阅读说明：本技术 基于国家自然科学基金立项项目管理方法及系统 (Project management method and system based on national natural science foundation project ) 是由 石建 于 2021-05-24 设计创作，主要内容包括：本发明公开了基于国家自然科学基金立项项目管理方法,该方法包括项目新增、国自然基金立项书图片附件上传、自动识别提取基金项目表单信息、自动填写系统中表单的基金项目信息、编辑、提交审批、审核基金项目信息、微信小程序查阅审核结果的步骤,并可自动发送站内消息；基金项目管理系统包括服务器、交互界面、图片识别模块、项目立项管理模块、纵向项目列表模块；该基金项目信息管理方法及系统,操作简单、维护容易；校级管理员审核人操作完毕后,系统可自动推送到微信小程序端,个人及时查看审核结果,节约流程时间,也可以及时告知申报人及参与人项目信息审核进展。(The invention discloses a project management method based on national natural science foundation, which comprises the steps of project increase, uploading of picture attachments of national natural foundation books, automatic identification and extraction of foundation project form information, automatic filling of foundation project information of forms in a system, editing, submission and approval, examination and verification of foundation project information, and inquiry and verification result lookup of WeChat small programs, and can automatically send out messages in a station; the fund project management system comprises a server, an interactive interface, a picture identification module, a project establishment management module and a longitudinal project list module; the fund project information management method and the fund project information management system are simple to operate and easy to maintain; after the operation of the level correction manager is finished, the system can be automatically pushed to the WeChat small program end, and an individual can check the verification result in time, so that the flow time is saved, and the project information verification progress of the applicant and the participators can be informed in time.)

1. The project management method based on national science fund project is characterized by comprising the following steps: the method comprises the following steps:

A. clicking project management-vertical project-project establishment;

B. clicking a new button, clicking an uploaded itemized book picture accessory, and automatically filling the identified fund project form information into the current project information form through an artificial intelligent image identification service based on CTPN & CRNN;

C. the personal can edit the project form information, and click 'submit approval' through the interactive interface to be submitted to the upper-level administrator for examination after the information is confirmed to be correct;

D. the project declaration plan updating declaration state is 'in declaration', and simultaneously sending an in-station audit message to display that the project is submitted in a longitudinal project list of project establishment;

E. in the operation column of the project declaration plan, operations such as cancellation, printing and the like can be carried out on the submitted project;

F. if the audit is passed, the state displayed on the vertical project list interface of the project establishment is 'audit passed', an internal message is sent to the applicant at the same time, and the internal message is also pushed to a WeChat small program end, and the establishment project is stored in school level, institution level and individual and participant libraries; if the audit is not passed, the state displayed on the vertical project list interface of the project establishment is 'no pass', and meanwhile, an internal message is sent to the applicant, and meanwhile, the internal message is also pushed to the WeChat applet end.

2. The method of managing projects based on national science foundation as claimed in claim 1, wherein the image recognition technology in the image recognition service: uploading a standing book picture or a file in a standing book PDF format, and automatically extracting characters in the file;

the image identification technology specifically comprises the following steps:

the method comprises the steps that a model based on a CTPN + CRNN network structure is used for extracting text information in a project book picture or a PDF file and then automatically inputting the text information;

CTPN (connectionist Text forward network), here CTPN variants, detects transversely arranged Text information, inputs a picture of a project book:

firstly, extracting features through a CNN feature extraction network Resnet50 network to obtain a conv5 feature map with the size of 1C H W, wherein C is the number of channels, H is the picture height, and W is the picture width;

then make 3 × 3 sliding windows on conv5, i.e. each point combines the surrounding 3 × 3 region features to obtain a feature vector with length 3 × C; outputting a feature map of 1 × 9C × H × W, which is obviously only the spatial feature learned by CNN;

then feature map is conducted to Reshape: 1 × 9C × H → (1 × H) W × 9C;

then, a data stream with N equal to 1 (here, 1 item book picture), so that Batch of Batch equal to NH and maximum time length T equal to W is input into a bidirectional LSTM (long and short term memory artificial intelligence network), and a sequence feature of each line is learned; bidirectional LSTM output (NH) × W256, then Reshape restores shape:

(NH) × W256 → N × 256 × H × W, which contains both spatial and LSTM learned sequence features;

then through the fully connected convolution layer, the characteristic of N512H W is changed, and finally through the RPN network similar to fast R-CNN, text messages are obtained;

CTPN has 3 outputs, text/non-text score, vertical coordinate V ═ V_c，v_hAnd left and right horizontal offsets O;

the CTPN contains 3 loss functions,a classification loss function, which distinguishes whether text is present or not by softmax loss,andare all regression functionsBy L₁And (4) calculating a function. The total loss function is as follows:

wherein λ 1 ═ 1 and λ 2 ═ 2

Then, carrying out character recognition on the taken text tokens, wherein the character recognition is carried out by using a network architecture of CRNN + CTC;

CRNN is divided into three parts: convolutional layer, cyclic network layer, transcription output layer;

the convolutional layer is a common CNN network used for extracting image features, a Resnet50 network is used for extracting the image features, the cyclic network layer is a BI-LSTM, a double-layer BI-LSTM is used for stack, the aim of continuously extracting character sequence features after convolution is achieved, and the output layer outputs the RNN as softmax and then outputs characters;

CTC (connectionist Temporal classification), the CRNN result obtained above is sent to a CTC algorithm model, which is a complete end-to-end model training, and only one input sequence and one output sequence are required for training without aligning data in advance; data alignment and one-to-one labeling are not needed, and CTC directly outputs the probability of sequence prediction without external post-processing; blank is introduced into CTC, that is, there is no predicted value at this moment, each predicted classification corresponds to a spike (peak) in a whole text, and other positions which are not peaks are regarded as blank; for a piece of text, the final output of a CTC is a sequence of spikes, not concerning how long each word lasts; CTC is by gradientAdjusting ω -parameter of LSTM to maximize p (l | x); (p (l | x) is the probability of the output l given the input x, ω is the LSTM hidden layer parameter); this results in processed text information.

3. The system of claim 2, wherein the fund project management system comprises a server, an interactive interface, a picture recognition module, a project management module, and a vertical project list module;

the server is used for installing an artificial intelligent image recognition module;

the interactive interface is used for inputting fund project form information, displaying an information list in the management module and displaying project establishment management states;

the picture identification module is used for identifying the picture or PDF of the national natural fund book;

the fund project form information required to be input in the project establishment management module comprises basic information, project members, budget information and project documents; the project establishment management module comprises a project establishment management module, and an information input sub-module and an information auditing sub-module are also arranged under the project establishment management module;

the information input sub-module is used for inputting fund project form information, interactively displaying the management state of the fund project form information, and sending audit wechat applet information and in-station information;

the information auditing submodule is used for auditing project establishment information, displaying the auditing state of the fund project through an interactive interface, and sending a WeChat applet message and an in-station message; entering an information auditing submodule by clicking a message or a functional tree node in a longitudinal project list auditing station of an interactive interface, and simultaneously providing auditing and editing operation functions by the information auditing submodule;

and the longitudinal project list module is used for inputting fund project form information, displaying an information list in the management module, displaying in-station information and WeChat applet information and displaying project establishment management states.

4. The national science foundation project management system of claim 3 wherein the management status includes "submission approval," pending approval, "" approved, "or" not approved.

5. The national science foundation project management system of claim 4, wherein in the vertical project list module, the information list is displayed in a classified manner according to all, approved or not approved, in-approved or temporary storage; the longitudinal project list module can retrieve, view, import and export fund project form information; the vertical item list module can export the entered item information through an export function, and the exported information is in an Excel format.

6. The national science foundation project management system according to claim 5, wherein the information entry submodule can enter the foundation project form information, and after the foundation project form information is entered, the information entry submodule enters the information auditing submodule by clicking 'submission approval'; if the audit is passed, the application form state displayed by the interactive interface is 'audit passed', and simultaneously sending an in-station message and a WeChat applet message to the applicant; and if the audit is not passed, the application form displayed by the interactive interface is in an 'audit failed' state, and simultaneously sending the station internal message and the WeChat applet message to the applicant.

7. The national science foundation project management system according to claim 6, wherein after the fund project form information is entered, the entered information can be modified and deleted in "scratch" before entering the information auditing sub-module.

8. The system according to claim 7, wherein after entering the information auditing submodule, a revocation function is performed before the auditing operation is not performed; after the revocation function is executed, the state of the updatable information is 'stopped', and the in-station message sent to the auditor is deleted.

9. The system according to claim 8, wherein when a new project is established or information is to be entered, a desired project picture or project PDF is uploaded and sent to an artificial intelligence image recognition module, and the image recognition module analyzes and extracts characters and automatically fills in a form; the information input sub-module automatically inputs the fund project form information identified by the image and can compare the input fund project form information

After the fund project form information is input, the project establishment management module inquires the added information according to the retrieval conditions such as project number, project name, responsible person, project classification, project state, starting date, plan ending date and the like; the basic information to be input in the project establishment management module comprises an approval number, a project name, a project number, a responsible person type, a responsible person name, a affiliated department, a primary subject, a project classification, a subject category, a project level, a project source unit, a project state, an establishment date, a start date, a plan ending date, superior subsidy expenses, a result form and remarks;

after the manual intelligent image recognition is used for automatically inputting the fund project form information, only the step of confirming the fund project form information without errors is needed;

filling the basic information into the information of the next confirmed project members, and checking the member types, names, unit names, titles, academic calendars, undertaking types and the like;

confirming budget information after confirming project member information, and then, confirming preselection information and then, going to a project document;

selecting a file type in the project document, and supporting uploading of files in a file name column; and simultaneously supports operations of adding new files and deleting files.

Technical Field

The invention belongs to the technical field of computer information management, and discloses a project management method and system based on national natural science fund establishment.

Background

In many colleges and universities in China, many colleges and universities independently develop new technologies, new products, release patents, papers, generate new achievements and the like, current scientific research activities and scientific research capabilities become important evaluation indexes in talent culture and evaluation of the colleges and universities, and the booster is also used for promoting teachers to improve business levels and teaching quality and social service capabilities. However, most research and development management work of these research projects currently stays in the paper office.

With the development of computer software and hardware technology and network technology, computer and network technology has penetrated all corners of society. The internal management of many enterprises and institutions has been networked, which not only increases the management efficiency and saves the management time, but also reduces many unnecessary troubles, and the advantages of the networked management are seen.

In order to strengthen scientific research management vigorously and control on scientific research projects in colleges and universities, fine management is comprehensively implemented to different scientific research modules and runs through various fields of various colleges, project management informatization and paperless office is realized, and it is necessary to develop a high-efficiency and reasonable scientific research information management system.

The invention discloses a scientific research management information system of a three-layer higher vocational school based on NET (network application number) in Chinese invention patent with application number 201110079403.6, wherein the system adopts ASP.NET + SQLSever2005 technology, selects B/S mode, and uses C # to realize the functions of system data setting, scientific research result management, scientific research subject management, inquiry, browsing and statistics and system maintenance and management. The system can basically meet the working requirements of scientific research management of some higher vocational schools, and can improve the working efficiency of the higher vocational schools to a certain extent. However, the operation of the system requires that all information is filled by individuals, the maintenance is difficult, the difference between the information filled by individuals and project filing information is not compared, and the intelligent scientific research management cannot be met at the same time.

Disclosure of Invention

The purpose of the invention is as follows: in order to overcome the defects of the prior art and reduce the recording time and labor cost, the invention provides a project management method and system based on national science foundation project, which has short manual time, is easy to learn, convenient to maintain and comprehensive in record.

The technical scheme is as follows: in order to achieve the purpose of the invention, the invention adopts the technical scheme that:

the project management method based on national science fund foundation includes the following steps:

A. clicking project management-longitudinal project-project establishment in a project establishment management module;

B. clicking a newly added button in the longitudinal project list module, clicking an uploaded project book image in the image recognition module to perform artificial intelligent image recognition, and automatically inputting the recognized fund project form information into a current form in the information input sub-module;

C. confirming that the user can submit the file by clicking 'submission approval' through an interactive interface; if an error exists, the device can be automatically modified;

D. updating the status of the item to be submitted in a longitudinal item list submodule of the item establishment management module, and simultaneously sending a message in the audit station;

E. the operation of canceling, printing and the like can be carried out on the submitted item in the column of the submitted operation of the longitudinal item list submodule of the item standing management module;

F. in the information auditing module, if the auditing is passed, the state displayed on a list interface in a longitudinal project submodule of the project establishment management module is 'approved', an internal message is sent to an applicant at the same time, meanwhile, the internal message is pushed from a WeChat small program end, and the establishment projects are stored in school, institution and individual and participant libraries; if the audit is not passed, the state displayed on the list interface in the longitudinal project submodule of the project establishment management module is 'no pass', and meanwhile, the message in the station is sent to the applicant, and meanwhile, the message is pushed from the WeChat small program end.

Further, an image recognition technique in the image recognition service: uploading a standing book picture or a file in a standing book PDF format, and automatically extracting characters in the file;

the image identification technology specifically comprises the following steps:

the method comprises the steps that a model based on a CTPN + CRNN network structure is used for extracting text information in a project book picture or a PDF file and then automatically inputting the text information;

CTPN (connectionist Text forward network), here CTPN variants, detects transversely arranged Text information, inputs a picture of a project book:

firstly, extracting features through a CNN feature extraction network Resnet50 network to obtain a conv5 feature map with the size of 1C H W, wherein C is the number of channels, H is the picture height, and W is the picture width;

then make 3 × 3 sliding windows on conv5, i.e. each point combines the surrounding 3 × 3 region features to obtain a feature vector with length 3 × C; outputting a feature map of 1 × 9C × H × W, which is obviously only the spatial feature learned by CNN;

then feature map is conducted to Reshape: 1 × 9C × H → (1 × H) W × 9C;

then, a data stream with N equal to 1 (here, 1 item book picture), so that Batch of Batch equal to NH and maximum time length T equal to W is input into a bidirectional LSTM (long and short term memory artificial intelligence network), and a sequence feature of each line is learned; bidirectional LSTM output (NH) × W256, then Reshape restores shape:

(NH) × W256 → N × 256 × H × W, which contains both spatial and LSTM learned sequence features;

then through the fully connected convolution layer, the characteristic of N512H W is changed, and finally through the RPN network similar to fast R-CNN, text messages are obtained;

CTPN has 3 outputs, text/non-text score, vertical coordinate V ═ V_c，v_hAnd left and right horizontal offsets O;

the CTPN contains 3 loss functions,a classification loss function, which distinguishes whether text is present or not by softmax loss,andare all regression functions, using L₁And (4) calculating a function. The total loss function is as follows:

where 1 and 2 are equal to each other

Then, carrying out character recognition on the taken text tokens, wherein the character recognition is carried out by using a network architecture of CRNN + CTC;

CRNN is divided into three parts: convolutional layer, cyclic network layer, transcription output layer;

the convolutional layer is a common CNN network used for extracting image features, a Resnet50 network is used for extracting the image features, the cyclic network layer is a BI-LSTM, a double-layer BI-LSTM is used for stack, the aim of continuously extracting character sequence features after convolution is achieved, and the output layer outputs the RNN as softmax and then outputs characters;

CTC (connectionist Temporal classification), the CRNN result obtained above is sent to a CTC algorithm model, which is a complete end-to-end model training, and only one input sequence and one output sequence are required for training without aligning data in advance; data alignment and one-to-one labeling are not needed, and CTC directly outputs the probability of sequence prediction without external post-processing; blank is introduced into CTC, that is, there is no predicted value at this moment, each predicted classification corresponds to a spike (peak) in a whole text, and other positions which are not peaks are regarded as blank; for a piece of text, the final output of a CTC is a sequence of spikes, not concerning how long each word lasts; CTC is by gradientAdjusting ω -parameter of LSTM to maximize p (l | x); (p (l | x) is the probability of the output l given the input x, ω is the LSTM hidden layer parameter); this results in processed text information.

The system comprises a server, an interactive interface, a picture identification module, a project establishment management module and a longitudinal project list module; the server is used for installing an artificial intelligent image recognition module;

the interactive interface is used for inputting fund project form information, displaying an information list in the management module and displaying project establishment management states;

the picture identification module is used for identifying the picture or PDF of the national natural fund book;

the fund project form information required to be input in the project establishment management module comprises basic information, project members, budget information and project documents; the project establishment management module comprises a project establishment management module, and an information input sub-module and an information auditing sub-module are also arranged under the project establishment management module;

the information input sub-module is used for inputting fund project form information, interactively displaying the management state of the fund project form information, and sending audit wechat applet information and in-station information;

the information auditing submodule is used for auditing project establishment information, displaying the auditing state of the fund project through an interactive interface, and sending a WeChat applet message and an in-station message; entering an information auditing submodule by clicking a message or a functional tree node in a longitudinal project list auditing station of an interactive interface, and simultaneously providing auditing and editing operation functions by the information auditing submodule;

and the longitudinal project list module is used for inputting fund project form information, displaying an information list in the management module, displaying in-station information and WeChat applet information and displaying project establishment management states.

Further, the management state includes "submission approval", "to be approved", "approval pass" or "approval fail".

Further, in the longitudinal item list module, the information list is displayed in a classified manner according to all, verification and audit pass, verification and audit fail, verification in process, temporary storage and the like; the longitudinal project list module can retrieve, view, import and export fund project form information; the vertical item list module can export the entered item information through an export function, and the exported information is in an Excel format.

Furthermore, the information input sub-module can input fund project form information, and after the fund project form information is input, the fund project form information enters the information auditing sub-module by clicking 'submission approval'; if the audit is passed, the application form state displayed by the interactive interface is 'audit passed', and simultaneously sending an in-station message and a WeChat applet message to the applicant; and if the audit is not passed, the application form displayed by the interactive interface is in an 'audit failed' state, and simultaneously sending the station internal message and the WeChat applet message to the applicant.

Furthermore, after the fund project form information is input and before the fund project form information enters the information auditing submodule, the input information can be modified and deleted in the temporary storage.

Furthermore, after the information auditing submodule is entered and before the auditing operation is not executed, a withdrawing function can be carried out; after the revocation function is executed, the state of the updatable information is 'stopped', and the in-station message sent to the auditor is deleted.

Further, when the new project is established or information is ready to be input, uploading a required project picture or project PDF and sending the project picture or the project PDF into an artificial intelligent image recognition module, and analyzing and extracting characters and automatically filling the characters into a form by the image recognition module;

the information input sub-module automatically inputs the fund project form information identified by the image and can compare the input fund project form information

After the fund project form information is input, the project establishment management module inquires the added information according to the retrieval conditions such as project number, project name, responsible person, project classification, project state, starting date, plan ending date and the like; the basic information to be input in the project establishment management module comprises an approval number, a project name, a project number, a responsible person type, a responsible person name, a affiliated department, a primary subject, a project classification, a subject category, a project level, a project source unit, a project state, an establishment date, a start date, a plan ending date, superior subsidy expenses, a result form and remarks;

after the manual intelligent image recognition is used for automatically inputting the fund project form information, only the step of confirming the fund project form information without errors is needed;

filling the basic information into the information of the next confirmed project members, and checking the member types, names, unit names, titles, academic calendars, undertaking types and the like;

confirming budget information after confirming project member information, and then, confirming preselection information and then, going to a project document;

selecting a file type in the project document, and supporting uploading of files in a file name column; and simultaneously supports operations of adding new files and deleting files.

Has the advantages that: compared with the prior art, the invention has the advantages that the classification management is clear at a glance, and the operation is simple and convenient;

the inquiry is convenient through multiple channels; the information is automatically input through an artificial intelligent picture recognition technology, and the operation flow is simple and easy to learn; after the operation of the applicant or the auditor is finished, the system can automatically remind and automatically send a message to inform the other side, so that the flow time is saved, and the other side can timely inform the audit progress; the revocation function can be timely revoked when the applicant finds that the input information is wrong or other reasons cause that the auditing step cannot be entered, so that some unnecessary operations are avoided; meanwhile, the project declaration can be edited and resubmitted again after being submitted, and the project declaration can be printed without withdrawing.

The system and the method can be suitable for the national science foundation project information management of teachers or students in colleges and universities.

Students, teachers, subject managers, academic achievement managers, student management departments and the like can all be maintained and managed in a grading mode through the system, and various national science foundation project information can be extracted from the library according to different people with different purposes.

Drawings

FIG. 1 is a block diagram of a teacher's project claim declaration management method based on a national science foundation project in accordance with the present invention;

Detailed Description

The present invention is further illustrated by the following specific examples, which are not intended to be limiting.

The system comprises a server, an information input interactive interface sub-module, a picture recognition module, a project establishment management module and a longitudinal project list module;

the server is used for installing an artificial intelligent image recognition module;

the interactive interface is used for inputting fund project form information, displaying an information list in the management module and displaying project establishment management states;

the picture identification module is used for identifying the picture or PDF of the national natural fund book;

the fund project form information required to be input in the project establishment management module comprises basic information, project members, budget information and project documents;

the project establishment management module comprises a project establishment management module, and an information input sub-module and an information auditing sub-module are also arranged under the project establishment management module;

the information input sub-module is used for inputting fund project form information, interactively displaying the management state of the fund project form information, and sending audit wechat applet information and in-station information;

the information auditing submodule is used for auditing project establishment information, displaying the auditing state of the fund project through an interactive interface, and sending a WeChat applet message and an in-station message; entering an information auditing submodule by clicking a message or a functional tree node in a longitudinal project list auditing station of an interactive interface, and simultaneously providing auditing and editing operation functions by the information auditing submodule;

and the longitudinal project list module is used for inputting fund project form information, displaying an information list in the management module, displaying in-station information and WeChat applet information and displaying project establishment management states.

Further in this example, the management status includes "submit approval", "pending audit", "audit passed" or "audit not passed".

In this embodiment, in the longitudinal item list module, the information list is displayed in a classified manner according to all, verification and audit pass, verification and audit fail, verification in the middle, temporary storage, and the like; the longitudinal project list module can retrieve, view, import and export fund project form information; the vertical item list module can export the entered item information through an export function, and the exported information is in an Excel format.

In this embodiment, the information entry sub-module may enter the fund project form information, and after entering the fund project form information, the information entry sub-module enters the information auditing sub-module by clicking "submit approval"; if the audit is passed, the application form state displayed by the interactive interface is 'audit passed', and simultaneously sending an in-station message and a WeChat applet message to the applicant; and if the audit is not passed, the application form displayed by the interactive interface is in an 'audit failed' state, and simultaneously sending the station internal message and the WeChat applet message to the applicant.

In this embodiment, after the fund project form information is entered, the entered information may be modified and deleted in the "temporary storage" before entering the information auditing sub-module.

In this embodiment, further, after entering the information auditing sub-module and before not performing the auditing operation, a revocation function may be performed; after the revocation function is executed, the state of the updatable information is 'stopped', and the in-station message sent to the auditor is deleted.

In this embodiment, when the new project is established or information is to be input, the needed project picture or project PDF is uploaded and sent to the artificial intelligent image recognition module, and the image recognition module analyzes and extracts characters and automatically fills in a form;

the information input sub-module automatically inputs the fund project form information identified by the image and can compare the input fund project form information.

After the fund project form information is input, the project establishment management module inquires the added information according to the retrieval conditions such as project number, project name, responsible person, project classification, project state, starting date, plan ending date and the like; the basic information to be input in the project establishment management module comprises an approval number, a project name, a project number, a responsible person type, a responsible person name, a affiliated department, a primary subject, a project classification, a subject category, a project level, a project source unit, a project state, an establishment date, a start date, a plan ending date, superior subsidy expenses, a result form and remarks;

after the manual intelligent image recognition is used for automatically inputting the fund project form information, only the step of confirming the fund project form information without errors is needed;

filling the basic information into the information of the next confirmed project members, and checking the member types, names, unit names, titles, academic calendars, undertaking types and the like;

confirming budget information after confirming project member information, and then, confirming preselection information and then, going to a project document; selecting a file type in the project document, and supporting uploading of files in a file name column; and simultaneously supports operations of adding new files and deleting files.

As shown in fig. 1, the project management method based on national science foundation includes the following steps:

G. clicking project management-longitudinal project-project establishment in a project establishment management module;

H. clicking a newly added button in the longitudinal project list module, clicking an uploaded project book image in the image recognition module to perform artificial intelligent image recognition, and automatically inputting the recognized fund project form information into a current form in the information input sub-module;

I. confirming that the user can submit the file by clicking 'submission approval' through an interactive interface; if an error exists, the device can be automatically modified;

J. updating the status of the item to be submitted in a longitudinal item list submodule of the item establishment management module, and simultaneously sending a message in the audit station;

K. the operation of canceling, printing and the like can be carried out on the submitted item in the column of the submitted operation of the longitudinal item list submodule of the item standing management module;

l, in the information auditing module, if the auditing is passed, the state displayed on a list interface in a longitudinal project submodule of the project establishment management module is 'approved', an in-station message is sent to an applicant at the same time, meanwhile, the in-station message is pushed from a WeChat small program end, and the establishment project is stored in a school level, a hospital level and a personal and participant library; if the audit is not passed, the state displayed on the list interface in the longitudinal project submodule of the project establishment management module is 'no pass', and meanwhile, the message in the station is sent to the applicant, and meanwhile, the message is pushed from the WeChat small program end.

In this example, the artificial intelligence image recognition module is a model based on a CTPN + CRNN network structure, and is used for extracting text information in a project establishment book picture or a project establishment PDF file, and then automatically inputting the text information;

CTPN (connectionist Text forward network), where CTPN variants detect horizontally arranged Text messages, and input a project book picture:

first, features are extracted through a Resnet50 network (a CNN feature extraction network) to obtain a conv5 feature map with a size of 1 × C × H × W (C is the number of channels, H is the picture height, and W is the picture width).

Then a 3 x 3 sliding window is made on conv5, i.e. each point combines the surrounding 3 x 3 region features to obtain a feature vector of length 3 x C. And outputting a feature map of 1 × 9C × H × W, wherein the feature obviously only has the spatial feature learned by the CNN.

Then feature map is conducted to Reshape: 1 × 9C × H × W → (1 × H) W × 9C

Then, since N is equal to 1 (here, 1 item book picture), a data stream of Batch NH and maximum time length T is W is input to a bidirectional LSTM (long and short memory artificial intelligence network), and sequence features of each line are learned. Bidirectional LSTM output (NH) × W256, then Reshape restores shape:

(NH) × W256 → N × 256 × H × W, which contains both spatial and LSTM learned sequence features.

Then through the fully connected convolution layer, it becomes the characteristic of N512H W, and finally through the RPN network like fast R-CNN, it obtains text messages.

CTPN has 3 outputs, text/non-text score, vertical coordinate V ═ V_c，v_hAnd left and right horizontal offsets O. The CTPN contains 3 loss functions,a classification loss function, which distinguishes whether text is present or not by softmax loss,andare all regression functions, using L₁And (4) calculating a function. The total loss function is as follows:

wherein λ 1 ═ 1 and λ 2 ═ 2

Then the text tokens taken in is subjected to character recognition, and the invention uses the network architecture of the CRNN + CTC to perform recognition.

CRNN is divided into three parts: convolutional layer, cyclic network layer, and transcription output layer.

The convolutional layer is a common CNN network used for extracting image features (the Resnet50 network is used in the invention), the cyclic network layer is a BI-LSTM (double-layer BI-LSTM is used for stack in the invention), the aim of continuously extracting character sequence features after convolution is achieved, and the output layer outputs the RNN as softmax and then outputs characters.

CTC (connectionist Temporal classification), the CRNN result obtained above is sent to a CTC algorithm model, which is a complete end-to-end model training, and only one input sequence and one output sequence are required for training without aligning data in advance. This eliminates the need for data alignment and labeling one-by-one, and the CTC outputs the probability of sequence prediction directly, without the need for external post-processing. CTC introduces a blank (no predicted value at this time), and each predicted class corresponds to a spike (peak) in the whole text, and other locations that are not peaks are considered blank. For a piece of text, the final output of a CTC is a sequence of spikes, and does not care how long each word lasts. CTC is by gradientThe ω -parameter of the LSTM is adjusted so that p (l | x) is maximized. (p (l | x) is the probability of the output l given the input x, ω is the LSTM hidden layer parameter). This results in processed text information.

And automatically and correspondingly filling the text result obtained by artificial intelligence image recognition into the itemized form in the information entry module.

Confirming whether filling has errors on the form interactive page, and confirming that the filling can be submitted by clicking 'submission approval' through the interactive interface; if there is error, it can be modified by itself.

And after the project declaration plan is submitted, updating the declaration state to be 'in declaration', simultaneously sending a message in the audit station, and displaying that the project is submitted in a longitudinal project list submodule of the project setting management module.

In the event of an operation in the project declaration plan, the operation of canceling, printing, or the like can be performed on the project to which the declaration is submitted

If the audit is passed, the state displayed on the interface of the longitudinal project list submodule of the project establishment management module is 'audit passed', and meanwhile, the message in the station is sent to the applicant and the WeChat applet information is pushed; and if the audit is not passed, displaying the status of 'failure' on an interface of a longitudinal project list submodule of the project establishment management module, and simultaneously sending the message in the station and pushing the WeChat applet information to the applicant.

Although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.