Intelligent paper marking content detection and identification method and system based on deep learning

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an intelligent paper marking content detection and identification method and system based on deep learning, and the method comprises the steps: determining a to-be-detected region through a locator; identifying a to-be-detected area based on the pixel value, and screening out an effective detection area; the uploaded characters are recognized and matched with a preset character database, and clear characters and fuzzy characters are distinguished; determining alternative preset characters corresponding to the fuzzy characters, performing screening through context semantics to obtain first preset characters, determining whether the same fuzzy characters exist or not, determining second preset characters for the same fuzzy characters, determining target characters according to the first preset characters or the second preset characters, or determining the target characters according to the target characters. Determining a writing model based on a deep learning algorithm and the plurality of clear characters, determining a third preset character based on the writing model and the fuzzy character, and determining a target character based on the first preset character and the third preset character. The accuracy of the intelligent paper marking result is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to an intelligent marking content detection and recognition method and system based on deep learning. Background Art

[0002] In recent years, deep learning technology has made great progress and demonstrated powerful capabilities in the fields of image and text processing. In terms of images, deep learning architectures such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) can effectively extract image features, have strong representation capabilities for various complex image contents, and can better handle the recognition tasks of handwritten text. In terms of text processing, deep learning can deeply understand and analyze the semantics, structure, etc. of text, which provides a technical basis for intelligent marking.

[0003] The patent document with the Chinese patent application publication number CN113222788A discloses an intelligent marking method, including the following steps: first, scan the complaint to obtain the complaint annotation and content, then magnify or rotate and correct the extracted annotation and content, identify the scanned text to obtain the complaint content text and confirm the keywords, and after corresponding the obtained keywords with the keywords in the case-filing form, fill the content into the case-filing form; retrieve and transfer relevant electronic files from the electronic file database according to the text keywords and select whether to translate according to the operator's needs, fill the electronic file content into the case-filing form, and review whether the content is accurate. If the filling is accurate, the case-filing form is completed. If it is inaccurate, modify it, synchronize the modified content to the electronic file and fill the modified content back into the case-filing form to complete the case-filing form.

[0004] In the prior art, due to factors such as the diversity of handwritten text, writing clarity, and scanning quality, there is a problem that blurred text is difficult to accurately identify. Through simple keyword matching, in-depth understanding and analysis of text semantics are not involved, resulting in incomplete text recognition results and inaccurate intelligent marking results. Summary of the Invention

[0005] Therefore, the present invention provides an intelligent marking content detection and recognition method and system based on deep learning. By combining strategies such as context semantic analysis and deep learning to construct a writing model to identify blurred text on the test paper, the problem of inaccurate intelligent marking results can be solved.

[0006] To achieve the above object, the present invention provides an intelligent marking content detection and recognition method based on deep learning, and the method includes:

[0007] Determine the area to be detected on the test paper based on the locator;

[0008] Identify the area to be detected based on the pixel values, and determine the effective detection area according to the recognition result;

[0009] Identify several uploaded characters in the effective detection area, match the several uploaded characters with a preset character database, and determine several clear characters and several blurred characters according to the matching results;

[0010] Determine several alternative preset characters corresponding to any blurred character, and screen the several alternative preset characters based on the context semantics to determine several first preset characters;

[0011] Based on several blurred characters, determine whether there are identical blurred characters, and determine several second preset characters corresponding to the blurred characters identical to any of the blurred characters based on the judgment result, so as to determine the target characters according to the several first preset characters or the several second preset characters, or, determine a writing model based on a deep learning algorithm and several clear characters, determine several third preset characters based on the writing model and the blurred characters, and determine the target characters based on the several first preset characters and the several third preset characters;

[0012] Mark the blurred characters according to the target characters to assist intelligent marking.

[0013] Further, the step of determining the effective detection area according to the recognition result includes:

[0014] Identify several actual pixel values corresponding to the area to be detected, and determine an initial character area based on a threshold segmentation algorithm and the several actual pixel values;

[0015] Adjust the initial character area based on dilation operation and erosion operation to obtain an adjusted character area;

[0016] Perform projection analysis on the adjusted character area based on the character arrangement feature to determine the effective detection area according to the analysis result.

[0017] Further, the step of determining several clear characters and several blurred characters according to the matching results includes:

[0018] Perform similarity matching between several uploaded characters and several preset characters in the preset character database to obtain several actual similarity values;

[0019] Compare the several actual similarity values with a preset similarity value to determine several clear characters or several blurred characters according to the comparison result.

[0020] Further, the step of determining several alternative preset characters corresponding to any blurred character includes:

[0021] Calculate the similarity between any of the blurred characters and the characters in the preset character database and sort them to obtain a first similarity sequence;

[0022] Determine a corresponding number of the alternative texts based on the first similarity sequence and a preset quantity threshold.

[0023] Further, the step of screening a number of alternative preset texts based on context semantics includes:

[0024] Locate any ambiguous text, determine its corresponding context content based on punctuation marks and convert it into a vector form based on a word vector model to obtain an upper word vector and a lower word vector;

[0025] Calculate the average semantic similarity between a number of the alternative preset texts and the upper word vector and the lower word vector respectively to obtain a first semantic similarity;

[0026] Compare a number of the first semantic similarities with a first preset semantic similarity and determine a number of first preset texts based on the comparison result.

[0027] Further, the step of determining a target text based on a number of the first preset texts or a number of second preset texts includes:

[0028] Calculate a number of second semantic similarities corresponding to a number of second preset texts corresponding to the same ambiguous text;

[0029] Select a number of fourth preset texts that are the same among a number of the first preset texts and a number of the second preset texts;

[0030] Determine a third semantic similarity corresponding to the fourth preset text based on the first semantic similarity and the second semantic similarity, and determine the target text based on a number of the third semantic similarities.

[0031] Further, the step of determining a writing model based on a deep learning algorithm and a number of clear texts includes:

[0032] Input a number of the clear texts into a preset deep learning model to train and obtain the writing model;

[0033] Extract features of the to-be-recognized ambiguous text through the writing model to obtain an ambiguous text feature vector;

[0034] Match the ambiguous text feature vector with preset text feature vectors in a preset text database to determine a number of third preset texts according to the matching result.

[0035] Further, the step of determining a target text based on a number of the first preset texts and a number of the third preset texts includes:

[0036] Calculate the average semantic similarity between a number of the third preset texts and the upper word vector and the lower word vector respectively to obtain a third semantic similarity;

[0037] Determine the target text based on the comparison results of a number of the first semantic similarities and a number of the third semantic similarities.

[0038] Further, the step of determining the area to be detected on the test paper based on the locator includes:

[0039] Determine the locator based on a preset shape and a preset pixel value;

[0040] Determine the horizontal locator and the vertical locator based on the straight-line distance between the centers of adjacent locators;

[0041] Correct the test paper based on the horizontal locator and the vertical locator to determine the area to be detected corresponding to the locator.

[0042] On the other hand, the present invention also provides a system for an intelligent marking content detection and recognition method based on deep learning, the system includes:

[0043] An area determination module, configured to determine the area to be detected on the test paper based on the locator, identify the area to be detected based on the pixel value, and determine the effective detection area according to the identification result;

[0044] A text recognition module, connected to the area determination module, configured to recognize a number of uploaded texts in the effective detection area, match the number of uploaded texts with a preset text database, and determine a number of clear texts and a number of fuzzy texts according to the matching result;

[0045] A screening module, connected to the text recognition module, configured to determine a number of alternative preset texts corresponding to any fuzzy text, and screen the number of alternative preset texts based on the context semantics to determine a number of first preset texts;

[0046] A target determination module, connected to the screening module, configured to determine whether there are the same fuzzy texts based on a number of fuzzy texts, determine a number of second preset texts corresponding to the fuzzy texts that are the same as any of the fuzzy texts based on the determination result, and determine the target text according to the number of first preset texts or the number of second preset texts, or, determine a writing model based on a deep learning algorithm and a number of clear texts, determine a number of third preset texts based on the writing model and the fuzzy texts, and determine the target text based on the number of first preset texts and the number of third preset texts;

[0047] A marking module, connected to the target determination module, configured to mark the fuzzy text according to the target text to assist intelligent marking.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows. The locator can accurately lock the specific area to be detected in the test paper, improving the processing efficiency and enhancing the adaptability of the method to different test paper templates. By analyzing the pixel values, using threshold segmentation, dilation and erosion can effectively remove the noise and irrelevant background information in the image, only retaining the effective area related to the text, improving the accuracy of subsequent text recognition, clarifying the effective detection area, avoiding invalid calculations, and further improving the processing efficiency. By quickly distinguishing clearly recognizable text from blurred text, it provides a basis for subsequent processing of different types of text. By matching with a preset text database, the quality of text recognition is preliminarily evaluated to determine whether it is clearly recognizable, which helps to improve the accuracy and reliability of marking. Using context semantics to screen alternative texts makes the determined first preset text more semantically consistent with the overall context of the text, improving the accuracy and rationality of recognizing blurred text. By using multiple strategies to determine the target text, the accuracy of recognizing blurred text is improved. Based on the deep learning algorithm, the writing model is learned from the clear text, which can better understand the writing characteristics of the text, thus making a more accurate inference for the blurred text, improving the adaptability. Marking the target text at the blurred text greatly improves the marking efficiency, reduces the time for recognizing blurred text, and improves the accuracy and fairness of marking.

[0049] In particular, by identifying the actual pixel values of the area to be detected, the information in the test paper image can be converted into a data form that can be quantified and processed, providing basic data for subsequent processing. The threshold segmentation algorithm divides the image into different regions according to the different pixel values, initially separating the region that may contain text and the background region, thus achieving the first step of screening out the region where the text is located from the complex test paper image, narrowing the range that needs further processing, and improving the pertinence and efficiency of subsequent processing. Through the dilation operation, the possibly disconnected parts of the text strokes are connected. For some texts with scribbled writing or broken strokes caused by scanning, through the dilation operation, the disconnected strokes can be connected into a complete shape, which is beneficial to subsequent text recognition and processing, making the text structure more complete. Through the erosion operation, the redundant noise points generated by dilation are removed, improving the quality of the text region and enhancing the continuity and integrity of the text region. According to the arrangement characteristics that text is usually distributed in rows and has a certain line spacing and character spacing, projection analysis can better organize and segment the text region according to natural rows and blocks. By accumulating the pixel values in the horizontal and vertical directions to obtain the projection curve, based on the peak and valley characteristics of the curve, the text line position and text block position can be accurately determined, which helps to divide the text content into the correct rows and paragraphs, avoiding misjudging the text in different rows as one row, ensuring the accuracy and orderliness of subsequent text content recognition and processing, and improving the accuracy and reliability of the entire intelligent marking process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flowchart of the intelligent marking content detection and recognition method based on deep learning provided by an embodiment of the present invention;

[0051] Figure 2 It is a schematic flowchart of determining an effective detection area in the intelligent marking content detection and recognition method based on deep learning provided by an embodiment of the present invention;

[0052] Figure 3 It is a schematic structural diagram of an area to be detected on a test paper in the intelligent marking content detection and recognition method based on deep learning provided by an embodiment of the present invention;

[0053] Figure 4 It is a block diagram of the structure of the intelligent marking content detection and recognition system based on deep learning provided by an embodiment of the present invention. Specific embodiments

[0054] In order to make the objectives and advantages of the present invention clearer, the present invention will be 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.

[0055] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0056] It should be noted that in the description of the present invention, terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for 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, and therefore should not be construed as a limitation of the present invention.

[0057] 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 directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0058] Please refer to Figure 1 As shown, an embodiment of the present invention provides an intelligent marking content detection and recognition method based on deep learning, and the method includes:

[0059] Step S100, determining an area to be detected on a test paper based on a locator;

[0060] Step S200: Identify the area to be detected based on pixel values, and determine the effective detection area according to the identification result.

[0061] Step S300: Identify several uploaded texts in the effective detection area, and match the several uploaded texts with a preset text database to determine several clear texts and several fuzzy texts according to the matching result.

[0062] Step S400: Determine several alternative preset texts corresponding to any fuzzy text, and screen the several alternative preset texts based on the context semantics to determine several first preset texts.

[0063] Step S500: Judge whether there are identical fuzzy texts based on the several fuzzy texts, determine several second preset texts corresponding to the fuzzy texts identical to any of the fuzzy texts based on the judgment result, so as to determine the target text according to the several first preset texts or the several second preset texts, or determine a writing model based on a deep learning algorithm and the several clear texts, determine several third preset texts based on the writing model and the fuzzy texts, and determine the target text based on the several first preset texts and the several third preset texts.

[0064] Step S600: Mark the fuzzy text according to the target text to assist intelligent marking.

[0065] It can be understood that the area to be detected in the test paper in the embodiment of the present invention may refer to the short-answer question area corresponding to the liberal arts test paper, and the answer area is all text descriptions without other formula or character explanations.

[0066] It can be understood that the preset text database in the embodiment of the present invention should include various types of words, phrases and other text contents that may be involved in liberal arts short-answer questions. These contents can be derived from teaching syllabuses, textbooks, answer keys of previous years' examination questions, etc. For example, for liberal arts short-answer questions in the history subject, the database will include historical event names, historical figure names, relevant historical terms, etc.; for the Chinese subject, there will be various literary genre names, rhetorical device names, words and expressions, etc. The texts in the database are usually stored in a standardized form, and each text or word has a corresponding standard representation.

[0067] Specifically, in the embodiments of the present invention, a locator can accurately lock a specific area to be detected in a test paper, improving the processing efficiency and enhancing the adaptability of the method to different test paper templates. By analyzing pixel values, threshold segmentation, dilation, and erosion are used to effectively remove noise and irrelevant background information in the image, only retaining the effective area related to the text, improving the accuracy of subsequent text recognition, clarifying the effective detection area, avoiding invalid calculations, and further improving the processing efficiency. By quickly distinguishing clearly recognizable text from blurred text, it provides a basis for subsequent processing of different types of text. By matching with a preset text database, the quality of text recognition is preliminarily evaluated to determine whether it is clearly recognizable, which helps to improve the accuracy and reliability of marking. Using context semantics to screen alternative texts makes the determined first preset text more semantically consistent with the overall context of the text, improving the accuracy and rationality of blurred text recognition. By using multiple strategies to determine the target text, the accuracy of blurred text recognition is improved. Based on a deep learning algorithm, the writing model is learned from clear text, which can better understand the writing characteristics of the text, thereby making a more accurate inference for blurred text, improving the adaptability. Marking the target text at the blurred text greatly improves the marking efficiency, reduces the time for recognizing blurred text, and improves the accuracy and fairness of marking.

[0068] See Figure 2 As shown, the step of determining the effective detection area according to the recognition result includes:

[0069] Step S210, recognizing a plurality of actual pixel values corresponding to the area to be detected, and determining an initial text area based on the threshold segmentation algorithm and the plurality of actual pixel values;

[0070] Step S220, adjusting the initial text area based on dilation operation and erosion operation to obtain an adjusted text area;

[0071] Step S230, performing projection analysis on the adjusted text area based on text arrangement characteristics to determine the effective detection area according to the analysis result.

[0072] Specifically, the embodiments of the present invention can convert the information in the test paper image into a data form that can be quantified and processed by identifying the actual pixel values of the area to be detected, providing basic data for subsequent processing. The threshold segmentation algorithm divides the image into different regions according to different pixel values, initially separating the regions that may contain text and the background regions, thus achieving the first step of screening out the text-containing regions from the complex test paper image, narrowing the scope that needs further processing, and improving the pertinence and efficiency of subsequent processing. By dilation operation, the possibly disconnected parts of the text strokes are connected. For some texts with scribbled writing or broken strokes caused by scanning, through the dilation operation, the disconnected strokes can be connected into a complete shape, which is beneficial to subsequent text recognition and processing and makes the text structure more complete. By erosion operation, the redundant noise points generated by dilation are removed, improving the quality of the text region and enhancing the continuity and integrity of the text region. Projection analysis is performed according to the arrangement characteristics that text is usually distributed in rows and has a certain line spacing and character spacing, which can better organize and segment the text region according to natural rows and blocks. By accumulating the pixel values in the horizontal and vertical directions, the projection curves are obtained. According to the peak and valley characteristics of the curves, the positions of text lines and text blocks are accurately determined, which helps to divide the text content into the correct rows and paragraphs, avoid misjudging the text in different rows as one row, and ensure the accuracy and orderliness of subsequent text content recognition and processing, improving the accuracy and reliability of the entire intelligent marking process.

[0073] It can be understood that the initial text region can be adjusted based on dilation operation and erosion operation in the embodiments of the present invention by first performing the dilation operation. By setting a suitable structuring element (such as a 3×3 rectangular structuring element), the possibly disconnected parts of the text strokes are connected; then the erosion operation is performed to remove the redundant noise points generated by dilation, making the text region clearer and more accurate.

[0074] It can be understood that the text arrangement characteristics in the embodiments of the present invention are that text is usually distributed in rows and has a certain line spacing and character spacing.

[0075] It can be understood that the steps of performing projection analysis on the adjusted text region in the embodiments of the present invention may include:

[0076] By accumulating the pixel values in the horizontal and vertical directions corresponding to the adjusted text region to obtain the horizontal projection curve and the vertical projection curve;

[0077] Determine the horizontal peak characteristics, horizontal valley characteristics, vertical peak characteristics, and vertical valley characteristics corresponding to the horizontal projection curve and the vertical projection curve respectively. Determine the text line position based on the horizontal peak characteristics and horizontal valley characteristics, and determine the text block position based on the vertical peak characteristics and vertical valley characteristics, so as to determine the effective detection area according to the text line position and the text block position.

[0078] It can be understood that in the embodiments of the present invention, the positions of text lines and text blocks are determined according to the peak and valley characteristics of the curve. For example, on the horizontal projection curve, the valley position corresponds to the blank area between text lines. By reasonably setting the valley threshold, different text line areas can be accurately divided.

[0079] Specifically, the steps of determining a number of clear texts and a number of fuzzy texts according to the matching result include:

[0080] Perform similarity matching between a number of uploaded texts and a number of preset texts in a preset text database to obtain a number of actual similarity values;

[0081] Compare the number of actual similarity values with a preset similarity value to determine a number of clear texts or a number of fuzzy texts according to the comparison result.

[0082] It can be understood that the preset similarity value in the embodiments of the present invention is 98%.

[0083] It can be understood that in the embodiments of the present invention, when the actual similarity value is greater than or equal to the preset similarity value, the corresponding uploaded text is determined as a clear text, and when the actual similarity value is less than the preset similarity value, the corresponding uploaded text is determined as a fuzzy text.

[0084] Specifically, the steps of determining a number of alternative preset texts corresponding to any fuzzy text include:

[0085] Calculate the similarity between any of the fuzzy texts and the texts in the preset text database and sort them to obtain a first similarity sequence;

[0086] Determine a corresponding number of the alternative texts based on the first similarity sequence and a preset quantity threshold.

[0087] It can be understood that the preset quantity threshold in the embodiments of the present invention is 5.

[0088] Specifically, the steps of screening a number of alternative preset texts based on context semantics include:

[0089] Locate any fuzzy text, determine its corresponding context content based on punctuation marks and convert it into a vector form based on a word vector model to obtain an upper word vector and a lower word vector;

[0090] Calculate the average semantic similarity between several of the alternative preset texts and the upper word vector and the lower word vector respectively to obtain the first semantic similarity;

[0091] Compare several first semantic similarities with a first preset semantic similarity, and determine several first preset texts based on the comparison results.

[0092] It can be understood that the embodiments of the present invention determine the corresponding context content based on punctuation marks and convert it into a vector form based on the word vector model, including:

[0093] Taking any of the fuzzy texts as the center, identify forward and backward based on a preset punctuation mark library, and use the text between the fuzzy text and the nearest previous punctuation mark as the upper context content, and the text between the fuzzy text and the nearest next punctuation mark as the lower context content;

[0094] Perform word segmentation processing on the determined upper context content and lower context content respectively to obtain several upper context word segments and several lower context word segments;

[0095] Based on the word vector model, convert several upper context word segments and several lower context word segments into several upper context word segment vectors and several lower context word segment vectors respectively.

[0096] It can be understood that the preset punctuation mark library in the embodiments of the present invention may include several punctuation marks determined in the prior art such as commas, periods, semicolons, etc.

[0097] It can be understood that the word segmentation processing in the embodiments of the present invention can be performed through natural language processing toolkits (such as NLTK, jieba, etc.).

[0098] It can be understood that the conversion of vectors based on the word vector model in the embodiments of the present invention can be achieved by, for each word, looking up the corresponding vector in the word list of the word vector model. If it is an OOV (Out-Of-Vocabulary) word, its vector representation can be generated by randomly initializing the vector or a character-based vector representation method.

[0099] It can be understood that the word vector model in the embodiments of the present invention is such as Word2Vec, GloVe or a word vector model trained for a specific field.

[0100] It can be understood that the first preset semantic similarity in the embodiments of the present invention is 0.9.

[0101] It can be understood that in the embodiments of the present invention, when the first semantic similarity is greater than the first preset semantic similarity, the corresponding alternative preset text is used as the first preset text.

[0102] Specifically, the steps of determining the target text according to several first preset texts or several second preset texts include:

[0103] Calculate the several second semantic similarities corresponding to several second preset texts corresponding to the same fuzzy text;

[0104] Select several fourth preset texts that are the same for several first preset texts and several second preset texts;

[0105] Determine the third semantic similarity corresponding to the fourth preset text based on the first semantic similarity and the second semantic similarity, and determine the target text based on several third semantic similarities.

[0106] It can be understood that the embodiments of the present invention further include that if there are no common elements in the first preset text set and the second preset text set, it can be processed according to specific situations. For example, according to the semantic similarity scores of the two sets, the elements in the set with a higher score can be used as the fourth preset text; or the two sets can be merged and the semantic evaluation can be performed again.

[0107] It can be understood that for each element in the fourth preset text set in the embodiments of the present invention, its corresponding third semantic similarity is obtained by a weighted average method according to its first semantic similarity in the first preset text set and its second semantic similarity in the second preset text set.

[0108] It can be understood that the third semantic similarity in the embodiments of the present invention is 0.9.

[0109] Specifically, the steps of determining the writing model based on the deep learning algorithm and several clear texts include:

[0110] Input several of the clear texts into a preset deep learning model to train and obtain the writing model;

[0111] Extract features of the fuzzy text to be recognized through the writing model to obtain a fuzzy text feature vector;

[0112] Match the fuzzy text feature vector with the preset text feature vectors in the preset text database to determine several third preset texts according to the matching results.

[0113] It can be understood that the preset deep learning model in the embodiments of the present invention can be a convolutional neural network model, a recurrent neural network model, a long short-term memory network model, etc., and is used to identify and learn the writing features of clear texts.

[0114] It can be understood that the preset text feature vector in the embodiments of the present invention is a vector representation obtained by extracting features from the texts in the preset text database, which may include various information such as the structure, strokes, semantics, etc. of the texts. One example of feature vector extraction can be that for each preset text, it is represented in the form of an image, and the convolutional layer in the CNN is used to perform convolutional operations on the image. The convolutional kernel slides on the image to extract local features at different levels, such as the edges and corners of the strokes; it is processed through multiple convolutional layers and pooling layers; and the feature map after pooling is converted into a vector of a fixed length through the fully connected layer.

[0115] Specifically, the steps of determining the target text based on a number of first preset texts and a number of third preset texts include:

[0116] Calculate the average semantic similarity between the number of third preset texts and the upper word vector and the lower word vector respectively to obtain the third semantic similarity;

[0117] Determine the target text based on the comparison results of the number of the first semantic similarities and the number of the third semantic similarities.

[0118] It can be understood that a possible example of determining the target text in the embodiments of the present invention is to select the preset text corresponding to the maximum value among the number of first semantic similarities and the number of third semantic similarities as the target text.

[0119] Specifically, the steps of determining the area to be detected on the test paper based on the locator include:

[0120] Determine the locator based on the preset shape and preset pixel value;

[0121] Determine the horizontal locator and the vertical locator based on the straight-line distance between the centers of adjacent locators;

[0122] Correct the test paper based on the horizontal locator and the vertical locator to determine the area to be detected corresponding to the locator.

[0123] It can be understood that the preset shape in the embodiments of the present invention is a template shape determined according to the actual designed shape features of the locators on the test paper. For example, the locator may be designed as a simple geometric shape such as a circle, a square, a triangle, etc., or a shape with a unique contour composed of specific line combinations. In practical applications, by analyzing the shape of the locator, key geometric features such as the radius of the circle, the side length of the square, the side length and angle of the triangle, etc. are extracted as the description parameters of the preset shape.

[0124] The preset pixel values are related to the color characteristics of the locators. When processing the test paper image, the preset pixel values are determined according to the pixel value range of the locator in the corresponding color space (such as RGB, HSV, etc.). For example, if the locator is red, in the RGB color space, the pixel value range corresponding to red may be (R: 200 - 255, G: 0 - 50, B: 0 - 50), which is an example of the preset pixel values. By setting such a pixel value range, the areas that may belong to the locators can be preliminarily screened out in the image, and then combined with the preset shape to accurately identify the locators.

[0125] See Figure 4 As shown, the embodiment of the present invention also provides a system for an intelligent marking content detection and recognition method based on deep learning. The system includes:

[0126] An area determination module 10, configured to determine the area to be detected on the test paper based on the locator, identify the area to be detected based on the pixel values, and determine the effective detection area according to the recognition result;

[0127] A text recognition module 20, connected to the area determination module 10, configured to recognize a number of uploaded texts in the effective detection area, match the number of uploaded texts with a preset text database, and determine a number of clear texts and a number of fuzzy texts according to the matching result;

[0128] A screening module 30, connected to the text recognition module 20, configured to determine a number of alternative preset texts corresponding to any fuzzy text, and screen the number of alternative preset texts based on the context semantics to determine a number of first preset texts;

[0129] A target determination module 40, connected to the screening module 30, configured to determine whether there are the same fuzzy texts based on the number of fuzzy texts, determine a number of second preset texts corresponding to the fuzzy texts that are the same as any of the fuzzy texts based on the determination result, and determine the target text according to the number of first preset texts or the number of second preset texts, or, determine a writing model based on the deep learning algorithm and the number of clear texts, determine a number of third preset texts based on the writing model and the fuzzy texts, and determine the target text based on the number of first preset texts and the number of third preset texts;

[0130] A marking module 50, connected to the target determination module 40, configured to mark the fuzzy text according to the target text to assist intelligent marking.

[0131] Specifically, the system for the intelligent marking content detection and recognition method based on deep learning provided by the embodiment of the present invention can execute the above-mentioned intelligent marking content detection and recognition method based on deep learning, achieving the same technical effects, which will not be elaborated here.

[0132] So far, the technical solution of the present invention has 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 all fall within the protection scope of the present invention.

[0133] The above are only the preferred embodiments of the present invention and are not used to limit the present invention; for those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting and identifying intelligent examination paper content based on deep learning, characterized in that: include: Determine the area of ​​the test paper to be tested based on the locator; Identify the area to be detected based on the pixel value, and determine the effective detection area according to the identification result; Identify a number of uploaded characters in the effective detection area, and match the uploaded characters with a preset character database to determine a number of clear characters and a number of fuzzy characters according to the matching results; Determine a number of candidate preset characters corresponding to any ambiguous character, and screen the number of candidate preset characters based on context semantics to determine a number of first preset characters; Determine whether there is an identical ambiguous character based on a plurality of ambiguous characters, determine a plurality of second preset characters corresponding to an ambiguous character identical to any of the ambiguous characters based on the determination result, and determine a target character based on a plurality of first preset characters or a plurality of second preset characters, or, determine a writing model based on a deep learning algorithm and a plurality of clear characters, determine a plurality of third preset characters based on the writing model and the ambiguous characters, and determine a target character based on a plurality of first preset characters and a plurality of third preset characters; The ambiguous text is annotated according to the target text to assist intelligent marking.

2. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 1 is characterized in that: The step of determining the effective detection area according to the recognition result comprises: Identify a number of actual pixel values ​​corresponding to the area to be detected, and determine an initial text area based on a threshold segmentation algorithm and the number of actual pixel values; Adjusting the initial text region based on a dilation operation and an erosion operation to obtain an adjusted text region; The adjusted text area is projected and analyzed based on the text arrangement characteristics to determine the effective detection area according to the analysis result.

3. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 2 is characterized in that: The steps of determining a plurality of clear characters and a plurality of fuzzy characters according to the matching results include: Perform similarity matching between a number of uploaded texts and a number of preset texts in a preset text database to obtain a number of actual similarity values; The actual similarity values ​​are compared with the preset similarity values ​​to determine a number of clear characters or a number of blurred characters according to the comparison result.

4. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 3 is characterized in that: The steps of determining a number of candidate preset characters corresponding to any fuzzy character include: Calculating the similarity between any of the fuzzy characters and characters in a preset character database and sorting them to obtain a first similarity sequence; The corresponding plurality of candidate characters are determined based on the first similarity sequence and a preset quantity threshold.

5. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 4 is characterized in that: The steps of screening several candidate preset texts based on context semantics include: Locate any ambiguous text, determine its corresponding context based on punctuation marks, and convert it into vector form based on the word vector model to obtain the upper word vector and the lower word vector; Calculate the average semantic similarity between the candidate preset characters and the upper word vector and the lower word vector respectively to obtain a first semantic similarity; The first semantic similarities are compared with the first preset semantic similarities, and the first preset characters are determined based on the comparison results.

6. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 5 is characterized in that: The step of determining the target text according to the plurality of first preset texts or the plurality of second preset texts comprises: Calculating a plurality of second semantic similarities corresponding to a plurality of second preset characters corresponding to the same fuzzy character; Selecting a plurality of fourth preset characters that are the same as the plurality of first preset characters and the plurality of second preset characters; A third semantic similarity corresponding to the fourth preset text is determined based on the first semantic similarity and the second semantic similarity, and the target text is determined based on a plurality of third semantic similarities.

7. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 6, characterized in that: The steps of determining the writing model based on the deep learning algorithm and several clear texts include: Inputting a number of the clear characters into a preset deep learning model to train and obtain the writing model; Extracting features of the fuzzy characters to be recognized by using the writing model to obtain fuzzy character feature vectors; The fuzzy character feature vector is matched with a preset character feature vector in a preset character database to determine a plurality of third preset characters according to the matching result.

8. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 7, characterized in that: The step of determining the target text based on the plurality of first preset texts and the plurality of third preset texts comprises: Calculate the average semantic similarity of the third preset characters with the upper word vector and the lower word vector respectively to obtain a third semantic similarity; The target text is determined based on comparison results of the first semantic similarities and the third semantic similarities.

9. The method for detecting and identifying intelligent examination paper content based on deep learning according to claim 8, characterized in that: The steps of determining the area to be detected on the test paper based on the locator include: Determine a locator based on a preset shape and a preset pixel value; Determine the horizontal locator and the vertical locator based on the linear distance between the centers of adjacent locators; The test paper is corrected based on the horizontal locator and the vertical locator to determine the area to be detected corresponding to the locator.

10. A system using the deep learning-based intelligent examination paper content detection and recognition method according to any one of claims 1 to 9, characterized in that: include: An area determination module is used to determine the area to be detected of the test paper based on the locator, identify the area to be detected based on the pixel value, and determine the effective detection area according to the identification result; A text recognition module, connected to the area determination module, for identifying a number of uploaded texts in the effective detection area, matching the uploaded texts with a preset text database, and determining a number of clear texts and a number of blurred texts according to the matching results; A screening module, connected to the text recognition module, for determining a number of candidate preset texts corresponding to any ambiguous text, and screening the number of candidate preset texts based on context semantics to determine a number of first preset texts; a target determination module connected to the screening module, for determining whether there is an identical fuzzy text based on a plurality of fuzzy texts, determining a plurality of second preset texts corresponding to the fuzzy texts identical to any of the fuzzy texts based on the determination result, determining the target text based on the plurality of first preset texts or the plurality of second preset texts, or determining a writing model based on a deep learning algorithm and a plurality of clear texts, determining a plurality of third preset texts based on the writing model and the fuzzy texts, and determining the target text based on the plurality of first preset texts and the plurality of third preset texts; A marking module is connected to the target determination module and is used to mark the ambiguous text according to the target text to assist intelligent marking.

Images