Automatic homework correcting system

Abstract

The invention discloses an automatic homework correcting system. Comprising a handwriting collection module, an OCR recognition module, a wrongly written character correction module and an intelligent scoring module. The handwriting acquisition module is used for acquiring handwriting; the OCR recognition module is used for recognizing the handwriting; the wrongly written character correction module is used for correcting wrongly written characters; the intelligent scoring module is used for comprehensive consideration; the test paper handwriting is collected and quickly converted into text data, the CR recognition module can accurately recognize the writing content of the test paper, the writing content can be accurately converted into texts even in the face of different writing styles and handwriting characteristics, and the wrongly written character correction module can ensure the text accuracy of examinee answers; the intelligent scoring module is combined with question requirements, a standard question solving process and an examinee answering process, deep analysis is carried out through an LLM large language model, objective and accurate scores and targeted comments can be given, and the accuracy and reliability of examination review are improved.

Description

Technical Field

[0001] The present invention belongs to the field of educational technology, and more specifically, particularly relates to an automatic homework grading system. Background Art

[0002] In the modern educational environment, with the advancement of digital teaching, the way of correcting students' homework urgently needs to be reformed. Currently, there are many problems in the handwriting recognition and text conversion links. Taking some existing homework correction software as an example, in terms of collecting handwriting, due to the complexity of handwriting (such as scribbled writing, connected writing, etc.), the recognition accuracy is seriously affected. The recognition system is difficult to accurately grasp key information such as the shape, direction, and connection method of strokes. For the handwriting written with some special writing tools (such as colored pens, highlighters, etc.), the recognition effect is even worse. The colors and fluorescent effects of these special handwritings will interfere with the normal operation of the recognition system, making it unable to accurately extract handwriting features.

[0003] In terms of text error correction, although some existing software can identify some misspelled words, the correction effect for some complex misspelled words (such as similar-shaped characters, near-homophonic characters, etc.) is not good. For easily confused words such as "already" and "oneself", "de" and "di", it may not be able to correct accurately. Moreover, for some professional terms in specific fields, due to the lack of corresponding thesaurus support, it may not be able to correctly identify and correct errors.

[0004] Existing intelligent scoring systems often have the problem of inaccurate scoring. Some systems may simply score according to the correctness of the answers, without considering the students' problem-solving ideas, steps, and methods used. For example, in math homework, a student may have the correct answer, but the problem-solving process is not standard or the wrong method is used, and the existing system may still give a full score, which is not conducive to the learning and growth of students.

[0005] The present invention mainly aims at the deficiencies existing in the existing homework grading system and solves the following key problems: Inaccurate handwriting recognition: Whether it is the complex handwriting (such as scribbled, connected writing, etc.) written by students, or the handwriting left by using special writing tools (such as colored pens, highlighters, etc.), the current system has a low accuracy rate when recognizing these handwritings and converting them into text content.

[0006] Weak text error correction ability: Existing software has defects in text error correction. For some complex misspelled words (such as similar-shaped characters, near-homophonic characters, etc.) and professional terms in specific fields, the correction effect is not good, and the accuracy of the text content cannot be guaranteed.

[0007] Unreasonable intelligent grading: Existing intelligent grading systems often only grade based on the correctness of the answer, ignoring the students' problem-solving ideas, steps and methods used. Unreasonable grading methods are not conducive to students' learning and growth. Therefore, we propose an automatic homework grading system. Summary of the invention

[0008] In view of the problems existing in the prior art, the purpose of the present invention is to provide an automatic homework grading system for improving the accuracy of handwriting recognition, enhancing the text error correction capability, and comprehensively considering the requirements of the question, the standard problem-solving process and the student's problem-solving process to give more reasonable and accurate scores and comments.

[0009] To achieve the above object, the present invention provides the following technical solutions: an automatic homework correction system, comprising a handwriting collection module, an OCR recognition module, a typo correction module and an intelligent scoring module; The handwriting acquisition module is used to realize the collection of handwriting during the writing process. The handwriting acquisition module records the moving path of the pen tip through the optical tracking method of the micro camera, and the strain gauge sensor records the pressure change of the pen tip; accurately records various characteristics of the handwriting, including the position, strength and writing speed information of the strokes; The OCR recognition module recognizes the handwriting collected by the handwriting collection module, outputs text that can be processed and recognized by the computer, and performs multi-dimensional analysis of the handwriting through recognition algorithms and technologies; first, the shape of the strokes is analyzed, including the length, thickness, and curvature of the strokes; second, the direction of the strokes is determined; and finally, the connection mode of the strokes is determined, whether they are connected, intersecting, or disconnected; The wrong character correction module is used to correct wrong characters in the recognized text output by the OCR recognition module; using rich vocabulary resources, including general vocabulary and specific field professional terminology vocabulary, for complex wrong characters, a combination of multiple recognition methods is used to correct them, and the accuracy of text error correction is improved through the font comparison and pronunciation comparison methods; The intelligent scoring module scores and comments on the homework by comprehensively considering the text information output by the OCR recognition module in combination with the question requirements, the standard problem-solving process and the student's problem-solving process; a comprehensive intelligent scoring model is established, taking into account multiple factors such as the question requirements, the standard problem-solving process and the student's problem-solving process. In the scoring process, the correctness of the answers is used to analyze the students' problem-solving ideas, steps and methods used, give students a more accurate evaluation, and mark the errors.

[0010] Specifically, the handwriting acquisition module captures the pen tip pressure, motion trajectory and writing speed of the handwriting in real time through graphic coding, and converts the features into digital signals to form handwriting data; The graphic code is designed as a specific graphic coding method through optical marking. It is composed of tiny graphics composed of lines of different shapes and directions, and the graphics are arranged on the paper surface according to a certain pattern to form a two-dimensional coordinate grid. Each graphic has unique coding information for determining the pen tip pressure, movement trajectory and writing speed.

[0011] Specifically, the micro camera is arranged near the pen tip, and photographs the paper surface at a fixed frame rate of 30 frames per second, and the obtained image includes the paper area around the pen tip and the corresponding coded pattern; The collected images are preprocessed by graying them first, converting color images into grayscale images, and then performing noise reduction processing. The median filtering method is used to remove noise points in the image and improve the image quality.

[0012] Specifically, the optical tracking method is used for mark recognition and coordinate calculation, and a template matching algorithm is used to identify the coded graphics in the image. For each captured image, a pre-designed coded graphic template is matched with each area in the image. When the matching degree reaches 80%, the location of the coded graphic is determined; based on the pre-designed coordinate rule of the coded graphic on the paper, the coordinate position of the pen tip relative to the paper is calculated.

[0013] Specifically, the OCR recognition module processes the handwriting image collected by the handwriting collection module, including removing noise, enhancing contrast and adjusting brightness; The OCR recognition module analyzes the stroke morphology including a stroke morphology analysis algorithm, a stroke direction analysis algorithm and a stroke connection mode analysis algorithm; The stroke morphology analysis algorithm adopts the Canny edge detection algorithm, which first calculates the gradient amplitude and direction of the image and determines the edge points by finding the local maximum value of the gradient amplitude; then performs non-maximum suppression on the edge points, removes points that are not edges, and further refines the edges; finally, double threshold processing is performed to determine the true edge contour; The stroke direction analysis algorithm adopts the method of directional derivative, firstly, the gradient of the handwriting image is calculated to obtain the gradient direction and gradient value of each point in the image; then, the direction of the stroke is determined according to the gradient direction of each point on the stroke edge contour; if the gradient direction points to the left, the stroke direction is leftward; if the gradient direction points upward, the stroke direction is upward; or a skeleton extraction-based method is adopted, first the skeleton of the handwriting image is extracted to obtain the skeleton structure of the stroke; then, the direction of the stroke is determined according to the direction of the skeleton structure; if the skeleton structure extends from left to right, the stroke direction is rightward; The stroke connection mode analysis algorithm adopts a connected domain analysis method, firstly, the handwriting image is binarized to distinguish the strokes from the background; then, the stroke connection mode is determined by calculating the number and shape of connected date domains; if the number of connected domains is 1 and the shape is a continuous curve, the stroke is in a connected mode; if the number of connected domains is 2 and the shape is two intersecting curves, the stroke is in an intersecting mode; if the number of connected domains is 3 and the shape is three disconnected curves, the stroke is in a disconnected mode; or a method based on graph theory is adopted, the strokes are regarded as nodes in the graph, the connection relationship between the strokes is regarded as the edge in the graph, and the connection mode of the strokes is determined by analyzing the connectivity and structure of the graph; if the graph is connected and all nodes are connected by edges, the stroke is in a connected mode; if two nodes in the graph intersect by edges, the stroke is in an intersecting mode; if three nodes in the graph are disconnected by edges, the stroke is in a disconnected mode.

[0014] Specifically, the typo correction module is used to conduct a comprehensive check on the recognized text content, compare common words with the general vocabulary; compare professional terms in specific fields with the corresponding professional terminology vocabulary; correct complex typos, such as pictographic characters, by combining multiple recognition methods; for pictographic characters, determine whether they are typos by comparing the glyphs, that is, comparing the appearance structure and stroke order.

[0015] Specifically, the intelligent scoring module corrects the homework through the LLM correction module; First, the key elements and correct steps determine the algorithm: conduct in-depth semantic analysis of the original question text, and use the pre-trained language model in natural language processing technology to identify the key concepts, conditions and requirements in the question; for math questions, accurately identify the involved mathematical concepts, known conditions and solution requirements; Build a knowledge graph based on question types and subject areas: The knowledge graph contains common question structures, key knowledge points, problem-solving steps, and logical relationships; Match the standard problem-solving process with the knowledge graph to determine the key elements and correct steps for answering questions; ensure that key information and steps can be accurately extracted under different question types and subject areas.

[0016] Specifically, the intelligent scoring module analyzes the students' problem-solving process, cleans the text of the students' problem-solving process, removes irrelevant punctuation marks, spaces and special characters, and makes the text more standardized; then, a word vector model is used to map the words in the text to a low-dimensional vector space to obtain the feature representation of the text.

[0017] Specifically, the intelligent scoring module implements scoring and commenting on assignments: Scoring rules: Different weights are set for key steps and ideas according to the question type and subject area; for questions with a strong theoretical focus, the rationality of the solution ideas is emphasized, with a 60% weight given to the rationality of the ideas, and a 40% weight given to the key steps; for questions with a strong computational focus, the correctness of the key steps is emphasized, with a 70% weight given to the key steps, and a 30% weight given to the rationality of the ideas; The students’ problem-solving process is scored based on the degree of match with the standard problem-solving process: if all the key steps are correct, but there are some minor flaws in the problem-solving ideas, a certain percentage of the score will be deducted, 10%-20% of the total score; If the student's answer is correct but the problem-solving process is not standardized, points will be deducted according to the degree of non-standardization and the degree of impact on the result; if steps are missing or the order is wrong during the problem-solving process, the corresponding proportion of points will be deducted according to the severity of the missing or wrong step. If an important step is missing, 30% of the total score will be deducted, and if the order of steps is wrong, 15% of the total score will be deducted; Comment generation: Generate targeted comment content based on the strengths and weaknesses of students in the problem-solving process.

[0018] Specifically, the steps of error marking are as follows: Rule-based recognition: Divide the questions into regions, obtain question region information and obtain incorrect problem-solving process information from the LLM correction module, determine the text region or image region corresponding to the incorrect problem-solving process, and if the problem-solving steps in a certain question region do not conform to the standard problem-solving process, it will be identified as an incorrect region; For some specific error types, such as formula errors and logic errors, corresponding rules are set for identification; if a mathematical formula contains symbols or structures that do not conform to mathematical operation rules, the area where the formula is located is identified as an error area; Annotation algorithm: Annotate the corresponding text display area or the original image; for the text area, use a specific text annotation format; For the original image, a graphics drawing algorithm is used to draw marks on the error area. A common graphics drawing library is used to draw a circle on the image to mark the error area, ensuring that the position and size of the mark accurately correspond to the actual position and range of the error area.

[0019] Technical effects and advantages of the present invention: The present invention can greatly improve the efficiency and accuracy of examination review. By collecting the handwriting on the examination paper and quickly converting it into text data, the tedious process of collecting and collating traditional paper examination papers is avoided. The OCR recognition module can accurately identify the written content of the examination paper, and even in the face of different writing styles and handwriting characteristics, it can accurately convert it into text, reducing misjudgments caused by problems such as unclear handwriting; the typo correction module can ensure the text accuracy of the examinee's answer and avoid affecting the fairness of the score due to typos; the intelligent scoring module combines the question requirements, the standard problem-solving process and the examinee's answering process, and conducts in-depth analysis through the LLM large language model, which can give objective and accurate scores and targeted comments, thereby improving the accuracy and reliability of examination review; The comments in the intelligent scoring module can help candidates understand their performance in the exam and identify their strengths and weaknesses. For incorrect problem-solving processes, the red circle marking function allows candidates to intuitively see where their mistakes are, making it easier to conduct targeted review and improvement. At the same time, teachers can also use the data analysis function to develop personalized study plans and guidance programs for each candidate based on the student's exam situation, thereby improving the candidate's learning effect and exam results. The present invention can effectively ensure the fairness of the examination. The precise handwriting recognition and OCR recognition technology can ensure that the answers of all examinees can be accurately recognized and reviewed, avoiding the influence of human factors on the review results. The standardized processing flow of typo correction and intelligent scoring system makes the scoring more objective and fair, reducing the interference of subjective factors. At the same time, the examination data can also be encrypted and securely managed to ensure the confidentiality and security of the examination results. The application of the present invention provides strong support for the development of examination informatization; through the intelligent examination review system, digital management and analysis of examination data can be realized, providing more comprehensive and accurate examination information and data analysis reports for education departments and schools; it helps education departments and schools understand students' learning conditions and teaching effects, and provides a scientific basis for teaching reform and education decision-making; at the same time, it also provides candidates with a more convenient and efficient examination experience, adapting to the development trend of modern education and examinations.

[0020] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the system structure provided by the present invention.

[0022] Figure 2 It is a working principle diagram of the automatic homework correction provided by the present invention.

[0023] Figure 3The present invention provides a flowchart of automatic homework correction. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] like Figure 1-Figure 3 As shown, an automatic homework correction system provided by an embodiment of the present invention includes a handwriting collection module, an OCR recognition module, a typo correction module and an intelligent scoring module; The handwriting acquisition module is used to realize the collection of handwriting during the writing process. The handwriting acquisition module records the moving path of the pen tip through the optical tracking method of the micro camera, and the strain gauge sensor records the pressure change of the pen tip; accurately records various characteristics of the handwriting, including the position, strength and writing speed information of the strokes; The OCR recognition module recognizes the handwriting collected by the handwriting collection module, outputs text that can be processed and recognized by the computer, and performs multi-dimensional analysis of the handwriting through recognition algorithms and technologies; first, the shape of the strokes is analyzed, including the length, thickness, and curvature of the strokes; second, the direction of the strokes is determined; and finally, the connection mode of the strokes is determined, whether they are connected, intersecting, or disconnected; The wrong character correction module is used to correct wrong characters in the recognized text output by the OCR recognition module; using rich vocabulary resources, including general vocabulary and specific field professional terminology vocabulary, for complex wrong characters, a combination of multiple recognition methods is used to correct them, and the accuracy of text error correction is improved through the font comparison and pronunciation comparison methods; The intelligent scoring module scores and comments on the homework by comprehensively considering the text information output by the OCR recognition module in combination with the question requirements, the standard problem-solving process and the student's problem-solving process; a comprehensive intelligent scoring model is established, taking into account multiple factors such as the question requirements, the standard problem-solving process and the student's problem-solving process. In the scoring process, the correctness of the answers is used to analyze the students' problem-solving ideas, steps and methods used, give students a more accurate evaluation, and mark the errors.

[0026] In this embodiment, preferably, the handwriting acquisition module captures the pen tip pressure, motion trajectory and writing speed of the handwriting in real time through graphic coding, and converts the features into digital signals to form handwriting data; The graphic code is designed as a specific graphic coding method through optical marking. It is composed of tiny graphics with lines of different shapes and directions, and the graphics are arranged on the surface of the paper according to a certain rule to form a two-dimensional coordinate grid. Each graphic has unique coding information for determining the pen tip pressure, movement trajectory and writing speed. It should be noted that, through the two-dimensional coordinate grid formed, each graphic has unique coding information, which can be used to determine its position on the paper, making it easier to determine the pen tip pressure, movement trajectory and writing speed.

[0027] In this embodiment, preferably, the micro camera is arranged near the pen tip, and the paper surface is photographed at a fixed frame rate of 30 frames per second, and the obtained image includes the paper area around the pen tip and the corresponding coded pattern; Preprocess the collected images by graying them first, converting color images into grayscale images, and then performing noise reduction, using the median filter method to remove noise points in the image and improve image quality; It should be noted that the image captured by the micro camera includes the paper area around the pen tip and the corresponding coded pattern, and the image is filtered to improve the image quality; The steps of median filtering are as follows: Choose the neighborhood size: select an odd-sized window (e.g. 3×3, 5×5, etc.); this window will be used to determine the new value of each pixel; Define neighborhood: For a pixel point p(x,y) in the image, its neighborhood is an area around the pixel point, and the set of pixels in the window is denoted as N(x,y); Sorting neighborhood pixels: Sort all pixel values ​​in the neighborhood area to obtain a sorted pixel value sequence S=[v1,v2,...,v n ], where n is the number of pixels in the neighborhood; Select the median: The median is the value in the middle of the sorted sequence, denoted by v median , if the neighborhood size is an odd number, then the median is the element in the middle of the sequence; the formula is:

[0028] Where n is the number of pixels in the window, usually n=k×k, where k is the size of the window (usually an odd number).

[0029] Replace pixel value: use median value v median Replace the value of the original pixel p(x,y).

[0030] In this embodiment, preferably, the optical tracking method is used for mark recognition and coordinate calculation, and the coded pattern in the image is recognized by using a template matching algorithm. For each captured image, a pre-designed coded pattern template is matched with each area in the image. When the matching degree reaches 80%, the location of the coded pattern is determined; according to the pre-designed coordinate rule of the coded pattern on the paper, the coordinate position of the pen tip relative to the paper is calculated; It should be noted that the coordinate position of the pen tip relative to the paper is calculated according to the pre-designed coordinate rule of the coded pattern on the paper. For example, if the coordinates of a coded pattern on the paper are (x0, y0), and the position of the coded pattern in the image is determined by the template matching algorithm; For example, if the position in the image is (x1, y1), then the coordinates (x, y) of the pen tip can be calculated by the relative position relationship. Assuming that the distance between the camera and the paper is d, and the focal length of the camera is known to be f, the coordinates (x, y) of the pen tip can be calculated by the principle of similar triangles: x=x1*(d / f)+x0, y=y1*(d / f)+y0. By continuously recording the coordinate position of the pen tip, the moving path of the pen tip on the paper can be obtained.

[0031] In this embodiment, the OCR recognition module processes the handwriting image collected by the handwriting collection module, including removing noise, enhancing contrast and adjusting brightness; The OCR recognition module analyzes the stroke morphology including a stroke morphology analysis algorithm, a stroke direction analysis algorithm and a stroke connection mode analysis algorithm; The stroke morphology analysis algorithm adopts the Canny edge detection algorithm, which first calculates the gradient amplitude and direction of the image and determines the edge points by finding the local maximum value of the gradient amplitude; then performs non-maximum suppression on the edge points, removes points that are not edges, and further refines the edges; finally, double threshold processing is performed to determine the true edge contour; By analyzing the curvature, length, thickness and other characteristics of the edge contour, the shape of the stroke can be determined. Specifically, by calculating the radius of each point on the edge contour, the curvature of the stroke can be determined; by counting the number of pixels of the edge contour, the thickness of the stroke can be obtained; by measuring the length of the edge contour, the shape of the stroke can be determined. On this basis, shape descriptor-based methods can be used, such as the Hu moment, which can quantitatively describe the shape of the stroke. By calculating the value of the Hu moment, the shape characteristics of the stroke, such as symmetry, flatness, etc., can be obtained, thereby assisting in determining the shape of the stroke; The stroke direction analysis algorithm adopts the method of directional derivative, firstly, the gradient of the handwriting image is calculated to obtain the gradient direction and gradient value of each point in the image; then, the direction of the stroke is determined according to the gradient direction of each point on the stroke edge contour; if the gradient direction points to the left, the stroke direction is leftward; if the gradient direction points upward, the stroke direction is upward; or a skeleton extraction-based method is adopted, first the skeleton of the handwriting image is extracted to obtain the skeleton structure of the stroke; then, the direction of the stroke is determined according to the direction of the skeleton structure; if the skeleton structure extends from left to right, the stroke direction is rightward; The stroke connection mode analysis algorithm adopts a connected domain analysis method, firstly, the handwriting image is binarized to distinguish the strokes from the background; then, the stroke connection mode is determined by calculating the number and shape of connected date domains; if the number of connected domains is 1 and the shape is a continuous curve, the stroke is in a connected mode; if the number of connected domains is 2 and the shape is two intersecting curves, the stroke is in an intersecting mode; if the number of connected domains is 3 and the shape is three disconnected curves, the stroke is in a disconnected mode; or a method based on graph theory is adopted, the strokes are regarded as nodes in the graph, the connection relationship between the strokes is regarded as the edge in the graph, and the connection mode of the strokes is determined by analyzing the connectivity and structure of the graph; if the graph is connected and all nodes are connected by edges, the stroke is in a connected mode; if two nodes in the graph intersect by edges, the stroke is in an intersecting mode; if three nodes in the graph are disconnected by edges, the stroke is in a disconnected mode; It should be noted that the above algorithm is used to calculate the direction and connection of the strokes, so as to facilitate the analysis and recognition of the handwriting. The calculation formula for noise removal is as follows: , in, Represented as the original data corresponding to time point t in the data information, Represented as a time point in the data information The corresponding smoothing value is is a smoothing factor used to assign weights to new data points and historical data points, and Indicates data information in scale and location The transformation structure on is the conjugate of the wavelet transform, is the scale parameter, is the translation parameter.

[0032] In this embodiment, preferably, the wrong character correction module is used to perform a comprehensive check on the recognized text content, compare common words with a general vocabulary; compare professional terms in a specific field with corresponding professional terminology vocabulary; correct complex wrong characters, such as similar characters, by combining multiple recognition methods; for similar characters, determine whether they are wrong characters by comparing their shapes, that is, comparing their appearance structure and stroke order; for similar characters, determine whether they are wrong characters by comparing their pronunciations, that is, comparing their pronunciations; It should be noted that by analyzing and identifying characters with similar shapes and similar pronunciations, it is convenient to determine errors in the text and correct them, but the errors in the text are not marked.

[0033] In this embodiment, preferably, the intelligent scoring module corrects the homework through the LLM correction module; First, the key elements and correct steps determine the algorithm: conduct in-depth semantic analysis of the original question text, and use the pre-trained language model in natural language processing technology to identify the key concepts, conditions and requirements in the question; for math questions, accurately identify the involved mathematical concepts, known conditions and solution requirements; Build a knowledge graph based on question types and subject areas: The knowledge graph contains common question structures, key knowledge points, problem-solving steps, and logical relationships; Match the standard problem-solving process with the knowledge graph to determine the key elements and correct steps for answering questions; ensure that key information and steps can be accurately extracted in different question types and subject areas; It should be noted that the original question text is deeply semantically analyzed through the underlying technical principles of the BERT and GPT series to identify the key concepts, conditions and requirements in the question. For math questions, the involved math concepts, such as functions and geometric figures, are accurately identified; known conditions, such as numerical values, the length of the side of the figure, etc.; and solution requirements, such as finding the area, proving equations, etc.; The knowledge graph contains common question structures, key knowledge points, problem-solving steps, and the logical relationships between them. For solving quadratic equations, the knowledge graph contains key knowledge points such as the definition of the equation, calculation of the discriminant, and application of the root-finding formula, as well as their order. Matching the standard problem-solving process with the knowledge graph makes it easier to extract the key elements and correct steps of answering questions, and to analyze and process the problem.

[0034] In this embodiment, preferably, the intelligent scoring module analyzes the student's problem-solving process, cleans the text of the student's problem-solving process, removes irrelevant punctuation marks, spaces and special characters, and makes the text more standardized; then, a word vector model is used to map the words in the text to a low-dimensional vector space to obtain a feature representation of the text; It should be noted that the word vector models of Word2Vec and GloVe are used to map the words in the text to a low-dimensional vector space to obtain the feature representation of the text; the feature representation can capture the semantic relationship between words and provide a basis for subsequent analysis.

[0035] In this embodiment, preferably, the intelligent scoring module implements scoring and commenting on the assignments: Scoring rules: Different weights are set for key steps and ideas according to the question type and subject area; for questions with a strong theoretical focus, the rationality of the solution ideas is emphasized, with a 60% weight given to the rationality of the ideas, and a 40% weight given to the key steps; for questions with a strong computational focus, the correctness of the key steps is emphasized, with a 70% weight given to the key steps, and a 30% weight given to the rationality of the ideas; The students’ problem-solving process is scored based on the degree of match with the standard problem-solving process: if all the key steps are correct, but there are some minor flaws in the problem-solving ideas, a certain percentage of the score will be deducted, 10%-20% of the total score; If the student's answer is correct but the problem-solving process is not standardized, points will be deducted according to the degree of non-standardization and the degree of impact on the result; if steps are missing or the order is wrong during the problem-solving process, the corresponding proportion of points will be deducted according to the severity of the missing or wrong step. If an important step is missing, 30% of the total score will be deducted, and if the order of steps is wrong, 15% of the total score will be deducted; Comment generation: Generate targeted comments based on the strengths and weaknesses of students in solving problems; It should be noted that the grading criteria can be used to accurately grade students' homework, making it easier to automatically grade papers or homework; Generate targeted comments based on the strengths and weaknesses of students in the problem-solving process; if students use novel methods or ideas in a certain step, give affirmation and encouragement; for example, "The [specific method] you used in this question is very creative and worth learning"; if students have shortcomings, clearly point out the problems and suggestions for improvement; for example, "You made [specific error description] in the problem-solving process, and you are advised to [suggested improvement content]"; in this way, provide students with targeted feedback to promote their learning and growth.

[0036] In this embodiment, preferably, the steps of error marking are as follows: Rule-based recognition: Divide the questions into regions, obtain question region information and obtain incorrect problem-solving process information from the LLM correction module, determine the text region or image region corresponding to the incorrect problem-solving process, and if the problem-solving steps in a certain question region do not conform to the standard problem-solving process, it will be identified as an incorrect region; For some specific error types, such as formula errors and logic errors, corresponding rules are set for identification; if a mathematical formula contains symbols or structures that do not conform to mathematical operation rules, the area where the formula is located is identified as an error area; Annotation algorithm: Annotate the corresponding text display area or the original image; for the text area, use a specific text annotation format; For the original image, a graphics drawing algorithm is used to draw a mark on the error area. A common graphics drawing library is used to draw a circle on the image to mark the error area, ensuring that the position and size of the mark accurately correspond to the actual position and range of the error area. It should be noted that by standardizing the wrong content, candidates can intuitively see where their mistakes are, and conduct targeted review and improvement. At the same time, teachers can also make corrections based on students' test results; For text areas, use a specific text annotation format, add specific symbols before and after the error area, [error start] and [error end]) or color marking (change the text color of the error area to red); For the original image, a common graphics drawing library (OpenCV) is used to draw a circle (red circle) on the image to mark the error area.

[0037] The principle of dividing the topic into regions: Determine the overall question area: When generating the test paper, for each question, check the starting coordinates (x, y) of the question elements (including question text, graphics, and all other elements that constitute the question) on the paper. At the same time, record the width and height of the question elements. Use this information to determine the area range of the entire question, that is, the coordinates of the upper left corner are (x, y), and the coordinates of the lower right corner are (x+w, y+h); Division of answering areas for different question types: Division of answering areas for multiple choice questions / true / false questions: Option element positioning: For multiple-choice and true-or-false questions (taking the question format of the subject network as an example, assuming that the options are included in the question), first find the option elements in the question; the options can be identified by text feature matching or specific tags; Answer area marking: insert a square box before each option as the answer area mark. The size of this box can be determined according to the pre-set standard size or the proportional relationship with the question text; at the same time, the coordinate information of each box is recorded for subsequent identification and processing; Fill-in-the-blank question answering area division: Specific element location: Look for underscores and brackets in the question, which are common answer hints for fill-in-the-blank questions. If these elements exist in the question, use them as a basis to determine the answer area. Adding styles to the answer area and obtaining coordinates: Add specific styles (such as changing the color, making it bold, etc.) and marks (such as adding a small icon to indicate the answer location) to the underline and bracket elements found, and then obtain the coordinates of these elements as the answer area. If there are no underline and bracket elements in the question, create a certain number of answer areas (determined by the number of answers or the requirements of the question) under the question based on the possible locations of the answers. Each answer area can be a rectangular area, and its size and position must also be reasonably set and the coordinate information must be recorded.

[0038] Essay question answering area division: Leave a large blank space under the question as the answer area. The size of the blank area can be determined based on the question type, the expected number of words in the answer, and the page layout. At the same time, record the coordinate information of the blank area to accurately identify and process the correspondence between the answer content and the question area.

[0039] Handwriting collection Structure: A specific smart pen cooperates with a signal processing unit. The smart pen has built-in pressure sensors, position sensors, speed sensors, etc.

[0040] Function: Accurately collect the physical characteristics of the user's handwriting and provide raw data for subsequent processing.

[0041] Principle: When students write, the smart pen uses internal sensors to capture the physical characteristics of the handwriting, such as pen tip pressure, movement trajectory, and writing speed. These characteristics are transmitted to the signal processing unit and converted into digital signals. After forming handwriting data, they are transmitted to OCR recognition. When writing a stroke, the pressure sensor senses the writing force, the position sensor records the movement path of the pen tip, and the speed sensor monitors the writing speed. After processing, the handwriting data of the stroke is formed.

[0042] OCR recognition Structure: includes image preprocessing module, multi-dimensional analysis engine, character matching module and coordinate marking unit.

[0043] Function: Recognize handwriting data and output recognized text with coordinate markings.

[0044] Principle: After receiving the handwriting data from the handwriting collection, the image preprocessing module first removes noise and enhances contrast to improve the clarity of the handwriting image. The multi-dimensional analysis engine analyzes the handwriting shape (such as length, thickness, curvature, etc.), direction (left, right, upward or downward) and connection mode (connected, intersecting or disconnected). The character matching module matches the analysis results with the predefined character model, recognizes the characters and combines them into text. The coordinate marking unit also marks the coordinates of each line of text to distinguish the question area. The recognized text and coordinate information are transmitted to the typo correction.

[0045] Typos correction Structure: It consists of a general vocabulary, a professional terminology vocabulary, a glyph comparison unit and a glyph comparison unit.

[0046] Function: Check and correct typos in OCR recognition results.

[0047] Principle: After receiving the text from OCR recognition, common words are compared with the general word library, and professional terms in specific fields are compared with the professional terminology word library. For complex typos such as similar-looking characters and similar-sounding characters, the shape comparison unit determines whether it is a typo by comparing the shape structure and stroke order of the characters, and the pronunciation comparison unit assists in the judgment by comparing the pronunciation of the characters. The corrected text is sent to the intelligent scoring.

[0048] Smart Scoring: Structure: It consists of a question analysis module, a problem-solving process evaluation module, a scoring and comment generation module, and an interactive interface with the LLM (large language model).

[0049] Function: Comprehensively consider the requirements of the questions, the standard problem-solving process and the students' problem-solving process, and give accurate scores and detailed comments.

[0050] Principle: After receiving the corrected answer text, question requirements, standard problem-solving process and other information from the typo correction part, the question analysis module determines the key elements and correct steps of the answer based on the question requirements and standard problem-solving process. The problem-solving process evaluation module deeply analyzes whether the student's problem-solving process meets the standards and whether the method is correct, and considers whether the problem-solving ideas are reasonable. Then, the question and the corrected answer text are passed to the LLM large language model for in-depth analysis. The LLM large language model uses its powerful language understanding and processing capabilities to conduct a comprehensive comparison of the question and the student's answer text, and further explores the advantages and disadvantages of the student's problem-solving process. The scoring and comment generation module gives more accurate and targeted scores and comments based on the analysis results of the LLM large language model and the question score. If an incorrect problem-solving process is found, it will be marked with a red circle in the original picture or the corresponding text display area so that students can clearly see the error, so as to better understand their own answer situation and promote learning and improvement.

[0051] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic homework grading system, characterized in that: It includes handwriting collection module, OCR recognition module, typo correction module and intelligent scoring module; The handwriting acquisition module is used to realize the collection of handwriting during the writing process. The handwriting acquisition module records the moving path of the pen tip through the optical tracking method of the micro camera, and the strain gauge sensor records the pressure change of the pen tip; accurately records various characteristics of the handwriting, including the position, strength and writing speed information of the strokes; The OCR recognition module recognizes the handwriting collected by the handwriting collection module, outputs text that can be processed and recognized by the computer, and performs multi-dimensional analysis of the handwriting through recognition algorithms and technologies; first, the shape of the strokes is analyzed, including the length, thickness, and curvature of the strokes; second, the direction of the strokes is determined; and finally, the connection mode of the strokes is determined, whether they are connected, intersecting, or disconnected; The wrong character correction module is used to correct wrong characters in the recognized text output by the OCR recognition module; using rich vocabulary resources, including general vocabulary and specific field professional terminology vocabulary, for complex wrong characters, a combination of multiple recognition methods is used to correct them, and the accuracy of text error correction is improved through the font comparison and pronunciation comparison methods; The intelligent scoring module scores and comments on the homework by comprehensively considering the text information output by the OCR recognition module in combination with the question requirements, the standard problem-solving process and the student's problem-solving process; a comprehensive intelligent scoring model is established, taking into account multiple factors such as the question requirements, the standard problem-solving process and the student's problem-solving process. In the scoring process, the correctness of the answers is used to analyze the students' problem-solving ideas, steps and methods used, give students a more accurate evaluation, and mark the errors.

2. The automatic homework marking system according to claim 1, characterized in that: The handwriting acquisition module captures the pen tip pressure, motion trajectory and writing speed of the handwriting in real time through graphic coding, and converts the features into digital signals to form handwriting data; The graphic code is designed as a specific graphic coding method through optical marking. It is composed of tiny graphics composed of lines of different shapes and directions, and the graphics are arranged on the paper surface according to a certain pattern to form a two-dimensional coordinate grid. Each graphic has unique coding information for determining the pen tip pressure, movement trajectory and writing speed.

3. The automatic homework marking system according to claim 1, characterized in that: The micro camera is arranged near the pen tip and photographs the paper surface at a fixed frame rate of 30 frames per second. The obtained image includes the paper area around the pen tip and the corresponding coded pattern. The collected images are preprocessed by graying them first, converting color images into grayscale images, and then performing noise reduction processing. The median filtering method is used to remove noise points in the image and improve the image quality.

4. The automatic homework marking system according to claim 1, characterized in that: The optical tracking method is used for mark recognition and coordinate calculation. The template matching algorithm is used to identify the coded pattern in the image. For each captured image, the pre-designed coded pattern template is matched with each area in the image. When the matching degree reaches 80%, the location of the coded pattern is determined; according to the pre-designed coordinate rule of the coded pattern on the paper, the coordinate position of the pen tip relative to the paper is calculated.

5. The automatic homework marking system according to claim 1, characterized in that: The OCR recognition module processes the handwriting image collected by the handwriting collection module, including removing noise, enhancing contrast and adjusting brightness; The OCR recognition module analyzes the stroke morphology including a stroke morphology analysis algorithm, a stroke direction analysis algorithm and a stroke connection mode analysis algorithm; The stroke morphology analysis algorithm adopts the Canny edge detection algorithm, which first calculates the gradient amplitude and direction of the image and determines the edge points by finding the local maximum value of the gradient amplitude; then performs non-maximum suppression on the edge points, removes points that are not edges, and further refines the edges; finally, double threshold processing is performed to determine the true edge contour; The stroke direction analysis algorithm adopts the method of directional derivative, firstly, the gradient of the handwriting image is calculated to obtain the gradient direction and gradient value of each point in the image; then, the direction of the stroke is determined according to the gradient direction of each point on the stroke edge contour; if the gradient direction points to the left, the stroke direction is leftward; if the gradient direction points upward, the stroke direction is upward; or a skeleton extraction-based method is adopted, first the skeleton of the handwriting image is extracted to obtain the skeleton structure of the stroke; then, the direction of the stroke is determined according to the direction of the skeleton structure; if the skeleton structure extends from left to right, the stroke direction is rightward; The stroke connection mode analysis algorithm adopts a connected domain analysis method, firstly, the handwriting image is binarized to distinguish the strokes from the background; then, the stroke connection mode is determined by calculating the number and shape of connected date domains; if the number of connected domains is 1 and the shape is a continuous curve, the stroke is in a connected mode; if the number of connected domains is 2 and the shape is two intersecting curves, the stroke is in an intersecting mode; if the number of connected domains is 3 and the shape is three disconnected curves, the stroke is in a disconnected mode; or a method based on graph theory is adopted, the strokes are regarded as nodes in the graph, the connection relationship between the strokes is regarded as the edge in the graph, and the connection mode of the strokes is determined by analyzing the connectivity and structure of the graph; if the graph is connected and all nodes are connected by edges, the stroke is in a connected mode; if two nodes in the graph intersect by edges, the stroke is in an intersecting mode; if three nodes in the graph are disconnected by edges, the stroke is in a disconnected mode.

6. The automatic homework marking system according to claim 1, characterized in that: The typo correction module is used to conduct a comprehensive check on the recognized text content, compare common words with the general vocabulary; compare professional terms in a specific field with the corresponding professional terminology vocabulary; correct complex typos, such as pictographic characters, by combining multiple recognition methods; for pictographic characters, determine whether they are typos by comparing the shape of the characters, that is, comparing the appearance structure and stroke order.

7. The automatic homework marking system according to claim 1, characterized in that: The intelligent scoring module corrects the homework through the LLM correction module; First, the key elements and correct steps determine the algorithm: conduct in-depth semantic analysis of the original question text, and use the pre-trained language model in natural language processing technology to identify the key concepts, conditions and requirements in the question; for math questions, accurately identify the involved mathematical concepts, known conditions and solution requirements; Build a knowledge graph based on question types and subject areas: The knowledge graph contains common question structures, key knowledge points, problem-solving steps, and logical relationships; Match the standard problem-solving process with the knowledge graph to determine the key elements and correct steps for answering questions; ensure that key information and steps can be accurately extracted under different question types and subject areas.

8. The automatic homework marking system according to claim 1, characterized in that: The intelligent scoring module analyzes the student's problem-solving process, cleans the text of the student's problem-solving process, removes irrelevant punctuation marks, spaces and special characters, and makes the text more standardized; then, a word vector model is used to map the words in the text to a low-dimensional vector space to obtain the feature representation of the text.

9. The automatic homework marking system according to claim 1, characterized in that: The intelligent scoring module can score and comment on the assignments: Scoring rules: Different weights are set for key steps and ideas according to the question type and subject area; for questions with a strong theoretical focus, the rationality of the solution ideas is emphasized, with a 60% weight given to the rationality of the ideas, and a 40% weight given to the key steps; for questions with a strong computational focus, the correctness of the key steps is emphasized, with a 70% weight given to the key steps, and a 30% weight given to the rationality of the ideas; The students’ problem-solving process is scored based on the degree of match with the standard problem-solving process: if all the key steps are correct, but there are some minor flaws in the problem-solving ideas, a certain percentage of the score will be deducted, 10%-20% of the total score; If the student's answer is correct but the problem-solving process is not standardized, points will be deducted according to the degree of non-standardization and the degree of impact on the result; if steps are missing or the order is wrong during the problem-solving process, the corresponding proportion of points will be deducted according to the severity of the missing or wrong step. If an important step is missing, 30% of the total score will be deducted, and if the order of steps is wrong, 15% of the total score will be deducted; Comment generation: Generate targeted comment content based on the strengths and weaknesses of students in the problem-solving process.

10. The automatic homework marking system according to claim 7, characterized in that: The steps of error marking are as follows: Rule-based recognition: Divide the questions into regions, obtain question region information and obtain incorrect problem-solving process information from the LLM correction module, determine the text region or image region corresponding to the incorrect problem-solving process, and if the problem-solving steps in a certain question region do not conform to the standard problem-solving process, it will be identified as an incorrect region; For some specific error types, such as formula errors and logic errors, corresponding rules are set for identification; if a mathematical formula contains symbols or structures that do not conform to mathematical operation rules, the area where the formula is located is identified as an error area; Annotation algorithm: Annotate the corresponding text display area or the original image; for the text area, use a specific text annotation format; For the original image, a graphics drawing algorithm is used to draw marks on the error area. A common graphics drawing library is used to draw a circle on the image to mark the error area, ensuring that the position and size of the mark accurately correspond to the actual position and range of the error area.

Images