contract images to be recognized uploaded by the user terminal are processed according to the cost target recognition model to obtain cost target recognition results; the cost target recognition results are superimposed based on existing cost summary values to obtain project cost update results; when the project cost update result is greater than or equal to a cost threshold, a cost warning signal is obtained and sent to the user terminal. The engineering contract structure diagram organization and management method and apparatus disclosed in this invention solves the technical problems of low recognition flexibility, poor accuracy, and impact on management efficiency, achieving the technical effects of improving the accuracy and adaptability of contract recognition and enhancing management efficiency. Attached Figure Description
[0015] Figure 1 A flowchart illustrating the engineering contract structure diagram organization and management method provided by the present invention; Figure 2 This is a schematic diagram of the structure of the engineering contract structure diagram organization and management device provided by the present invention.
[0016] The attached diagram lists the components represented by each number as follows: Data generation module 11, identification model training module 12, cost target identification module 13, project cost update module 14, and cost early warning module 15. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0019] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.
[0020] Example 1: like Figure 1 As shown in the figure, this embodiment of the invention provides a method for organizing and managing engineering contract structure diagrams.
[0021] Image encoding is performed using a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one.
[0022] Specifically, firstly, a generative adversarial neural network is used to encode the contract image to generate a virtual contract image, which has a one-to-one corresponding cost object truth data. The generative adversarial neural network generates the virtual contract image through a generator, and the cost object truth data is structured data related to the contract content, including price, quantity, contract terms, etc.
[0023] In some embodiments, image encoding is performed using a generative adversarial neural network to obtain ground truth data of the virtual contract image and the cost object, including: A cost-encoded material library is obtained, wherein any coded material in the cost-encoded material library has a material attribute tag and a material target tag, wherein the material target represents a specific value; a white template of the contract target is obtained, wherein the white template of the contract target is a contract template in which the monetary value area is empty and the attribute area is not empty, wherein the monetary value area and the attribute area correspond one-to-one; the white template of the contract target, the material attribute tag, and the material target tag are encoded through the generative adversarial neural network to obtain the virtual contract image and the ground truth data of the cost target.
[0024] Specifically, a material library containing various cost-coded materials is first established as the basis for generating virtual contract images. This cost-coded material library contains various contract-related material elements, such as text, charts, and tables, and involves different contract terms and amounts. Each material has material attribute tags that describe the characteristics or category of the material, such as product name, price, quantity, payment terms, etc.; and material object tags that contain specific numerical values or numerical ranges.
[0025] Specifically, for the target interaction scenario or target user, based on the management task requirements, the corresponding basic contract template is obtained as the white template of the contract subject. The monetary value area in the white template of the contract subject is empty, and the attribute area has been defined. In other words, the non-empty attribute area reflects the purpose, attributes, task requirements and other characteristics of the white template of the contract subject, while the numerical part is empty and will be filled in according to the specific data in the subsequent process.
[0026] Specifically, by using a generative adversarial neural network, the white template of the contract subject, the material attribute labels, and the material subject labels are encoded. This includes matching the corresponding material attribute labels to the attribute regions in the white template and encoding appropriate material subject labels to generate a virtual contract image and extracting the corresponding cost subject ground truth data. Here, the cost subject ground truth data refers to the values of the encoded material subject labels.
[0027] In some implementations, the white template of the contract subject matter, the attribute tags of the materials, and the tags of the material subject matter are encoded to obtain the virtual contract image and the ground truth data of the cost subject matter, including: The generative adversarial neural network includes an encoder and a discriminator. It performs a one-to-one matching between the attribute regions and the material attribute labels to obtain a first set of material attribute labels and a first set of material target labels. Based on the first set of material attribute labels, it calculates the cost target of the first set of material target labels to obtain the ground truth data of the first cost target. It inputs the first set of material attribute labels, the first set of material target labels, the attribute regions, and the white template of the contract target into the encoder to obtain a first coded contract image. The discriminator judges the first coded contract image; if the judgment result is true, the first coded contract image is added to the virtual contract image, and the ground truth data of the first cost target is added to the ground truth data of the cost target.
[0028] Specifically, the white template of the contract subject is input into the encoder of the generative adversarial neural network. First, attribute matching is performed, that is, based on the attributes of the attribute region, multiple material attribute labels and multiple material target labels are determined to generate the first set of material attribute labels and the first set of material target labels. Among them, the first set of material target labels reflects all possible values of the first set of material attribute labels.
[0029] Specifically, based on the first set of material attribute tags, the values of the corresponding first set of material target tags are counted to form the specific content of each field in the contract image, i.e., the true value data of the first cost target; then, the first set of material attribute tags, the first set of material target tags, the attribute area and the white template of the contract target are used as input data and input into the encoder to generate the first coded contract image, i.e., the contract image filled with specific values (such as amount, delivery time, etc.).
[0030] Then, the generated first coded contract image is discriminated against by a discriminator. The discriminator determines whether the image conforms to the format and content of a real contract, such as whether the amount field and contract terms are logically consistent. If the discriminant result is true, that is, the generated contract image is similar to the real contract image, then the generated contract image is considered "real" and can be used as part of the virtual contract image. The image is added to the virtual contract image set, and at the same time, the corresponding first cost object truth data (such as contract amount) is also added to the cost object truth dataset.
[0031] The above process uses a generative adversarial neural network encoder and discriminator to generate and discriminate images, ensuring that the content in the generated virtual contract image is consistent with the true value data of the cost object, thereby generating an image that conforms to the format of a real contract.
[0032] Furthermore, the encoder training step includes: Collect the first set of material attribute record data, the first set of material target record data, attribute region record data, and a white template of contract target record data, wherein the attribute region record data has a one-to-one corresponding position in the white template of contract target record data; input the first set of material attribute record data, the first set of material target record data, the attribute region record data, and the white template of contract target record data into the encoder for unsupervised training to obtain a trained encoded contract image; when the discriminator's discrimination result for the trained encoded contract image is true for a preset number of consecutive training iterations, the encoder is generated.
[0033] Specifically, the first step is to obtain the first set of material attribute record data, the first set of material target record data, the attribute area record data, and the contract target record white template. The contract target record white template is a sample white template extracted from existing real contracts. The aforementioned first set of material attribute record data, first set of material target record data, attribute area record data, and contract target record white template contain the mapping relationship between material attributes, material targets, and attribute areas.
[0034] Specifically, the encoder is trained in an unsupervised manner using the first set of material attribute record data, the first set of material target record data, attribute region record data, and the white template of the contract target record as training data. During the training process, the encoder learns how to map the material attribute record data, material target record data, and attribute region data to the corresponding regions of the contract template, thereby enabling the encoder to acquire the ability to generate a contract image that is consistent with the structure of the real contract and has reasonable content.
[0035] Specifically, the discriminator performs a judgment after each generated contract image and gives a "true" or "false" result. If the generated contract image meets the discriminator's requirements, the judgment result is "true"; otherwise, it is "false". The encoder training requires multiple rounds of iteration. If the discriminator judges the generated contract images as "true" for a preset number of times (such as 10 times, 20 times, etc.), it means that the contract image generated by the encoder is close to the real contract image and can correctly fill in the relevant field content, and the training is considered complete.
[0036] Furthermore, the discriminator training step includes: Configure virtual contract image recording data, wherein the virtual contract image recording data has preset supervisory contract image recording data; wherein the first supervisory truth value of the virtual contract image recording data is false, and the second supervisory truth value of the supervisory contract image recording data is true; based on the first supervisory truth value and the second supervisory truth value, retrieve the virtual contract image recording data and the supervisory contract image recording data for supervised training, and generate the discriminator when the discriminator is correctly identified for a preset number of consecutive times.
[0037] Specifically, multiple generated supervisory contract image record data are acquired as virtual contract image record data. The supervisory contract image record data can be real contract images or virtual contract images generated by the encoder, and each virtual contract image has a corresponding supervisory contract image record data. By comparison, a discriminator can be trained to identify the differences between the generated images and real contract images.
[0038] Specifically, the first supervision truth value of the virtual contract image recording data is false, which means that the contract image was generated by the generator and is therefore false; while the second supervision truth value of the supervision contract image recording data is true, which means that the contract image is a real contract image and is therefore true.
[0039] Specifically, through supervised learning, the discriminator is trained to distinguish between real and virtual contract images. During training, the discriminator receives virtual and supervised contract images as input and attempts to classify them correctly. When a virtual contract image is input, the discriminator should output "false"; when a real contract image is input, the discriminator should output "true". The training process adjusts the error by comparing the discriminator's output with the actual supervised ground truth (i.e., "false" or "true"). When the error is large, the discriminator's parameters are updated using optimization methods such as gradient descent until it can more accurately identify the authenticity of contract images.
[0040] Specifically, a preset number of consecutive predictions is set as the standard for error-free judgment. That is, if the discriminator makes the correct predictions for a certain number of consecutive times in multiple iterations (e.g., 10 times or more), the training can be considered successful, meaning that the discriminator has successfully learned how to distinguish between virtual contract images and real contract images.
[0041] Using the ground truth data of the cost target as supervision and the virtual contract image as input to the convolutional neural network, a cost target recognition model is trained.
[0042] Specifically, the true value data of the cost target is the identification target result of the cost target identification, which corresponds to the virtual contract image input into the convolutional neural network. This cost target identification model based on the convolutional neural network determines the corresponding material target as the true value data of the cost target through convolution processing and material identification of the virtual contract image.
[0043] The cost target identification model is used to process the contract image uploaded by the user to obtain the cost target identification result.
[0044] In some embodiments, the cost target identification model is used to process the contract image to be identified uploaded by the user to obtain the cost target identification result, including: Obtain historical recognition data of the contract template of the contract image to be identified; analyze the historical recognition data of the contract template and calculate the historical recognition accuracy; when the historical recognition accuracy is less than or equal to a preset accuracy, copy K cost target recognition models and process the contract image to be identified in parallel to obtain K sets of cost target recognition results; evaluate the average value of the K sets of cost target recognition results with the same attribute to obtain the cost target recognition result.
[0045] Specifically, before processing the contract image to be identified, relevant historical recognition data of contract templates is first obtained, that is, the recognition results data of similar contract templates in history, including the accuracy and recognition status of each element (such as amount, quantity, etc.) in the contract template. This historical recognition data is used to evaluate the performance of historical recognition, that is, the historical recognition accuracy obtained by statistical analysis.
[0046] Specifically, if the historical recognition accuracy is lower than or equal to the preset accuracy threshold, it indicates that the recognition effect of the contract template is insufficient and additional computing resources are needed to improve its recognition accuracy. Correspondingly, the existing cost target recognition model is copied K times, and the contract image to be recognized is input into the K copied models for parallel processing, where K is a positive integer greater than 1.
[0047] Specifically, the obtained K sets of identification results are the different identification results given by K models. To ensure the accuracy of the identification results, the mean of the same attribute targets is further evaluated for all K sets of identification results. For example, targets with the same attribute (such as amount, time, etc. in the contract) are extracted from the output results of the K sets of models, and the mean of these targets with the same attribute is calculated as the final identification result. This helps to eliminate possible biases or errors in individual models and enhance the stability and accuracy of the identification results.
[0048] By using the parallel processing method described above, the prediction results of multiple models can be used to reduce errors and improve the accuracy and stability of identification.
[0049] In some embodiments, the method further includes: when the historical recognition accuracy is greater than the preset accuracy, or the historical recognition accuracy is empty, processing the contract image to be identified according to the cost target recognition model to obtain the cost target recognition result.
[0050] Specifically, if the historical recognition accuracy rate is greater than the preset accuracy rate, it can be considered that the recognition difficulty of the contract to be identified is relatively low, and it can be directly processed by a single cost target recognition model, thus having high recognition efficiency.
[0051] Specifically, if the historical identification accuracy rate is empty, it means that the contract to be identified is a contract template that appears for the first time. It is also directly processed through a single cost target identification model to obtain the identification result. Optionally, the identification accuracy rate analysis and judgment are performed on the identification results of the above-mentioned contracts to be identified with empty historical identification accuracy rates to further improve the stability and accuracy of the identification results.
[0052] Based on the existing cost summary values, the identification results of the cost targets are superimposed to obtain the updated project cost results.
[0053] Furthermore, based on the existing cost summary values, the cost target identification results are superimposed to obtain the project cost update results. For example, this includes correcting erroneous values, supplementing missing values, and adding new values to the existing cost summary values based on the cost target identification results to obtain the project cost update results.
[0054] When the updated cost result of the project is greater than or equal to the cost threshold, a cost warning signal is sent to the user terminal.
[0055] Specifically, the system determines whether to trigger an alert based on the updated project cost and cost threshold. If the updated project cost is greater than or equal to the cost threshold, there is a risk of the project cost exceeding the limit, and a corresponding cost alert signal is generated to provide timely warnings, thereby helping to better control costs.
[0056] The engineering contract structure diagram organization and management method provided in this embodiment of the invention has at least the following technical effects: Image encoding is performed using a generative adversarial neural network (GAN) to obtain virtual contract images and ground truth data of cost targets, with a one-to-one correspondence between the virtual contract images and the ground truth data of cost targets. A cost target recognition model is trained using the ground truth data of cost targets as supervision and the virtual contract images as input to a convolutional neural network. The cost target recognition model is then used to process the contract images uploaded by the user to obtain the cost target recognition results. Based on existing cost summary values, the cost target recognition results are overlaid to obtain updated project cost results. When the updated project cost result is greater than or equal to a cost threshold, a cost warning signal is generated and sent to the user. This achieves the technical effect of improving the accuracy and adaptability of contract recognition and enhancing management efficiency.
[0057] Example 2: like Figure 2 As shown, based on the same inventive concept as the engineering contract structure diagram organization and management method provided in Embodiment 1, this embodiment of the invention also provides an engineering contract structure diagram organization and management device.
[0058] Based on the same concept as the engineering contract structure diagram organization and management method in the above embodiments, the present invention also provides an engineering contract structure diagram organization and management device comprising: The data generation module 11 is used to encode images through a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one.
[0059] The recognition model training module 12 is used to train a cost target recognition model using the ground truth data of the cost target as supervision and the virtual contract image as input to a convolutional neural network.
[0060] The cost target identification module 13 is used to process the contract image to be identified uploaded by the user terminal according to the cost target identification model to obtain the cost target identification result.
[0061] The project cost update module 14 is used to overlay the cost target identification results based on the existing cost summary value to obtain the project cost update result.
[0062] The cost early warning module 15 is used to obtain a cost early warning signal and send it to the user terminal when the updated cost result of the project is greater than or equal to the cost threshold.
[0063] The data generation module 11 includes: The material acquisition unit is used to obtain a cost-coded material library, wherein any coded material in the cost-coded material library has a material attribute tag and a material target tag, wherein the material target represents a specific numerical value.
[0064] The white template acquisition unit is used to obtain a white template of the contract subject matter, wherein the white template of the contract subject matter is a contract template in which the monetary value area is empty and the attribute area is not empty, and the monetary value area and the attribute area correspond one-to-one.
[0065] The encoding unit is used to encode the white template of the contract subject, the material attribute tags, and the material subject tags through the generative adversarial neural network to obtain the virtual contract image and the ground truth data of the cost subject.
[0066] In some embodiments, the generative adversarial neural network in the encoding unit includes an encoder and a discriminator, and the execution steps of the encoding unit further include: Based on the attribute region and the material attribute tag, a first set of material attribute tags and a first set of material tag labels are obtained.
[0067] Based on the first group of material attribute tags, the cost target of the first group of material target tags is statistically analyzed to obtain the true value data of the first cost target.
[0068] The first set of material attribute tags, the first set of material target tags, the attribute area, and the white template of the contract target are input into the encoder to obtain the first coded contract image.
[0069] The discriminator judges the first coded contract image. When the judgment result is true, the first coded contract image is added to the virtual contract image, and the truth data of the first cost object is added to the truth data of the cost object.
[0070] In some embodiments, the encoder training step in the encoding unit includes: Collect the first set of material attribute record data, the first set of material target record data, attribute area record data, and the contract target record white template, wherein the attribute area record data has a one-to-one corresponding position in the contract target record white template.
[0071] The first set of material attribute record data, the first set of material target record data, the attribute area record data, and the white template of the contract target record are input into the encoder for unsupervised training to obtain the trained encoded contract image.
[0072] When the discriminator is trained for a preset number of consecutive times and the discrimination result of the trained encoded contract image is true for all of them, the encoder is generated.
[0073] In some embodiments, the discriminator training step in the coding unit includes: Configure virtual contract image recording data, wherein the virtual contract image recording data has preset supervision contract image recording data.
[0074] Wherein, the first supervision truth value of the virtual contract image recording data is false, and the second supervision truth value of the supervision contract image recording data is true.
[0075] Based on the first supervised truth value and the second supervised truth value, the virtual contract image recording data and the supervised contract image recording data are retrieved for supervised training. When the discriminator is generated after a preset number of consecutive correct judgments, it is then generated.
[0076] In some embodiments, the cost target identification module 13 includes: The contract template historical recognition data acquisition unit is used to obtain the contract template historical recognition data of the contract image to be recognized.
[0077] The historical recognition accuracy analysis unit is used to analyze the historical recognition data of the contract template and to calculate the historical recognition accuracy.
[0078] The accuracy judgment and model replication unit is used to replicate K cost target recognition models when the historical recognition accuracy is less than or equal to a preset accuracy, and to perform parallel processing on the contract image to be recognized to obtain K sets of cost target recognition results.
[0079] The identification result mean evaluation unit is used to evaluate the mean of the identification results of the K groups of cost targets with the same attribute, and obtain the identification results of the cost targets.
[0080] In some embodiments, the execution steps of the cost target identification module 13 further include: When the historical recognition accuracy is greater than the preset accuracy, or when the historical recognition accuracy is empty, the image of the contract to be identified is processed according to the cost target recognition model to obtain the cost target recognition result.
[0081] It should be understood that the focus of the embodiments mentioned in this specification is their difference from other embodiments. The specific embodiments in the aforementioned Embodiment 1 are also applicable to the engineering contract structure diagram organization and management device described in Embodiment 2. For the sake of brevity, they will not be further elaborated here.
[0082] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.
[0083] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for organizing and managing engineering contract structure diagrams, characterized in that, include: Image encoding is performed using a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one. Using the true value data of the cost target as supervision and the virtual contract image as input to the convolutional neural network, a cost target recognition model is trained. The cost target identification model is used to process the contract image uploaded by the user to be identified, and the cost target identification result is obtained. Based on the existing total cost value, the identification results of the cost targets are superimposed to obtain the updated project cost results; When the updated cost result of the project is greater than or equal to the cost threshold, a cost warning signal is sent to the user terminal.
2. The method as described in claim 1, characterized in that, Image encoding is performed using a generative adversarial neural network to obtain ground truth data for virtual contract images and cost targets, including: Obtain a cost-coded material library, wherein any coded material in the cost-coded material library has a material attribute tag and a material target tag, wherein the material target represents a specific value; Obtain a white template of the contract subject matter, wherein the white template of the contract subject matter is a contract template in which the monetary value area is empty and the attribute area is not empty, and the monetary value area and the attribute area correspond one-to-one. The generative adversarial neural network is used to encode the white template of the contract subject, the material attribute tags, and the material subject tags to obtain the virtual contract image and the ground truth data of the cost subject.
3. The method as described in claim 2, characterized in that, The generative adversarial neural network is used to encode the white template of the contract subject matter, the attribute tags of the materials, and the tags of the material subject matter to obtain the virtual contract image and the ground truth data of the cost subject matter, including: The generative adversarial neural network includes an encoder and a discriminator; Based on the attribute region and the material attribute tag, a first set of material attribute tags and a first set of material tag tags are obtained; Based on the first group of material attribute tags, the cost target of the first group of material target tags is statistically analyzed to obtain the true value data of the first cost target; The first set of material attribute tags, the first set of material target tags, the attribute area, and the white template of the contract target are input into the encoder to obtain the first coded contract image; The discriminator judges the first coded contract image. When the judgment result is true, the first coded contract image is added to the virtual contract image, and the truth data of the first cost object is added to the truth data of the cost object.
4. The method as described in claim 3, characterized in that, The encoder training steps include: Collect the first set of material attribute record data, the first set of material target record data, attribute area record data, and the contract target record white template, wherein the attribute area record data has a one-to-one corresponding position in the contract target record white template; The first set of material attribute record data, the first set of material target record data, the attribute area record data, and the white template of the contract target record are input into the encoder for unsupervised training to obtain the trained encoded contract image; When the discriminator is trained for a preset number of consecutive times and the discrimination result of the trained encoded contract image is true for all of them, the encoder is generated.
5. The method as described in claim 3, characterized in that, The discriminator training steps include: Configure virtual contract image recording data, wherein the virtual contract image recording data includes preset supervision contract image recording data; Wherein, the first supervision truth value of the virtual contract image recording data is false, and the second supervision truth value of the supervision contract image recording data is true; Based on the first supervised truth value and the second supervised truth value, the virtual contract image recording data and the supervised contract image recording data are retrieved for supervised training. When the discriminator is generated after a preset number of consecutive correct judgments, it is then generated.
6. The method as described in claim 1, characterized in that, The cost target identification model is used to process the contract image uploaded by the user to obtain the cost target identification result, including: Obtain historical recognition data of the contract template for the contract image to be identified; Analyze the historical identification data of the contract template and calculate the historical identification accuracy rate; When the historical recognition accuracy is less than or equal to the preset accuracy, K cost target recognition models are copied, and the contract image to be recognized is processed in parallel to obtain K sets of cost target recognition results; The cost target identification results of the K groups are evaluated by the mean value of targets with the same attribute to obtain the cost target identification results.
7. The method as described in claim 6, characterized in that, Also includes: When the historical recognition accuracy is greater than the preset accuracy, or when the historical recognition accuracy is empty, the image of the contract to be identified is processed according to the cost target recognition model to obtain the cost target recognition result.
8. An engineering contract structure diagram organization and management device, characterized in that, The apparatus is used to execute the engineering contract structure diagram organization and management method according to any one of claims 1-7, the apparatus comprising: A data generation module is used to encode images through a generative adversarial neural network to obtain virtual contract images and ground truth data of cost targets, wherein the virtual contract images and the ground truth data of cost targets correspond one-to-one. The recognition model training module is used to train a cost target recognition model by using the ground truth data of the cost target as supervision and the virtual contract image as input to a convolutional neural network. A cost target identification module is used to process the contract image to be identified uploaded by the user terminal according to the cost target identification model to obtain the cost target identification result; The project cost update module is used to overlay the cost target identification results based on the existing cost summary value to obtain the project cost update result; A cost early warning module is used to obtain a cost early warning signal and send it to the user terminal when the updated cost result of the project is greater than or equal to a cost threshold.
9. The apparatus as claimed in claim 8, characterized in that, The data generation module includes: The material acquisition unit is used to obtain a cost-coded material library, wherein any coded material in the cost-coded material library has a material attribute tag and a material target tag, wherein the material target represents a specific value; The white template acquisition unit is used to obtain a white template of the contract subject matter, wherein the white template of the contract subject matter is a contract template in which the monetary value area is empty and the attribute area is not empty, and the monetary value area and the attribute area correspond one-to-one. An encoding unit is used to encode the white template of the contract subject, the material attribute tags, and the material subject tags through the generative adversarial neural network to obtain the virtual contract image and the ground truth data of the cost subject.
10. The apparatus as claimed in claim 9, characterized in that, The execution steps of the encoding unit further include: The generative adversarial neural network includes an encoder and a discriminator; Based on the attribute region and the material attribute tag, a first set of material attribute tags and a first set of material tag tags are obtained; Based on the first group of material attribute tags, the cost target of the first group of material target tags is statistically analyzed to obtain the true value data of the first cost target; The first set of material attribute tags, the first set of material target tags, the attribute area, and the white template of the contract target are input into the encoder to obtain the first coded contract image; The discriminator judges the first coded contract image. When the judgment result is true, the first coded contract image is added to the virtual contract image, and the truth data of the first cost object is added to the truth data of the cost object.