An image classification and recognition system based on multi-cooperative neural networks

Abstract

The present invention relates to the technical field of image classification and recognition, and in particular to an image classification and recognition system based on a multi-cooperative neural network, comprising: scanning original data, and recognizing pictures of the original data, determining users corresponding to the editing authority of each original data, and sorting out data with the same editing authority, identifying similar graphics in the user's editing pool, and fitting the similar graphics, recording the text corresponding to the similar graphics in each text data, correcting the fitted text using a semantic recognition strategy, and forming a record text for unrecognizable content, and storing the record text in a database for further recognition. The present invention forms the writing features corresponding to the user by performing font recognition on the original document, normalizing the original document according to the writing features, and correcting the normalized document using semantic recognition, thereby effectively improving the stability of text maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of image classification and recognition, and in particular to an image classification and recognition system based on multiple collaborative neural networks. Background Art

[0002] With the development of science and technology, the preservation form of text materials has also undergone tremendous changes. For paper documents that have been preserved for a long time, due to the limitations of printing technology and material processing, a large amount of text materials in paper documents are lost. Moreover, after scanning, due to the backward recognition technology, the actual recorded text is quite different from the stored text.

[0003] Chinese Patent Authorization Announcement No.: CN117454957B discloses an inference training system for an image processing neural network model, including a model access module, a model parsing module, a training monitoring module, an inference parsing module and a model feedback module, wherein the model access module is used to connect with an image processing neural network model, the model parsing module is used to parse the layout of neuron nodes in the image processing neural network model, the training monitoring module is used to monitor the parameter changes of neuron nodes in the image processing neural network model during the training process, the inference parsing module is used to perform inference analysis on the parameter changes, and the model feedback module is used to feed back the inference analysis results to the image processing neural network model; the invention accelerates the learning speed of the image processing neural network model and improves the problem of wasting computer resources and poor real-time computing.

[0004] However, the above method has the following problems: it is unable to effectively identify erroneous information in the text, resulting in the inability to effectively maintain the text data. Summary of the invention

[0005] To this end, the present invention provides an image classification and recognition system and system based on a multi-cooperative neural network to overcome the problem in the prior art that erroneous information in the text cannot be effectively identified, resulting in the inability to effectively maintain the text data, thereby reducing the maintenance stability of text management.

[0006] To achieve the above object, the present invention provides an image classification and recognition system based on a multi-cooperative neural network, comprising:

[0007] A scanning module, which is used to scan the original data to form corresponding image data;

[0008] A classification module, which is connected to the scanning module, is used to determine the editing authority of the corresponding user according to the editing authority of the user on the original data, and classify the image data into a corresponding classification pool;

[0009] An identification module, which is connected to the classification module, and is used to identify and correct each image data according to the classification pool to form corresponding record data;

[0010] A storage module, connected to the identification module, for recording the recorded data;

[0011] A timing module, connected to the scanning module and the storage module, for maintaining the system after a preset time has passed;

[0012] Wherein, the preset time duration is related to the damage cycle of the original data;

[0013] When establishing the classification pool, the classification module also sorts the data with the same authority according to the user corresponding to the editing authority of each original data to form a feature pool for the user;

[0014] For a single user, the recognition module is provided with an editing pool of the user, the recognition module recognizes similar graphics in the editing pool of the user, and fits the similar graphics to form a writing fitting pool of the user.

[0015] The recognition module also uses a neural network to record the text corresponding to the similar graphics in each text data to form a fitting text, and uses a plurality of fitting texts to coordinate with the fitting text to perform deviation correction, and forms a record text for unrecognizable content.

[0016] Further, the classification module verifies the permissions of the user to classify the permissions of the user;

[0017] Among them, when classifying the user's authority, the text data is classified according to the user's authority.

[0018] If the user has the permission to write, the text data is stored in the writing pool of the user;

[0019] If the user has editing authority, the text data is stored in the editing pool of the user;

[0020] The classification module combines the writing pool and the editing pool of the corresponding user to form a feature pool of the user.

[0021] Furthermore, the recognition module determines the recognition characters of the user according to the content of each text material in the writing pool corresponding to the single user;

[0022] Wherein, when determining the recognition characters, a preset reference number of text materials with a text volume greater than a text volume threshold is set as a standard file, and the characters in each standard file are respectively used as the recognition characters;

[0023] The text volume is the number of characters in the text material, and the text volume threshold is related to the average text volume of each text material in the writing pool;

[0024] Wherein, the recognition characters correspond to pictures and text information.

[0025] Further, for the single recognized character, when the recognition module recognizes the character, if the text category corresponding to the character is greater than one, the character is marked as an abnormal character, and the character is fuzzy recognized to correspond to a single text, and the recognized characters and the corresponding texts are combined into the writing fitting pool;

[0026] The text categories include characters and character groups.

[0027] Further, for the single user, when establishing the handwriting fitting pool corresponding to the user, the recognition module sets a fuzziness rate corresponding to the fuzzy recognition, outputs the text corresponding to each recognized character whose fitting degree exceeds a preset recognition rate according to the fuzziness rate, and forms a corresponding fitting text;

[0028] Wherein, the preset recognition rate is related to the blur rate corresponding to each character in the fitting text;

[0029] The fitting text includes text materials in the writing pool, but does not include text materials in the editing pool.

[0030] Furthermore, for the single fitting text, the recognition module also performs semantic recognition on the fitting text. If the fitting text contains any unrecognizable text content, the fitting text is corrected in a preset correction manner to form a record text and a corresponding correction record.

[0031] Among them, the preset correction method is to correct the deviation through customary expressions.

[0032] Furthermore, when the preset maintenance time has passed, the timing module adjusts the text volume threshold and the fuzziness rate according to the writing pool of the single user, and identifies each of the original materials stored in the preset maintenance time according to the text volume threshold and the fuzziness rate to form a pre-recorded text.

[0033] Furthermore, when the recognition module recognizes the recorded text, if there is a difference between the recognition information of any of the recognized characters in the recorded text and the corresponding text in the pre-recorded text, the recognized character is determined to be incorrect, and the incorrect information is alerted to the corresponding user for correction by the user.

[0034] Furthermore, when collecting the original data, the scanning module divides the original data into several sections, and identifies the pictures corresponding to each section to form corresponding text data;

[0035] Among them, a single area corresponds to a single text document.

[0036] Furthermore, when the timing module identifies the original data stored in the preset maintenance duration, its sample does not include the record text.

[0037] Compared with the prior art, the beneficial effect of the present invention lies in that, by performing font recognition on the original document, the writing features corresponding to the user are formed, the original document is normalized according to the writing features, and the normalized document is corrected by using semantic recognition, thereby effectively improving the stability of text maintenance.

[0038] Furthermore, by determining the users corresponding to the editing and writing permissions of each original document, text management is made more secure and orderly. The unified permission document organization can form a feature pool for specific users, improve the efficiency of text management and use, and further improve maintenance stability.

[0039] Furthermore, by identifying similar graphics in the user editing pool and fitting them, the user's writing style and characteristics can be abstracted, providing a basis for subsequent text recognition and error correction, further improving maintenance stability.

[0040] Furthermore, by recording the text corresponding to similar graphics in each text material, a fitting text is formed, which can improve the readability and usage efficiency of the text and further improve the maintenance stability.

[0041] Furthermore, the fitted text is corrected through semantic recognition strategies, and unrecognizable content is recorded, thereby improving the accuracy and usability of the text and further improving maintenance stability.

[0042] Furthermore, the recorded text will be stored in the database and processed again after the preset maintenance time, so as to improve the maintainability and scalability of the text. The data that exceeds the storage time threshold will be sealed, so that the storage resources can be used rationally, and the maintenance stability will be further improved.

[0043] Furthermore, by extracting text data from the writing pool and determining the user's recognition characters, we can effectively improve our understanding of the user's writing style and habits, improve the accuracy of text recognition, and further improve maintenance stability.

[0044] Furthermore, fuzzy recognition of abnormal characters can improve the recognition ability of non-standard texts, enhance the robustness of the system, and further improve maintenance stability.

[0045] Furthermore, by setting the fuzziness rate, the result of fuzzy recognition can be adjusted to make the fitting text more consistent with the actual situation, thereby improving the recognition accuracy and further improving the maintenance stability.

[0046] Furthermore, by adjusting the text volume threshold and fuzziness rate for identifying stored text data, dynamic management of text data can be achieved, the flexibility of text management can be improved, and maintenance stability can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a structural block diagram of the image classification and recognition system based on multiple collaborative neural networks of the present invention;

[0048] Figure 2 It is a flow chart of the image classification and recognition system based on multiple collaborative neural networks of the present invention;

[0049] Figure 3 Schematic diagram of identifying font blur rate according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0052] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0053] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0054] See also Figure 1 As shown, it is a structural block diagram of the image classification and recognition system based on multiple collaborative neural networks of the present invention, and the image classification and recognition system based on multiple collaborative neural networks includes:

[0055] A scanning module, which is used to scan the original data to form corresponding image data;

[0056] A classification module, which is connected to the scanning module, is used to determine the editing authority of the corresponding user according to the editing authority of the user on the original data, and classify the image data into a corresponding classification pool;

[0057] An identification module, which is connected to the classification module, and is used to identify and correct each image data according to the classification pool to form corresponding record data;

[0058] A storage module, connected to the identification module, for recording the recorded data;

[0059] A timing module, connected to the scanning module and the storage module, for maintaining the system after a preset time has passed;

[0060] Wherein, the preset time duration is related to the damage cycle of the original data;

[0061] When establishing a classification pool, the classification module also sorts out the data with the same permissions according to the user corresponding to the editing permissions of each original data to form a feature pool for the user;

[0062] For a single user, the recognition module is provided with the editing pool of the user. The recognition module recognizes similar graphics in the editing pool of the user and fits the similar graphics to form a writing fitting pool of the user.

[0063] The recognition module also uses a neural network to record the text corresponding to similar graphics in each text data to form a fitting text, and uses several fitting texts to collaboratively fit the text for correction, and forms a recorded text for unrecognizable content.

[0064] In practice, paper documents have a damage cycle, such as 5 years or 15 years, due to material and storage limitations. Identifying and storing them as images within the damage cycle can better maintain the data.

[0065] Specifically, the classification module verifies the user's permissions to classify the user's permissions;

[0066] Among them, when classifying the user's authority, the text data is classified according to the user's authority.

[0067] If the user's permission is writing permission, the text data is stored in the user's writing pool;

[0068] If the user's permission is editing permission, the text data is stored in the user's editing pool;

[0069] The classification module combines the writing pool and the editing pool of the corresponding user to form a feature pool of the user.

[0070] Specifically, the recognition module determines the recognition characters of a single user according to the content of each text material in the writing pool corresponding to the single user;

[0071] When determining the recognition characters, a preset reference amount of text data with a text volume greater than a text volume threshold is set as a standard file, and the characters in each standard file are respectively used as the recognition characters;

[0072] The text volume refers to the number of characters in the text material, and the text volume threshold is related to the average text volume of each text material in the writing pool;

[0073] Among them, the recognized characters correspond to the pictures and text information.

[0074] Specifically, for a single recognized character, when the recognition module recognizes the character, if the text category corresponding to the character is greater than one, the character is marked as an abnormal character, and the character is fuzzy recognized to correspond to a single text, and each recognized character and the corresponding text are combined into a writing fitting pool;

[0075] The text categories include characters and character groups.

[0076] Specifically, for a single user, when establishing the writing fitting pool corresponding to the user, the recognition module sets the fuzziness rate of the corresponding fuzzy recognition, outputs the text corresponding to each recognized character whose fitting degree exceeds the preset recognition rate according to the fuzziness rate, and forms the corresponding fitting text;

[0077] Among them, the preset recognition rate is related to the fuzziness rate corresponding to each character in the fitting text;

[0078] The fitting text includes the text data in the writing pool, but does not include the text data in the editing pool. Specifically, for a single fitting text, the recognition module also performs semantic recognition on the fitting text. If the fitting text contains any unrecognizable text content, the fitting text is corrected in a preset correction method to form a record text and a corresponding correction record;

[0079] Among them, the preset correction method is to correct the error through idiomatic expressions.

[0080] Specifically, when the preset maintenance time has passed, the timing module adjusts the text volume threshold and the fuzziness rate according to the writing pool of a single user, and identifies each original material stored in the preset maintenance time according to the text volume threshold and the fuzziness rate to form a pre-recorded text.

[0081] Specifically, when the recognition module recognizes the recorded text, if there is a difference between the recognition information of any recognized character in the recorded text and the corresponding text in the pre-recorded text, the recognized character is determined to be incorrect, and the incorrect information is alerted to the corresponding user for correction by the user.

[0082] Specifically, when collecting the original data, the scanning module divides the original data into several sections and recognizes the pictures corresponding to each section to form corresponding text data;

[0083] Among them, a single area corresponds to a single text document.

[0084] Specifically, when the timing module identifies the original data stored in the preset maintenance time, its sample does not include the recorded text.

[0085] See also Figure 2 As shown, it is a flow chart of the image classification and recognition system based on multiple collaborative neural networks of the present invention, and the image classification and recognition system based on multiple collaborative neural networks includes:

[0086] Step S1, scanning the original data and identifying the image of the original data;

[0087] Step S2, determining the user corresponding to the editing authority of each original document, and sorting the documents with the same authority to form a feature pool of the user;

[0088] Step S3, identifying similar graphics in the user's editing pool, and fitting the similar graphics to form the user's writing fitting pool;

[0089] Step S4, recording the text corresponding to the similar graphics in each text data to form a fitting text;

[0090] Step S5, using a semantic recognition strategy to correct the fitted text, and forming a recorded text with the unrecognizable content;

[0091] Step S6, storing the recorded text in a database, and after a preset maintenance time, repeating steps S2 to S5, and sealing the data that exceeds the storage time threshold.

[0092] The present invention forms the writing features corresponding to the user by performing font recognition on the original document, normalizes the original document according to the writing features, and corrects the normalized document by using semantic recognition, thereby effectively improving the stability of text maintenance.

[0093] Specifically, in step S2, the maintenance method verifies the user's authority, including:

[0094] Step S21, confirming the user's authority;

[0095] Step S22, judging whether the user is the author of the text material according to the authority;

[0096] Step S23, further distinguishing the text data according to the user's authority, and classifying the text data according to the user's authority, including:

[0097] Step S231, if the user's permission is writing permission, the text data is stored in the user's writing pool;

[0098] Step S232, if the user's authority is editing authority, the text data is stored in the user's editing pool;

[0099] Step S24: combining the writing pool and the editing pool of the corresponding user to form a feature pool of the user.

[0100] By determining the users with editing and writing permissions for each original document, text management becomes more secure and orderly. Unified permission document organization can form a feature pool for specific users, improve the efficiency of text management and use, and further improve maintenance stability.

[0101] Taking medical text as an example, in implementation, for users who write text materials, they write more text in the corresponding text materials, and for users who can edit text materials, they are in the process of revising or modifying the text materials, and they write less text in the text materials;

[0102] It is understandable that, for a single user, it is difficult to infer the user's writing habits based on text materials with less written text.

[0103] By determining the users with editing and writing permissions for each original document, text management becomes more secure and orderly. Unified permission document organization can form a feature pool for specific users, improve the efficiency of text management and use, and further improve maintenance stability.

[0104] Specifically, in step S3, the identification characters of a single user are determined according to the content of each text material in the writing pool corresponding to the single user;

[0105] When determining the recognition characters, a preset reference amount of text data with a text volume greater than a text volume threshold is set as a standard file, and the characters in each standard file are respectively used as the recognition characters;

[0106] The text volume refers to the number of characters in the text material, and the text volume threshold is related to the average text volume of each text material in the writing pool;

[0107] Among them, the preset reference number can be set arbitrarily, as long as it is not less than 1.

[0108] Among them, the recognized characters correspond to the pictures and text information.

[0109] In the implementation, take user A as an example:

[0110] The corresponding writing pool includes 5 text materials, among which the amount of text corresponding to each material is:

[0111] 60, 200, 300, 120, 700, the corresponding average text volume is 276, the preset reference number is set to 3, then the text volume threshold of user A is 276, and the text volume greater than this value is 300, 700, that is, the text materials with a text volume of 300, 700 are used as standard files;

[0112] Take user B as an example:

[0113] The corresponding writing pool includes 5 text materials, among which the amount of text corresponding to each material is:

[0114] 200, 200, 200, 200, 220, the corresponding average text volume is 204, the preset reference quantity is set to 1, then the text volume threshold of user B is 204, and the text volume greater than this value is 220, that is, the text material with a text volume of 220 is used as the standard file;

[0115] Take user C as an example:

[0116] The corresponding writing pool includes 3 text materials, among which the amount of text corresponding to each material is:

[0117] 60, 60, 60. The corresponding average text volume is 60. The text volume threshold of user B is 60. The preset reference quantity is set to 2. Any two texts of this user are used as standard files.

[0118] Specifically, in step S3, for a single recognized character, if the text category corresponding to the character is greater than one when the character is recognized, the character is marked as an abnormal character, and the character is fuzzy recognized to correspond to a single text, and each recognized character and the corresponding text are combined into a writing fitting pool;

[0119] The text categories include characters and character groups.

[0120] By identifying and fitting similar graphics in the user editing pool, the user's writing style and characteristics can be abstracted, providing a basis for subsequent text recognition and error correction, and further improving maintenance stability.

[0121] By recording the text corresponding to similar graphics in each text material to form a fitting text, the readability and usage efficiency of the text can be improved, and the maintenance stability is further improved.

[0122] Specifically, in step S4, for a single user, when establishing the writing fitting pool corresponding to the user, the fuzziness rate of the corresponding fuzzy recognition is set, and the text corresponding to each recognized character whose fitting degree exceeds the preset recognition rate is output according to the fuzziness rate, and the corresponding fitting text is formed;

[0123] Among them, the preset recognition rate is related to the fuzziness rate corresponding to each character in the fitting text;

[0124] The fitting text includes the text materials in the writing pool, but does not include the text materials in the editing pool.

[0125] In implementation, for the text in the editing pool, since the amount of text corresponding to it is small, the fitting text should not select the text data in the editing pool.

[0126] By extracting text data from the writing pool and determining the user's recognition characters, we can effectively improve our understanding of the user's writing style and habits, improve the accuracy of text recognition, and further improve maintenance stability.

[0127] Specifically, in step S5, for a single fitting text, when semantic recognition is performed on the fitting text, if there is any unrecognizable text, the text is corrected in a preset correction method to form a record text and a corresponding correction record;

[0128] Among them, the preset correction method is to correct the error through idiomatic expressions.

[0129] It is understandable that idiomatic expressions should conform to industry norms and spoken language usage, which will not be elaborated here.

[0130] By means of a semantic recognition strategy, the fitted text is corrected, and the content that cannot be recognized is recorded, thereby improving the accuracy and usability of the text and further enhancing the maintenance stability.

[0131] Specifically, in step S6, when the preset maintenance duration is reached, the text volume threshold and the fuzziness rate are adjusted according to the writing pool of a single user, and each original material stored during the preset maintenance duration is recognized according to the text volume threshold and the fuzziness rate to form a pre-recorded text;

[0132] Among them, when recognizing the original materials stored during the preset maintenance duration, the sample does not include the recorded text.

[0133] Performing fuzzy recognition on abnormal characters can improve the recognition ability for non-standard texts, enhance the robustness of the system, and further improve the maintenance stability.

[0134] Specifically, in step S6, if there is a difference between the recognition information of any recognized character in the recorded text and the corresponding text in the pre-recorded text, it is determined that the recognized character is incorrect, and the error information is alerted to the corresponding user for correction by the user.

[0135] By adjusting the text volume threshold and the fuzziness rate for recognizing the stored text materials, dynamic management of the text materials can be achieved, the flexibility of text management can be improved, and the maintenance stability can be further improved.

[0136] Please refer to Figure 3 shown in the figure, which is a schematic diagram for recognizing the font fuzziness rate of an embodiment of the present invention. In the figure, the character of the original file is "化", and at different fuzziness levels, it can be recognized as different characters "化", "什", "代";

[0137] In implementation, for a single file, it corresponds to an original fuzziness rate. If the character recognized by this fuzziness rate cannot be recognized semantically, the fuzziness rate is adjusted to change the recognized character. Please continue to refer to Figure 2 shown in the figure. In the figure, if it is recognized as "化" and forms the word "消化", this recognition is successful;

[0138] If it is recognized as "化" and forms the word "化么", it is determined that this recognition is incorrect, and the fuzziness rate is increased;

[0139] After increasing the fuzziness rate, the recognized character changes to "什" and forms the word "什么", and this recognition is successful.

[0140] Specifically, in step S1, the original data is composed of a plurality of sections, and when the image of the original data is recognized, each area corresponds to a single text data.

[0141] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An image classification and recognition system based on multiple collaborative neural networks, characterized in that: include: A scanning module, which is used to scan the original data to form corresponding image data; A classification module, which is connected to the scanning module, is used to determine the editing authority of the corresponding user according to the editing authority of the user on the original data, and classify the image data into a corresponding classification pool; When classifying the user's permissions, the text data is classified according to the user's permissions. If the user has the permission to write, the text data is stored in the writing pool of the user; If the user has editing authority, the text data is stored in the editing pool of the user; The classification module combines the writing pool and editing pool of the corresponding user to form a feature pool of the user. An identification module, which is connected to the classification module, and is used to identify and correct each image data according to the classification pool to form corresponding record data; A storage module, connected to the identification module, for recording the recorded data; A timing module, connected to the scanning module and the storage module, for maintaining the system after a preset time has passed; Wherein, the preset time duration is related to the damage cycle of the original data; For a single user, the recognition module is provided with an editing pool of the user, the recognition module recognizes similar graphics in the editing pool of the user, and fits the similar graphics to form a writing fitting pool of the user. The recognition module also uses a neural network to record the text corresponding to the similar graphics in each text data to form a fitting text, and uses a plurality of fitting texts to coordinate with the fitting text to perform deviation correction, and forms a record text for unrecognizable content.

2. The image classification and recognition system based on multiple collaborative neural networks according to claim 1 is characterized in that: The recognition module determines the recognition characters of the user according to the content of each text material in the writing pool corresponding to the single user; Wherein, when determining the recognition characters, a preset reference number of text materials with a text volume greater than a text volume threshold is set as a standard file, and the characters in each standard file are respectively used as the recognition characters; The text volume is the number of characters in the text material, and the text volume threshold is related to the average text volume of each text material in the writing pool; Wherein, the recognition characters correspond to pictures and text information.

3. The image classification and recognition system based on multiple collaborative neural networks according to claim 2 is characterized in that: For a single recognized character, when the recognition module recognizes the character, if the text category corresponding to the character is greater than one, the character is marked as an abnormal character, and the character is fuzzy recognized to correspond to a single text, and the recognized characters and the corresponding texts are combined into the writing fitting pool; The text categories include characters and character groups.

4. The image classification and recognition system based on multiple collaborative neural networks according to claim 3 is characterized in that: For the single user, when establishing the handwriting fitting pool corresponding to the user, the recognition module sets a fuzziness rate corresponding to the fuzzy recognition, outputs the text corresponding to each recognized character whose fitting degree exceeds the preset recognition rate according to the fuzziness rate, and forms a corresponding fitting text; Wherein, the preset recognition rate is related to the blur rate corresponding to each character in the fitting text; The fitting text includes text materials in the writing pool, but does not include text materials in the editing pool.

5. The image classification and recognition system based on multiple collaborative neural networks according to claim 4 is characterized in that: For a single fitting text, the recognition module also performs semantic recognition on the fitting text. If the fitting text contains any unrecognizable text content, the fitting text is corrected in a preset correction method to form a record text and a corresponding correction record; Among them, the preset correction method is to correct the deviation through customary expressions.

6. The image classification and recognition system based on multiple collaborative neural networks according to claim 5 is characterized in that: When the preset maintenance time has passed, the timing module adjusts the text volume threshold and the fuzziness rate according to the writing pool of the single user, and identifies each of the original materials stored in the preset maintenance time according to the text volume threshold and the fuzziness rate to form a pre-recorded text.

7. The image classification and recognition system based on multiple collaborative neural networks according to claim 6 is characterized in that: When the recognition module recognizes the recorded text, if there is a difference between the recognition information of any of the recognized characters in the recorded text and the corresponding text in the pre-recorded text, the recognized character is determined to be incorrect, and the incorrect information is alerted to the corresponding user for correction by the user.

8. The image classification and recognition system based on multiple collaborative neural networks according to claim 1 is characterized in that: When collecting the original data, the scanning module divides the original data into several sections, and identifies the pictures corresponding to each section to form corresponding text data; Among them, a single area corresponds to a single text document.

9. The image classification and recognition system based on multiple collaborative neural networks according to any one of claims 7 or 8, characterized in that: When the timing module identifies the original data stored in the preset maintenance duration, its sample does not include the record text.

Images