A character recognition method and system for military scientific research and production data

Abstract

The embodiment of the present invention provides a character recognition method for military scientific research and production data, including: obtaining military scientific research and production data after preprocessing, using text detection technology to frame the text area of ​​the image, and outputting the initial recognition result and recognition accuracy of each character after character recognition in the text area; obtaining the initial image of the text area corresponding to the extracted character, and distinguishing it through the initial image of the preset recognition module based on the preset discrimination standard; if the discrimination result of the preset discrimination module is consistent with the initial recognition result, then the initial recognition result is determined to be correct; if the discrimination result of the preset discrimination module is inconsistent with the initial recognition result, then the initial recognition result is determined to be wrong; the extraction probability of the extracted character is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors. Combined with the document recognition requirements and the recognition accuracy of the characters, the characters are randomly extracted for recognition to improve the proofreading efficiency.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a character recognition method and system for military scientific research and production data. Background Art

[0002] In the long-term military scientific research and production process, a large number of paper documents have been accumulated, and it is urgent to rescue the historical data and use automated methods to quickly identify paper documents and form electronic data. Since military scientific research and production are extremely demanding work, the currently commonly used OCR (Optical Character Recognition) technology is difficult to meet the accuracy requirements.

[0003] At present, OCR technology in general fields is relatively mature. For example, Google OCR, Baidu OCR, Alibaba OCR, Hanwang OCR, etc. all provide convenient API interfaces, and users can directly call them to implement OCR recognition functions. In addition, some companies have also carried out targeted optimization for vertical business, such as bill recognition, ID card recognition, license plate number recognition, etc.

[0004] At present, OCR technology is relatively mature. Calling the corresponding API interface can conveniently realize the function of character recognition. However, OCR technology only supports direct calling, and users cannot directly view the recognition accuracy of each character. At the same time, the recognition accuracy of existing technology is not high. For the military scientific research and production fields, after the machine recognition, professionals are required to proofread the text again, which is time-consuming and labor-intensive, and the proofreading results cannot be reused. In addition, the proofreading results are difficult to feed back to the training process, resulting in a lack of data in the training process and difficulty in improving the recognition accuracy. Summary of the invention

[0005] The embodiments of the present invention provide a method and system for character recognition for military scientific research and production data, which can improve the accuracy of character recognition and facilitate the sorting of military scientific research and production data.

[0006] The embodiment of the present invention provides a character recognition method for military scientific research and production data, including:

[0007] Obtain pre-processed military scientific research and production data, use text detection technology to select the text area of ​​the image, perform character recognition on the text area, and output the initial recognition result and recognition accuracy of each character;

[0008] Acquire an initial image of the text area corresponding to the extracted character, and perform discrimination on the initial image through a preset discrimination module based on a preset discrimination criterion;

[0009] If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct;

[0010] If the identification result of the preset identification module is inconsistent with the initial identification result, it is determined that the initial identification result is wrong;

[0011] The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors.

[0012] In some embodiments of the present invention, the initial image is discriminated by a preset discrimination module based on a preset discrimination criterion, including:

[0013] Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively;

[0014] If the first discrimination results of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result;

[0015] If they are consistent, it is determined that the initial recognition result is correct;

[0016] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0017] In some embodiments of the present invention, if the first discrimination results of the first discrimination module and the first discrimination results of the second discrimination module are inconsistent, then,

[0018] Based on the secondary discrimination standard, the initial image is discriminated by a third discrimination module and a fourth discrimination module respectively;

[0019] If the second discrimination results of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result;

[0020] If they are consistent, it is determined that the initial recognition result is correct;

[0021] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0022] In some embodiments of the present invention, if the second discrimination results of the third discrimination module and the fourth discrimination module are inconsistent, then,

[0023] Determine a third discrimination standard based on the field category and research direction category corresponding to the military scientific research and production data, and discriminate the initial image through a fifth discrimination module based on the third discrimination standard;

[0024] Comparing the third discrimination result of the fifth discrimination module with the initial recognition result;

[0025] If they are consistent, it is determined that the initial recognition result is correct;

[0026] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0027] In some embodiments of the present invention, the extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the recognition error frequency, specifically:

[0028]

[0029] Among them, p i is the extraction probability;

[0030] d is the document identification requirement, expressed as a percentage;

[0031] n i Represents the i-th character, i is the serial number, counting starts from 1;

[0032] a i is the recognition accuracy, expressed as a percentage;

[0033] e is the recognition error frequency, which is initially 1.

[0034] In some embodiments of the present invention, the recognition accuracy of the secondary discrimination criterion is higher than the recognition accuracy of the primary discrimination criterion;

[0035] The recognition accuracy of the third-order discrimination standard is higher than the recognition accuracy of the second-order discrimination standard.

[0036] In some embodiments of the present invention, the recognition accuracy of the third discrimination module and the fourth discrimination module is higher than the recognition accuracy of the first discrimination module and the second discrimination module;

[0037] The recognition accuracy of the fifth discrimination module is higher than the recognition accuracy of the third discrimination module and the fourth discrimination module.

[0038] The embodiment of the present invention also provides a character recognition system for military scientific research and production data, including:

[0039] A character recognition module is used to select the text area of ​​the image based on the pre-processed military scientific research and production data using text detection technology, and output the initial recognition result and recognition accuracy of each character after character recognition in the text area;

[0040] A preset discrimination module, which is used to discriminate the initial image of the text area corresponding to the extracted characters obtained by the recognition result extraction module based on a preset discrimination standard;

[0041] If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct;

[0042] If the identification result of the preset identification module is inconsistent with the initial identification result, it is determined that the initial identification result is wrong;

[0043] The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors.

[0044] In some embodiments of the present invention, the preset identification module is further used to:

[0045] Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively;

[0046] If the first discrimination results of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result;

[0047] If they are consistent, it is determined that the initial recognition result is correct;

[0048] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0049] In some embodiments of the present invention, the preset identification module is further used to:

[0050] If the first discrimination results of the first discrimination module and the second discrimination module are inconsistent, then,

[0051] Based on the secondary discrimination standard, the initial image is discriminated by a third discrimination module and a fourth discrimination module respectively;

[0052] If the second discrimination results of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result;

[0053] If they are consistent, it is determined that the initial recognition result is correct;

[0054] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0055] The character recognition method and system for military scientific research and production data provided by the embodiment of the present invention have the following advantages: it combines the document recognition requirements and the recognition accuracy of each character, randomly extracts characters for recognition, and dynamically adjusts the recognition error rate, thereby improving the proofreading efficiency. At the same time, by comparing the discrimination results of multiple discrimination modules, the existing manual proofreading process is optimized, ensuring the accuracy of recognition and proofreading. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flowchart of a character recognition method for military scientific research and production data according to an embodiment of the present invention;

[0057] Figure 2 The figure is a flow chart of a character recognition method for military scientific research and production data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with the accompanying drawings and specific implementation methods.

[0059] The phrases "in one embodiment", "in another embodiment", "in yet another embodiment", "in an embodiment", "in some embodiments" or "in other embodiments" may be used in this specification to refer to one or more of the same or different embodiments according to the present invention.

[0060] Specific embodiments of the present invention are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments invented are merely embodiments of the present invention, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to determine the true intent based on the user's historical operations and to avoid unnecessary or redundant details that obscure the present invention. Therefore, the specific structural and functional details of the present invention are not intended to be limiting, but only serve as the basis and representative basis for the claims to teach those skilled in the art to use the present invention in a variety of substantially any appropriate detailed structures.

[0061] The embodiment of the present invention provides a character recognition method for military scientific research and production data, such as Figure 1 and Figure 2 As shown, the following steps are included:

[0062] Step 1: Obtain military scientific research and production data after preprocessing, use text detection technology to select the text area of ​​the image, perform character recognition on the text area, and output the initial recognition result and recognition accuracy of each character. Specifically, military scientific research and production refers to scientific research, technological development, technological transformation, scientific and technological innovation, new product development, model spectrum development, model development, finalized production and test platform construction for the purpose of national defense construction; character recognition can be optical character recognition, that is, OCR (Optical Character Recognition), which refers to the process of analyzing and identifying the image file of text data to obtain text and layout information, that is, recognizing the text in the image and returning it in the form of text.

[0063] In this embodiment, the military scientific research and production data can be preprocessed first, the text detection technology is used to select the text area of ​​the image, the OCR engine Tesseract is used for character recognition, and the recognition accuracy of each character is output. In the actual operation process, human-computer collaborative recognition can be performed, and the results of recognition errors can be collected in real time. The results of recognition errors are regularly imported and saved, such as imported into the character recognition module, and training tasks are carried out to improve the accuracy of character recognition. Among them, human-computer collaboration refers to the use of a reasonable mechanism for domain experts and computers to collaborate to complete specific tasks, combining the respective advantages of machines and humans to improve the quality of task completion.

[0064] Step 2: Obtain an initial image of the text area corresponding to the extracted characters, and perform discrimination on the initial image through a preset discrimination module based on a preset discrimination standard, wherein the discrimination on the initial image through the preset discrimination module may include multiple discriminations on the initial image, so as to further improve the discrimination accuracy by comparing the multiple discrimination results. Specifically, when discriminating the initial image based on the preset discrimination standard, a method of combining experts to participate may be adopted to improve the discrimination accuracy. As an example, ordinary experts may be randomly extracted from the expert database by a machine.

[0065] Step 3: If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct;

[0066] Step 4: If the identification result of the preset identification module is inconsistent with the initial identification result, then determining that the initial identification result is wrong;

[0067] The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the recognition error frequency. Specifically, the extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the recognition error frequency. The formula is:

[0068]

[0069] Among them, p i is the extraction probability;

[0070] d is the document identification requirement, expressed as a percentage;

[0071] n i Represents the i-th character, i is the serial number, counting starts from 1;

[0072] a i is the recognition accuracy, expressed as a percentage;

[0073] e is the recognition error frequency, which is initially 1.

[0074] In some embodiments of the present invention, the initial image is discriminated by a preset discrimination module based on a preset discrimination criterion, including:

[0075] Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively. In the actual discrimination process, the first discrimination module and the second discrimination module can be image recognition modules with a first recognition capability (such as resolution, clarity, etc.), and can be combined with experts for manual discrimination, such as combining two ordinary experts to perform discrimination based on the original image, wherein the ordinary experts can be randomly selected from the corresponding expert database.

[0076] If the first discrimination results (which may be B1 and B2 respectively) of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result (which may be set to B);

[0077] If they are consistent, it is determined that the initial recognition result is correct;

[0078] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0079] Further, in this embodiment, if the first discrimination results of the first discrimination module and the first discrimination results of the second discrimination module are inconsistent (ie, B1 and B2 are inconsistent), then,

[0080] Based on the secondary discrimination standard, the initial image is discriminated by the third discrimination module and the fourth discrimination module respectively. At this time, two other ordinary experts can also be combined to discriminate based on the original image. Similarly, the ordinary experts can be randomly selected from the corresponding expert database;

[0081] If the second discrimination results (which may be C1 and C2 respectively) of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result B;

[0082] If they are consistent, it is determined that the initial recognition result is correct;

[0083] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0084] In this embodiment, if the second discrimination results of the third discrimination module and the fourth discrimination module are inconsistent (ie, C1 and C2 are inconsistent), then,

[0085] Based on the field categories and research direction categories corresponding to the military scientific research and production data, three discrimination criteria are determined, and based on the three discrimination criteria, the initial image is discriminated by a fifth discrimination module. At this time, it can be combined with senior experts to discriminate based on the original image, and the senior experts can determine based on the field categories and research direction categories corresponding to the military scientific research and production data;

[0086] Comparing the third discrimination result D of the fifth discrimination module with the initial recognition result B;

[0087] If they are consistent, it is determined that the initial recognition result is correct;

[0088] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0089] The above technical solution combines the document recognition requirements and the recognition accuracy of each character, randomly selects characters for recognition, and dynamically adjusts the recognition error rate to improve the proofreading efficiency. At the same time, by comparing the recognition results of multiple recognition modules, the existing manual proofreading process is optimized to ensure the accuracy of recognition and proofreading.

[0090] In some embodiments of the present invention, the recognition accuracy of the secondary discrimination criterion is higher than the recognition accuracy of the primary discrimination criterion;

[0091] The recognition accuracy of the third-order discrimination standard is higher than the recognition accuracy of the second-order discrimination standard.

[0092] In some embodiments of the present invention, the recognition accuracy of the third discrimination module and the fourth discrimination module is higher than the recognition accuracy of the first discrimination module and the second discrimination module;

[0093] The recognition accuracy of the fifth discrimination module is higher than the recognition accuracy of the third discrimination module and the fourth discrimination module.

[0094] The embodiment of the present invention also provides a character recognition system for military scientific research and production data, including:

[0095] A character recognition module is used to select the text area of ​​the image based on the pre-processed military scientific research and production data using text detection technology, and output the initial recognition result and recognition accuracy of each character after character recognition in the text area;

[0096] A preset discrimination module, which is used to discriminate the initial image of the text area corresponding to the extracted characters obtained by the recognition result extraction module based on a preset discrimination standard;

[0097] If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct;

[0098] If the identification result of the preset identification module is inconsistent with the initial identification result, it is determined that the initial identification result is wrong;

[0099] The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors.

[0100] In some embodiments of the present invention, the preset identification module is further used to:

[0101] Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively;

[0102] If the first discrimination results of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result;

[0103] If they are consistent, it is determined that the initial recognition result is correct;

[0104] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0105] In some embodiments of the present invention, the preset identification module is further used to:

[0106] If the first discrimination results of the first discrimination module and the second discrimination module are inconsistent, then,

[0107] Based on the secondary discrimination standard, the initial image is discriminated by a third discrimination module and a fourth discrimination module respectively;

[0108] If the second discrimination results of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result;

[0109] If they are consistent, it is determined that the initial recognition result is correct;

[0110] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0111] Further, if the second discrimination results of the third discrimination module and the fourth discrimination module are inconsistent, then,

[0112] Determine a third discrimination standard based on the field category and research direction category corresponding to the military scientific research and production data, and discriminate the initial image through a fifth discrimination module based on the third discrimination standard;

[0113] Comparing the third discrimination result of the fifth discrimination module with the initial recognition result;

[0114] If they are consistent, it is determined that the initial recognition result is correct;

[0115] If they are inconsistent, it is determined that the initial recognition result is wrong.

[0116] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.

Claims

1. A character recognition method for military scientific research and production data, characterized in that: include: Obtain pre-processed military scientific research and production data, use text detection technology to select the text area of ​​the image, perform character recognition on the text area, and output the initial recognition result and recognition accuracy of each character; Acquire an initial image of the text area corresponding to the extracted character, and perform discrimination on the initial image through a preset discrimination module based on a preset discrimination criterion; If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct; If the identification result of the preset identification module is inconsistent with the initial identification result, it is determined that the initial identification result is wrong; The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors; The step of performing discrimination based on a preset discrimination standard and using a preset discrimination module on the initial image includes: Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively; If the first discrimination results of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result; If they are consistent, it is determined that the initial recognition result is correct; If they are inconsistent, it is determined that the initial recognition result is wrong; If the first discrimination results of the first discrimination module and the second discrimination module are inconsistent, then, Based on the secondary discrimination standard, the initial image is discriminated by a third discrimination module and a fourth discrimination module respectively; If the second discrimination results of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result; If they are consistent, it is determined that the initial recognition result is correct; If they are inconsistent, it is determined that the initial recognition result is wrong.

2. The character recognition method for military scientific research and production data according to claim 1 is characterized in that: If the second discrimination results of the third discrimination module and the fourth discrimination module are inconsistent, then, Determine a third discrimination standard based on the field category and research direction category corresponding to the military scientific research and production data, and discriminate the initial image through a fifth discrimination module based on the third discrimination standard; Comparing the third discrimination result of the fifth discrimination module with the initial recognition result; If they are consistent, it is determined that the initial recognition result is correct; If they are inconsistent, it is determined that the initial recognition result is wrong.

3. The character recognition method for military scientific research and production data according to claim 2 is characterized in that: The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the recognition error frequency, specifically: Among them, pi is the extraction probability; d is the document recognition requirement, expressed in percentage; ni represents the i-th character, i is the sequence number, counting from 1; ai is the recognition accuracy, expressed in percentage; e is the recognition error frequency, initially 1.

4. The character recognition method for military scientific research and production data according to claim 3 is characterized in that: The recognition accuracy of the secondary discrimination standard is higher than the recognition accuracy of the primary discrimination standard; The recognition accuracy of the third-order discrimination standard is higher than the recognition accuracy of the second-order discrimination standard.

5. The character recognition method for military scientific research and production data according to claim 4 is characterized in that: The recognition accuracy of the third discrimination module and the fourth discrimination module is higher than the recognition accuracy of the first discrimination module and the second discrimination module; The recognition accuracy of the fifth discrimination module is higher than the recognition accuracy of the third discrimination module and the fourth discrimination module.

6. A character recognition system for military scientific research and production data, characterized in that: include: A character recognition module is used to select the text area of ​​the image based on the pre-processed military scientific research and production data using text detection technology, and output the initial recognition result and recognition accuracy of each character after character recognition in the text area; A preset discrimination module, which is used to discriminate the initial image of the text area corresponding to the extracted characters obtained by the recognition result extraction module based on a preset discrimination standard; If the identification result of the preset identification module is consistent with the initial identification result, then the initial identification result is determined to be correct; If the identification result of the preset identification module is inconsistent with the initial identification result, it is determined that the initial identification result is wrong; The extraction probability of the extracted characters is determined based on the document recognition requirements, the number of characters, the recognition accuracy of each character, and the frequency of recognition errors; The preset identification module is also used for: Based on a primary discrimination criterion, the initial image is discriminated by a first discrimination module and a second discrimination module respectively; If the first discrimination results of the first discrimination module and the second discrimination module are consistent, then the first discrimination result is compared with the initial recognition result; If they are consistent, it is determined that the initial recognition result is correct; If they are inconsistent, it is determined that the initial recognition result is wrong; The preset identification module is also used for: If the first discrimination results of the first discrimination module and the second discrimination module are inconsistent, then, Based on the secondary discrimination standard, the initial image is discriminated by a third discrimination module and a fourth discrimination module respectively; If the second discrimination results of the third discrimination module and the fourth discrimination module are consistent, then the second discrimination result is compared with the initial recognition result; If they are consistent, it is determined that the initial recognition result is correct; If they are inconsistent, it is determined that the initial recognition result is wrong.

Images