Preschool education interaction system based on image big data

Abstract

The invention discloses a preschool education interaction system based on image big data, and relates to the technical field of preschool education. The image data acquisition module acquires images in multiple scenes by using high-definition equipment, has intelligent tracking and image enhancement functions, and accurately captures the state of an infant; the image big data storage module adopts distributed and cloud storage; the image analyzing and processing module accurately analyzes image information by applying a deep learning algorithm and combining migration and multi-task learning technologies; the interactive teaching module generates personalized teaching contents and tasks according to an analysis result, and has a self-adaptive learning mechanism; the user interaction module enables children and teachers to interact conveniently through a touch screen and an intelligent bracelet, and the intelligent bracelet can transmit physiological data of the children in real time. According to the invention, images are accurately acquired and efficiently analyzed, personalized teaching is provided, and the learning effect is improved; the social module promotes child communication, and the parent module strengthens family communication; the advanced technology guarantees data safety and assists preschool children to grow in all directions.

Description

Technical Field

[0001] The present invention relates to the technical field of preschool education systems, and in particular to a preschool education interactive system based on image big data. Background Art

[0002] Preschool education is a critical stage in children's growth and plays a vital role in the cognitive, emotional, social and other development of children. However, the traditional preschool education model has many limitations and cannot meet the diverse learning needs of modern children and parents' expectations for quality education.

[0003] Traditional preschool education mainly relies on teachers’ experience and limited teaching resources. The teaching methods are relatively simple and it is difficult to provide personalized education for each child. Teachers often use the same teaching content and methods in the classroom and fail to pay enough attention to the differences in each child’s interests, abilities and learning progress. As a result, some children may not be able to fully realize their potential, while others may find learning difficult or lacking challenges.

[0004] In terms of evaluating children's learning, traditional methods mainly rely on teachers' subjective observations and regular simple tests, lacking comprehensive, objective, and real-time data support. This makes it difficult for teachers to accurately grasp each child's learning status and progress, and it is impossible to adjust teaching strategies in time to meet children's needs. At the same time, parents have limited knowledge of their children's learning and living conditions in kindergartens, and can only obtain partial information through teachers' verbal feedback, making it difficult for them to deeply participate in their children's education process.

[0005] With the rapid development of information technology, technologies such as image recognition and big data analysis have been widely used in various fields. However, in the field of preschool education, the application of these technologies is still relatively lagging. Although there are some educational software and online courses, most of them lack in-depth understanding and application of children's actual learning scenarios, and cannot capture children's learning behaviors and emotional states in real time, making it difficult to achieve truly personalized education.

[0006] In addition, preschool children are at an important stage of developing their image thinking. As an intuitive and vivid information carrier, images play an important role in promoting children's cognition and learning. However, the existing preschool education system rarely makes full use of image big data to carry out teaching activities, and is unable to mine valuable information from a large amount of image data to provide children with learning content and interactive experiences that are more in line with their needs.

[0007] Therefore, developing a preschool education interactive system based on image big data has important practical significance. The system can collect image data of children in their study and life in real time, use advanced image analysis and big data processing technology to deeply understand the learning characteristics and needs of each child, and provide personalized teaching content and interactive tasks for them. At the same time, the system can also strengthen the interaction and communication between teachers, parents and children, promote home-school co-education, improve the quality and effect of preschool education, and promote the development of preschool education in the direction of intelligence and personalization. Summary of the invention

[0008] The present invention proposes a preschool education interactive system based on image big data to solve the problems mentioned in the above-mentioned prior art.

[0009] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: a preschool education interactive method based on image big data, comprising:

[0010] Image data collection steps: Use the cameras in the classroom and smart tablet devices to collect image data of children in classroom activities, game interactions, and life scenes I = {i1, i2, ..., i n}, where n is the number of images, i j represents the jth image;

[0011] Data preprocessing step: noise reduction is performed on the collected image data, using the mean filter algorithm, for each pixel x in the image ij , where i represents the row index, j represents the column index, and the processed pixel value Remove salt and pepper noise and Gaussian noise from the image; then perform image enhancement, using the adaptive histogram equalization CLAHE algorithm, divide the image into multiple sub-blocks, and redistribute the grayscale values ​​of the pixels in each sub-block to enhance the contrast and details of the image; finally, perform normalization to unify the pixel values ​​of the image data into the range of [0, 1]. For the pixel value, the normalized value Where min(X) and max(X) are the minimum and maximum values ​​of all pixel values ​​in the image, respectively;

[0012] Feature extraction step: Use convolutional neural network CNN to extract features from the preprocessed image. Suppose the input image is I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conv The value at a certain position in Among them, m and n are the indices in the convolution kernel. After the pooling layer, for the pooling window, the output feature map F pool The value at a certain position in Finally, the feature vector F = {f1, f2, ..., f m}, where m is the number of features, fk is the kth feature;

[0013] Behavior analysis step: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s}, where s is the number of behavior patterns, pl is the feature vector corresponding to the lth behavior pattern, and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library. Determine the emotional state and learning status of young children;

[0014] Interaction strategy generation steps: Based on the behavior analysis results, the reinforcement learning algorithm is used to define the state S as the behavior state of the child, the action A as various interaction strategies, and the reward R as the effect feedback after the interaction. Generate an interactive strategy, where γ is a discount factor in [0, 1] and t is the time step.

[0015] Interactive feedback implementation steps: The generated interactive strategies are fed back to teachers or directly pushed to smart teaching devices. Teachers adjust their teaching methods based on the strategies, and smart devices play corresponding teaching content and conduct interactive games based on the strategies.

[0016] Furthermore, the method further comprises the following steps:

[0017] Scene recognition step: The image data is analyzed by the support vector machine (SVM) algorithm to identify the current educational scene, including classroom teaching, game activities, and rest time. For the input image feature vector x, SVM solves the optimization problem The constraint is y i (w T φ(x i )+b)≥1-ξ i ,ξ i ≥0, where w is the weight vector, b is the bias, ξ i is the slack variable, C is the penalty parameter, y i For sample x i The category label of φ(x i ) is the sample x i Functions that map to higher-dimensional spaces.

[0018] Knowledge graph construction steps: Integrate preschool education related knowledge, including children's cognitive development theory, teaching methods, knowledge points, and construct a knowledge graph; the knowledge graph consists of nodes N = {n1, n2, ..., n q} represents knowledge elements and edges E = {e1, e2, ..., e r} represents the relationship between knowledge elements. For node n i , and its updated feature representation Where σ is the activation function, N i is the set of neighbor nodes of node i, d i and d j are the degrees of nodes i and j respectively, W l and b l is the weight matrix and bias vector of the lth layer.

[0019] Furthermore, in the data preprocessing step, during image enhancement, an adaptive histogram equalization CLAHE algorithm is used to perform histogram equalization according to the characteristics of the local area of ​​the image.

[0020] Furthermore, in the scene recognition step, transfer learning technology is used to migrate the model pre-trained on the image dataset to the preschool education scene recognition task, reducing training time and data requirements.

[0021] Furthermore, in the knowledge graph construction step, the graph convolutional neural network GCN is used to update and optimize the knowledge graph.

[0022] A system using the preschool education interactive method based on image big data, comprising:

[0023] Image data acquisition module: collects image data of children in classroom activities, game interactions, and life scenes through cameras and smart tablet devices in the classroom. n};

[0024] Data preprocessing module: denoise, enhance and normalize the collected image data; use the mean filter algorithm to calculate the average value of each pixel x in the image. ij , where i represents the row index, j represents the column index, and the processed pixel value Remove noise; use the adaptive histogram equalization CLAHE algorithm to enhance the image; normalize the image pixel values ​​to the range [0, 1], and for the pixel value x, the normalized value Where min(X) and max(X) are the minimum and maximum values ​​of all pixel values ​​in the image respectively;

[0025] Feature extraction module: Use convolutional neural network CNN to extract features from the preprocessed image. Let the input image be I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conv The value at a certain position in Among them, m and n are the indices in the convolution kernel, and then after the pooling layer, for the pooling window, the output feature map F pool The value at a certain position in Generate feature vector F = {f1, f2, ..., f m};

[0026] Behavior analysis module: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s}Where s is the number of behavior patterns, pl is the feature vector corresponding to the lth behavior pattern, and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library Determine the child's emotional state, engagement, and learning status;

[0027] Interaction strategy generation module: Based on the behavior analysis results, the reinforcement learning algorithm is used to define the state S as the behavior state of the child, the action A as various interaction strategies, and the reward R as the effect feedback after the interaction. Generate an interactive strategy, where γ is a discount factor, ranging from [0, 1], and t is the time step;

[0028] Interactive feedback implementation module: The generated interactive strategies are fed back to teachers or directly pushed to intelligent teaching devices. Teachers adjust teaching methods based on the strategies, and intelligent devices play corresponding teaching content, conduct interactive games, etc. according to the strategies to achieve effective interaction with children and collect feedback data from children.

[0029] Furthermore, the following modules are also included:

[0030] Scene recognition module: Use the support vector machine (SVM) algorithm to analyze image data and identify current educational scenes, including classroom teaching, game activities, and rest time. For the input image feature vector, SVM solves the optimization problem. The constraint is y i (w T φ(x i )+b)≥1-ξ i ,ξ i ≥0, where w is the weight vector, b is the bias, ξ i is the slack variable, C is the penalty parameter, y i For sample x i The category label of φ(x i ) is the sample x i Functions that map to higher-dimensional spaces.

[0031] Interaction strategy generation module: Genetic algorithm is used to optimize the interaction strategy; define the strategy population P = {p1, p2, ..., p z}, where p i is an interactive strategy individual, and each strategy individual is encoded; by calculating the fitness function Fitness(p i ) evaluates the adaptability of each strategy individual to the current behavior state of the child, and performs selection operations based on fitness, using the roulette wheel selection method.i Probability of being selected

[0032] Privacy protection module: Federated learning technology is used to encrypt image data on the data collection device side. The homomorphic encryption algorithm is used. For the original image data x, the encrypted data y = Enc(x). During model training, each device side is trained based on the local encrypted data, uploads the model parameters, and updates the global model through the federated aggregation algorithm; the server collects the model parameters θ of each device side i , calculate the global model parameters Where N is the number of participating devices.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] In terms of personalized teaching, the image analysis and processing module uses advanced algorithms to accurately identify children's expressions, movements and works. The interactive teaching module dynamically adjusts the teaching content and difficulty based on these analysis results through the Bayesian learning algorithm, realizing truly personalized education, which can fully stimulate the learning potential of each child and improve learning effects. The system has powerful data acquisition and storage capabilities. The image data acquisition module uses intelligent tracking and image enhancement technology to obtain clear image data in various complex environments, providing rich and accurate materials for subsequent analysis. The image big data storage module uses hash algorithms and redundant backup strategies to ensure efficient storage and safe and reliable massive image data. The social interaction module builds a social network for children, promotes communication and cooperation between children through social influence formulas, and cultivates children's social skills. The parent participation module provides parents with a convenient remote access channel, uses the AES encryption algorithm to ensure data security, and regularly pushes learning reports and suggestions generated based on the comprehensive evaluation formula, strengthening communication and collaboration between home and school, and realizing home-school co-education. In terms of technical application, the transfer learning and multi-task learning technologies of the image analysis and processing module have greatly improved analysis efficiency and reduced resource consumption. The adaptive learning mechanism of the interactive teaching module has a high adjustment accuracy and can effectively meet the learning needs of children. The touch screen and smart bracelet devices of the user interaction module provide a convenient and efficient way of interaction for children and teachers, improving the user experience. In summary, this system has comprehensively improved the quality and effectiveness of preschool education by making full use of image big data and advanced technology, and provided strong support for the growth and development of children. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 A schematic block diagram of a preschool education interactive system based on image big data proposed by the present invention;

[0036] Figure 2This is a schematic block diagram of a preschool education interactive method based on image big data proposed by the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0039] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. In addition, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, and it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0040] Reference Figure 1-2 :A preschool education interactive method based on image big data, the steps are as follows:

[0041] Image data collection steps: Through the cameras, smart tablets and other devices in the classroom, we can collect all-round image data of children in classroom activities, game interactions, and life scenes. n}, where n is the number of images, i jrepresents the jth image. These images cover information such as children’s expressions, movements, and classroom participation, providing raw materials for subsequent analysis.

[0042] Use a variety of devices to comprehensively collect various images of young children in preschool education scenarios, from daily activities to classroom performance, from facial expression changes to body movements, and collect a wide range of information to provide rich original data for subsequent analysis and interaction based on image data, which helps to gain a deeper understanding of children's learning and living conditions.

[0043] Data preprocessing step: noise reduction is performed on the collected image data, using the mean filter algorithm, for each pixel x in the image ij (i represents the row index, j represents the column index), the processed pixel value Remove salt and pepper noise, Gaussian noise, etc. in the image; then perform image enhancement, using the adaptive histogram equalization (CLAHE) algorithm to divide the image into multiple sub-blocks, and for each pixel in the sub-block, redistribute its grayscale value to enhance the contrast and details of the image; finally, perform normalization to unify the pixel values ​​of the image data into the range of [0, 1]. For the pixel value, the normalized value Among them, min(X) and max(X) are the minimum and maximum values ​​of all pixel values ​​in the image, respectively, which is convenient for subsequent model processing.

[0044] First, the mean filter algorithm is used to remove noise interference to ensure image clarity; then the adaptive histogram equalization algorithm is used to enhance image quality and highlight key information based on local image characteristics; finally, normalization is performed to unify the data format. These operations can improve data availability and lay a good foundation for subsequent analysis and model training using image data.

[0045] Feature extraction step: Use convolutional neural network (CNN) to extract features from the preprocessed image. Suppose the input image is I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conv The value at a certain position in (m, n are the indices in the convolution kernel), and then through the pooling layer, such as the maximum pooling, for the pooling window, the output feature map F pool The value at a certain position in Finally, the feature vector F = {f1, f2, ..., f m}, where m is the number of features, f k is the kth feature. CNN learns the spatial features in the image through different convolutional layers and pooling layers, such as the facial expression features and body posture features of children.

[0046] Utilizing the powerful feature extraction capability of convolutional neural networks, local features of the image are extracted through convolution operations, and then the feature map resolution is reduced through the pooling layer while retaining key features. Complex image information is converted into feature vectors that are easy to analyze, providing strong support for subsequent data analysis and interactive strategy formulation.

[0047] Behavior analysis step: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s}(s is the number of behavior patterns, pl is the feature vector corresponding to the lth behavior pattern), and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library Determine the child's emotional state (such as happy, sad, focused, etc.), participation (positive, negative), and learning status (understanding, confusion), etc. By comparing the feature vector with the standard pattern in the behavior pattern library, the analysis results of the child's behavior are obtained.

[0048] Based on the extracted feature vectors, the cosine similarity algorithm is used to analyze the behavior pattern library. By calculating the similarity, the emotions, participation and learning status of the children can be judged, so that educators can understand the status of children in learning and activities, and provide a basis for adjusting teaching strategies and interaction methods to better meet the learning needs of children.

[0049] Interaction strategy generation steps: Based on the results of behavior analysis, the reinforcement learning algorithm is used to define the state S as the behavior state of the child (derived from behavior analysis), the action A as various interaction strategies (such as explaining knowledge, playing games, etc.), and the reward R as the effect feedback after the interaction (if the child's participation is improved, a positive reward will be given). By maximizing the cumulative reward (γ is a discount factor, ranging from [0, 1], and t is a time step) to generate interactive strategies. If the child is confused, the system generates a strategy to explain the knowledge point in detail; if the child's participation is low, a strategy for interesting interactive activities is generated. By continuously optimizing the strategy, the best educational interaction effect can be achieved.

[0050] According to the results of the children's behavior analysis, the reinforcement learning algorithm is used to develop interaction strategies by maximizing the cumulative rewards based on the definition of state, action and reward. According to different children's states, corresponding strategies are generated, such as detailed explanations, interesting interactions, etc. By continuously optimizing the strategy, the interaction is more in line with the needs of children, and the effect and quality of preschool education are improved.

[0051] Implementation steps of interactive feedback: The generated interactive strategy is fed back to the teacher or directly pushed to the intelligent teaching device. The teacher adjusts the teaching method according to the strategy, and the intelligent device plays the corresponding teaching content, conducts interactive games, etc. according to the strategy to achieve effective interaction with the children, and collects the children's feedback data to provide a basis for subsequent optimization.

[0052] The generated interactive strategies are applied to actual teaching. Teachers adjust teaching accordingly, and smart devices perform corresponding operations, such as playing teaching content, conducting interactive games, etc., to achieve interaction with children. At the same time, feedback from children is collected to provide data support for subsequent strategy optimization and system improvement, forming a closed loop of teaching interaction and continuously improving the interactive effect.

[0053] The present invention also includes the following steps:

[0054] Scene recognition step: Use the support vector machine (SVM) algorithm to analyze the image data and identify the current educational scene, such as classroom teaching, game activities, break time, etc. For the input image feature vector x, SVM solves the optimization problem The constraint is y i (w T φ(x i )+b)≥-ξ i ,ξ i ≥0, where w is the weight vector, b is the bias, ξ i is the slack variable, C is the penalty parameter, y i For sample x i Category labels (corresponding to different educational scenarios), φ(x i ) is to transform the sample x i A function that maps to a high-dimensional space. By calculating the distance between the sample and the classification hyperplane, the scene category to which the image belongs is determined, so that the interaction strategy can be generated more accurately.

[0055] The SVM algorithm is used to analyze image data, and by solving specific optimization problems, the distance between samples and the classification hyperplane is used to judge educational scenes, such as classroom, game, rest, etc. After clarifying the scene, more targeted interaction strategies can be generated to improve the accuracy and effectiveness of interaction and meet the needs of educational interaction in different scenarios.

[0056] Knowledge graph construction steps: Integrate preschool education related knowledge, such as children's cognitive development theory, teaching methods, knowledge points, etc., to construct a knowledge graph. The knowledge graph consists of nodes N = {n1, n2, ..., n q} (representing knowledge elements) and edge E = {e1, e2, ..., e r} (representing the relationship between knowledge elements), and the graph convolutional neural network (GCN) is used to update and optimize the knowledge graph. i , and its updated feature representation Where σ is the activation function, N i is the set of neighbor nodes of node i, d i and d j are the degrees of nodes i and j respectively, W l and b lis the weight matrix and bias vector of the lth layer. Through GCN, we learn the relationship between nodes in the knowledge graph, mine new knowledge, and update the knowledge graph so that the knowledge graph can better reflect the latest knowledge and dynamics in the field of preschool education.

[0057] Integrate various aspects of preschool education knowledge to build a knowledge graph, and use graph convolutional neural networks to update and optimize the knowledge graph. GCN can learn the relationship between nodes in the knowledge graph, mine new knowledge, and continuously update the content of the knowledge graph. This enables the knowledge graph to keep up with the development of the preschool education field, provide more timely and accurate knowledge support for behavior analysis and interactive strategy generation, and improve the adaptability and effectiveness of the system.

[0058] In the present invention, in the data preprocessing step, during image enhancement, a CLAHE algorithm is used to perform histogram equalization according to the characteristics of the local area of ​​the image, so as to better enhance the image details, improve the image quality, and make the subsequent feature extraction more accurate.

[0059] Summary: In the image enhancement link of data preprocessing, an adaptive histogram equalization algorithm is used. This algorithm processes based on the local characteristics of the image. Compared with traditional methods, it can better enhance image details, highlight key information in the image, and improve the overall image quality. It creates favorable conditions for the subsequent use of convolutional neural networks to accurately extract image features and improve the accuracy and effectiveness of feature extraction.

[0060] In the present invention, in the scene recognition step, the transfer learning technology is used to migrate the model pre-trained on a large-scale image dataset to the preschool education scene recognition task, thereby reducing the training time and data requirements and improving the efficiency and accuracy of scene recognition.

[0061] Summary: Transfer learning technology is used in scene recognition to migrate the model pre-trained on a large-scale image dataset to the preschool education scene recognition task. By using the common image features learned by the pre-trained model, the training time and data volume requirements in the preschool education scene are reduced, and the education scene is quickly and accurately identified, the efficiency and accuracy of scene recognition are improved, and the system operation cost is reduced.

[0062] In the present invention, in the knowledge graph construction step, a graph convolutional neural network (GCN) is used to update and optimize the knowledge graph. GCN learns the relationship between nodes in the knowledge graph, mines new knowledge, and updates the knowledge graph, so that the knowledge graph can better reflect the latest knowledge and dynamics in the field of preschool education.

[0063] In the construction of the knowledge graph, a graph convolutional neural network is used to update and optimize the knowledge graph. GCN can learn the relationship between nodes in the knowledge graph, mine new knowledge, and continuously update the content of the knowledge graph. This enables the knowledge graph to keep up with the development of the field of preschool education, provide more timely and accurate knowledge support for behavior analysis and interactive strategy generation, and improve the adaptability and effectiveness of the system.

[0064] The present invention also discloses a system of a preschool education interactive method based on image big data, comprising:

[0065] Image data acquisition module: Through cameras, smart tablets and other devices in the classroom, image data of children in classroom activities, game interactions, and life scenes are collected in all directions. n}, providing the system with raw image data support, covering various behaviors and status information of young children.

[0066] With the help of various devices, we comprehensively collect images of children in preschool education scenes. From daily classes to game activities, from daily life to learning moments, we collect image data extensively, providing rich original materials for subsequent data processing, analysis and interactive strategy formulation, which is the basis for the operation of the entire system.

[0067] Data preprocessing module: performs noise reduction, image enhancement and normalization on the collected image data. The mean filter algorithm is used to calculate the average value of each pixel x in the image. ij (i represents the row index, j represents the column index), the processed pixel value Remove noise; use the adaptive histogram equalization (CLAHE) algorithm to enhance the image; normalize the image pixel values ​​to the range [0, 1]. For pixel value x, the normalized value Among them, min(X) and max(X) are the minimum and maximum values ​​of all pixel values ​​in the image, respectively, which improves the quality of image data and facilitates subsequent model processing.

[0068] First, the mean filter is used to remove noise and make the image clear; then the adaptive histogram equalization is used to enhance the image quality and highlight the key information; finally, the normalization process is performed to unify the data format. After this series of operations, the usability of the image data is improved, laying a good foundation for subsequent feature extraction and analysis.

[0069] Feature extraction module: Use convolutional neural network (CNN) to extract features from the preprocessed image. Suppose the input image is I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conv The value at a certain position in (m, n are the indices in the convolution kernel), and then through the pooling layer, such as the maximum pooling, for the pooling window, the output feature map F pool The value at a certain position in Finally, the feature vector F = {f1, f2, ..., f m}, converting complex image information into feature vectors that are easy to analyze, providing a data basis for subsequent behavioral analysis.

[0070] Utilizing the powerful feature extraction capabilities of convolutional neural networks, local features of the image are extracted through convolution operations, and then the feature map resolution is reduced through a pooling layer while retaining key features, converting the image into a feature vector. These vectors are an important basis for subsequent analysis of children's behavior and status, providing key support for the system to understand the image content.

[0071] Behavior analysis module: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s}(s is the number of behavior patterns, pl is the feature vector corresponding to the lth behavior pattern), and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library Judge children's emotional state, participation, learning status, etc., to provide a basis for generating interactive strategies.

[0072] Based on the feature vector and behavior pattern library, the cosine similarity algorithm is used to interpret the children's behavior. By calculating the similarity, the emotions, participation and learning status of the children are judged, which helps educators understand the status of the children in learning and activities, and provides a reference for the subsequent formulation of interactive strategies, so that the interactive strategies are more in line with the actual needs of the children.

[0073] Interaction strategy generation module: Based on the results of behavior analysis, the reinforcement learning algorithm is used to define the state S as the behavior state of the child (derived from behavior analysis), the action A as various interaction strategies (such as explaining knowledge, playing games, etc.), and the reward R as the effect feedback after the interaction (if the child's participation is improved, a positive reward will be given). By maximizing the cumulative reward (γ is a discount factor, ranging from [0, 1], and t is a time step) to generate interactive strategies. According to the different behavioral states of young children, corresponding interactive strategies are generated to improve the interactive effect of preschool education.

[0074] According to the results of the children's behavior analysis, the reinforcement learning algorithm is used to develop interactive strategies by maximizing the cumulative rewards based on the definition of state, action and reward. According to the different states of children such as confusion and low participation, strategies such as detailed explanations and interesting interactions are generated, and the strategies are continuously optimized to improve the interactive effect, meet the learning and development needs of children, and improve the quality and effect of preschool education.

[0075] Interactive feedback implementation module: The generated interactive strategies are fed back to teachers or directly pushed to intelligent teaching devices. Teachers adjust teaching methods based on the strategies, and intelligent devices play corresponding teaching content, conduct interactive games, etc. according to the strategies to achieve effective interaction with children and collect feedback data from children.

[0076] The generated interactive strategies are applied to actual teaching. Teachers adjust teaching accordingly, and smart devices perform corresponding operations, such as playing teaching content, conducting interactive games, etc., to achieve interaction with children. At the same time, feedback from children is collected to provide data support for subsequent strategy optimization and system improvement, forming a closed loop of teaching interaction and continuously improving the interactive effect and education quality.

[0077] The present invention also includes the following modules:

[0078] Scene recognition module: Use the support vector machine (SVM) algorithm to analyze image data and identify current educational scenes, such as classroom teaching, game activities, break time, etc. For the input image feature vector, SVM solves the optimization problem.

[0079] Interaction strategy generation module: Genetic algorithm is used to optimize the interaction strategy. Define the strategy population P = {p1, p2, ..., p z}, where p i is an interactive strategy individual, and each strategy individual is encoded. By calculating the fitness function Fitness(p i ) evaluates the adaptability of each strategy individual to the current behavior state of the child. The higher the fitness, the better the strategy. The selection operation is based on the fitness, using the roulette selection method. The individual p i Probability of being selected The selected individuals are crossovered and mutated to generate new strategy individuals, which are continuously iterated and optimized to obtain better interaction strategies. Through genetic algorithms, we can search in a wider strategy space to find interaction strategies that better meet the needs of children and improve the interaction effect.

[0080] Genetic algorithms are introduced into the interactive strategy generation module to perform operations such as encoding, fitness evaluation, selection, crossover and mutation on the strategy population. Through continuous iterative optimization, better strategies are screened from a wider strategy space, and more targeted and efficient interactive strategies are provided for the complex and diverse behavioral states of young children, further improving the quality and effectiveness of preschool education interaction.

[0081] Privacy protection module: Federated learning technology is used to encrypt image data on the data collection device side. The homomorphic encryption algorithm is used. For the original image data x, the encrypted data y = Enc(x) to ensure the security of the data during transmission and storage. During model training, each device side is trained based on local encrypted data, and the model parameters are uploaded instead of the original data. The global model is updated through the federated aggregation algorithm. For example, in the federated average algorithm, the server collects the model parameters θ of each device side. i , calculate the global model parameters Where N is the number of participating devices. The privacy protection module not only ensures the privacy security of children's image data, but also enables effective learning and system optimization using multi-device data.

[0082] The system adds a privacy protection module, using federated learning technology and homomorphic encryption algorithms. Image data is encrypted at the data collection end, and encrypted data is transmitted and stored to ensure security; when training the model, each device is trained based on local encrypted data, and the global model is updated through a federated aggregation algorithm. This module protects the privacy of children while making full use of multi-device data resources to achieve safe and efficient operation of the system.

[0083] The above are only preferred specific implementation modes of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical solutions and inventive concepts of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A preschool education interactive method based on image big data, characterized in that: include: Image data collection steps: Use the cameras in the classroom and smart tablet devices to collect image data of children in classroom activities, game interactions, and life scenes I = {i1, i2, ..., i n }, where n is the number of images, i j Represents the jth image Data preprocessing step: noise reduction is performed on the collected image data, using the mean filter algorithm, for each pixel x in the image ij , where i represents the row index, j represents the column index, and the processed pixel value Remove noise from the image; then perform image enhancement, use the adaptive histogram equalization CLAHE algorithm to divide the image, and for the pixels in each sub-block, redistribute the grayscale value to enhance the contrast and details of the image; Finally, normalization is performed to unify the pixel values ​​of the image data into the range of [0, 1]. For the pixel value, the normalized value Where min(X) and max(X) are the minimum and maximum values ​​of all pixels in the image, respectively. Feature extraction step: Use convolutional neural network CNN to extract features from the preprocessed image. Suppose the input image is I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conυ The value at a certain position in Among them, m and n are the indices in the convolution kernel. After the pooling layer, for the pooling window, the output feature map F pool The value at a certain position in Finally, the feature vector F = {f1, f2, ..., f m }, where m is the number of features, f k is the kth feature; Behavior analysis step: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s }, where s is the number of behavior patterns, p1 is the feature vector corresponding to the first behavior pattern, and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library. Determine the emotional state and learning status of young children; Interaction strategy generation steps: Based on the behavior analysis results, the reinforcement learning algorithm is used to define the state S as the behavior state of the child, the action A as various interaction strategies, and the reward R as the effect feedback after the interaction. Generate an interactive strategy, where γ is a discount factor, ranging from [0, 1], and t is the time step; Interactive feedback implementation steps: The generated interactive strategy is fed back to the teacher or pushed to the teaching equipment. The teacher adjusts the teaching method according to the strategy, and the equipment plays the corresponding teaching content and conducts interactive games according to the strategy.

2. The interactive method for preschool education based on image big data according to claim 1, characterized in that: Also includes: Scene recognition step: Analyze the image data through the support vector machine (SVM) algorithm to identify the current educational scene, including classroom teaching, game activities, and rest time; For the input image feature vector x, SVM solves the optimization problem The constraint condition is yi(w T φ(xi)+b)≥1-ξi,ξi≥0, where w is the weight vector, b is the bias, ξ i is the slack variable, C is the penalty parameter, y i For sample x i The category label of φ(x i ) is the sample x i Functions that map to higher-dimensional spaces.

3. The interactive method for preschool education based on image big data according to claim 1, characterized in that: Also includes: Steps for constructing the knowledge map: Integrate knowledge related to preschool education, including theories of children's cognitive development, teaching methods, and knowledge points, and construct a knowledge map; The knowledge graph consists of nodes N = {n1, n2, ..., n q } represents knowledge elements and edges E = {e1, e2, ..., e r } represents the relationship between knowledge elements. For node n i , and its updated feature representation Where σ is the activation function, N i is the set of neighbor nodes of node i, d i and d j are the degrees of nodes i and j respectively, W l and b l is the weight matrix and bias vector of the lth layer.

4. The interactive method for preschool education based on image big data according to claim 1, characterized in that: In the data preprocessing step, when enhancing the image, the adaptive histogram equalization CLAHE algorithm is used to perform histogram equalization according to the characteristics of the local area of ​​the image.

5. The interactive method for preschool education based on image big data according to claim 2, characterized in that: In the scene recognition step, transfer learning technology is used to migrate the model pre-trained on the image dataset to the preschool education scene recognition task, reducing training time and data requirements.

6. The interactive method for preschool education based on image big data according to claim 3 is characterized in that: In the knowledge graph construction step, the graph convolutional neural network GCN is used to update and optimize the knowledge graph.

7. A system for implementing the preschool education interactive method based on image big data as described in any one of claims 1 to 6, characterized in that: include: Image data acquisition module: collects image data of children in classroom activities, game interactions, and life scenes through cameras and smart tablet devices in the classroom. n }; Data preprocessing module: denoise, enhance and normalize the collected image data; use the mean filter algorithm to calculate the average value of each pixel x in the image. ij , where i represents the row index, j represents the column index, and the processed pixel value Remove noise; use the adaptive histogram equalization CLAHE algorithm to enhance the image by normalizing the image pixel values ​​to the range of [0, 1]. For the pixel value x, the normalized value Where min(X) and max(X) are the minimum and maximum values ​​of all pixel values ​​in the image respectively; Feature extraction module: Use convolutional neural network CNN to extract features from the preprocessed image. Let the input image be I. After the convolution operation of the convolution layer, for the convolution kernel K, the output feature map F conυ The value at a certain position in Among them, m and n are the indices in the convolution kernel, and then after the pooling layer, for the pooling window, the output feature map F pool The value at a certain position in Generate feature vector F = {f1, f2, ..., f m }; Behavior analysis module: Based on the extracted feature vector F, combined with the preset behavior pattern library P = {p1, p2, ..., p s }, where s is the number of behavior patterns, p1 is the feature vector corresponding to the first behavior pattern, and the cosine similarity algorithm is used to calculate the similarity between the feature vector and each pattern in the behavior pattern library. Determine the child's emotional state, engagement, and learning status; Interaction strategy generation module: Based on the behavior analysis results, the reinforcement learning algorithm is used to define the state S as the behavior state of the child, the action A as various interaction strategies, and the reward R as the effect feedback after the interaction. Generate an interactive strategy, where γ is a discount factor, ranging from [0, 1], and t is the time step; Interactive feedback implementation module: The generated interactive strategies are fed back to teachers or pushed to teaching equipment. Teachers adjust teaching methods based on the strategies. The equipment plays corresponding teaching content and conducts interactive games according to the strategies to achieve interaction with children and data collection.

8. The preschool education interactive system based on image big data according to claim 7, characterized in that: Also includes: Scene recognition module: Analyze image data through the support vector machine (SVM) algorithm to identify the current educational scene, including classroom teaching, game activities, and rest time; For the input image feature vector, SVM solves the optimization problem. The constraint condition is yi(w T φ(xi)+b)≥1-ξi,ξi≥0, where w is the weight vector, b is the bias, ξ i is the slack variable, C is the penalty parameter, y i For sample x i The category label of φ(x i ) is the sample x i Functions that map to higher-dimensional spaces.

9. The preschool education interactive system based on image big data according to claim 7, characterized in that: Also includes: Interaction strategy generation module: using genetic algorithm to optimize interaction strategy; Define the strategy population P = {p1, p2, ..., p z }, where p i is an interactive strategy individual, and each strategy individual is encoded; by calculating the fitness function Fitness(p i ) evaluates the adaptability of each strategy individual to the current behavior state of the child, and performs selection operations based on fitness, using the roulette wheel selection method. i Probability of being selected 10. The preschool education interactive system based on image big data according to claim 7, characterized in that: Also includes: Privacy protection module: Federated learning technology is used to encrypt image data on the data collection device side. The homomorphic encryption algorithm is used. For the original image data x, the encrypted data y = Enc(x). During model training, each device side is trained based on the local encrypted data, uploads the model parameters, and updates the global model through the federated aggregation algorithm; the server collects the model parameters θ of each device side i , calculate the global model parameters Where N is the number of participating devices.

Images