A method and system for identifying classroom listening characteristics

Abstract

This invention relates to a method and system for identifying classroom listening characteristics, belonging to the field of classroom student state recognition technology. By analyzing head support time and arm pressure distribution data collected by desk pressure sensors using a decision tree algorithm, it can automatically and in real-time identify students sleeping on their desks. This overcomes the limitations of low efficiency in traditional manual supervision, significantly reducing the burden of classroom management for teachers. Compared to image recognition technology, it relies solely on pressure data for detection, resulting in lower costs and fundamentally avoiding privacy risks, making it more acceptable to students and parents. Furthermore, the decision tree algorithm has a clear structure, low hardware requirements, and is easy to deploy in ordinary classrooms. The pressure sensor detection method is unaffected by environmental factors such as light and obstruction, making it suitable for various classroom scenarios. It also has the potential to expand by combining data from more sensors to identify other classroom states, effectively filling the gap in existing technologies for identifying students sleeping on their desks and enriching the intelligent monitoring technology system in the education field.

Description

Technical Field

[0001] This invention relates to the field of classroom student status recognition technology, specifically a classroom listening feature recognition method based on decision tree algorithm, which identifies the characteristic of sleeping while lying on the desk by analyzing desk pressure sensor data and triggers a micro-vibration of the desk to wake the student up. Background Technology

[0002] Students sleeping at their desks is a common phenomenon in classroom teaching, which seriously affects learning outcomes. Traditional manual supervision methods are inefficient and cannot accurately identify the status of each student in real time. With the development of sensor technology and machine learning algorithms, it has become possible to use sensors to collect data and combine it with algorithms to identify student status. Decision tree algorithms, as a commonly used classification algorithm, are characterized by clear structure, ease of understanding and implementation, and are suitable for binary classification of student classroom status.

[0003] Most existing technologies focus on detecting infants' or patients' prone sleeping postures. For example, patent CN115641603A provides a method, device, and equipment for monitoring infants' prone sleeping postures based on multimodal data fusion. This method includes: acquiring real-time visible light and real-time thermal infrared images of the infant's care area; calibrating the real-time visible light and thermal infrared images to obtain a homography matrix; extracting a first infant region image based on the infant's body temperature characteristics; outputting a second infant region image from the real-time visible light image based on the homography matrix; detecting the infant's head in the second infant region image; and when the infant's head is detected, performing prone sleeping detection and outputting the infant's prone sleeping detection result. This patent requires image equipment and processing chips, which is costly and difficult to use widely in schools. Currently, there is no relevant research on detecting student status in the classroom. This invention addresses this problem without relying on image technology, providing a method for recognizing classroom listening characteristics. Summary of the Invention

[0004] This invention aims to provide an intelligent recognition solution capable of sensing students' classroom status in real time. By analyzing the time students spend supporting their heads and the distribution of arm pressure collected by sensors on their desks, the system automatically determines whether a student is napping at their desk. When a student is detected napping at their desk, the system will silently remind them with a slight vibration of the desk, helping the student maintain a good listening attitude and thus improving their concentration and learning efficiency in the classroom.

[0005] The purpose of this invention is to provide a method for recognizing classroom listening characteristics. By analyzing desk pressure sensor data through a decision tree algorithm, the method can accurately identify the characteristic of sleeping while slumped over a desk and trigger a micro-vibration of the desk to wake up the student, thereby improving classroom learning efficiency.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Step 1, Data Acquisition: Pressure data of the student's head contact area and pressure distribution data of the arm contact area are collected in real time by pressure sensors installed on the desk.

[0008] Furthermore, the head contact area is located in the upper left corner of the desk, and the arm contact area is located in the armrest areas on both sides of the desk.

[0009] Step 2, Data Preprocessing and Feature Extraction:

[0010] Step 2.1: Preprocess the collected data;

[0011] Preprocessing includes filtering, noise reduction, and cleaning to remove noise and outliers, making the data comparable.

[0012] Step 2.2: Extract two types of features from the preprocessed data: head support duration and arm pressure distribution;

[0013] (1) Let P be the actual pressure collected in the head contact area. head Set the head contact area pressure trigger threshold P th (e.g., 5N represents the minimum pressure required for the head to lightly touch the table). When P head >P th When P is determined to be in "head support state"; head ≤P th When the pressure rises from "unsupported state" to "head support state", the timer is started; when the pressure falls back to "unsupported state", the timer is stopped, and the duration of head support is recorded.

[0014] (2) Calculate the symmetry index of the pressure distribution in the arm contact area, and record it as the arm pressure distribution data;

[0015] (3) Let the head support duration be t. i (i = 1, 2, ..., n), arm pressure distribution data is p i (i = 1, 2, ..., n); and normalize the two types of data to obtain the normalized feature t. i 'and p i '. The details are as follows:'

[0016] The head support duration data is normalized using the following formula:

[0017]

[0018] Where t min t represents the minimum value of the head support duration data. max This represents the maximum value of the head support duration data.

[0019] The arm pressure distribution data is normalized using the following formula:

[0020]

[0021] Where p min p represents the minimum value of the arm pressure distribution data. max This represents the maximum value of the arm pressure distribution data.

[0022] Step 3, Decision Tree Model Construction: Utilizing the two types of features t' extracted in Step 2 i and p' i As input X, it is used to train the decision tree model, which performs binary classification and outputs the student's state as either normal or napping.

[0023] The process of constructing the decision tree model includes:

[0024] (1) Calculate the Gini index for each feature: Select the optimal feature and its optimal split point. Let A1(X) and A2(X) represent the head support duration t', respectively. i and arm pressure distribution p' i Two features are used. After sorting each feature, the median value of adjacent values ​​is taken as the candidate split point 'a'. i Calculate the Gini index for each candidate point;

[0025] For dataset D, the Gini index is calculated as follows:

[0026]

[0027] Where C k Represents dataset D i The sample set of class k, where k=1 represents the napping state and k=2 represents the normal state, |C k | is the number of samples in the k-th class, |D i |For dataset D i The total number of samples; D is a set containing the entire data, D1 = {X∈D|A i (X)≤a i} corresponds to the data set where the selected feature is less than or equal to the split point, D2={X∈D|A i (X)>a i} corresponds to the data set where the selected feature is greater than the dividing point.

[0028] Choose feature A that minimizes the Gini index. * and the dividing point a * Perform node splitting, where (A) * ,a * ) = argminA,a Gini(D,A,a).

[0029] (2) Root node partitioning: Apply the optimal partitioning (A) to the entire training set at the root node. * ,a * ), generate left and right subtrees.

[0030] (3) Subtree recursion: Repeat the above steps for each child node until one of the following termination conditions is met: the number of node samples is less than the threshold N. min The node's Gini index is less than the threshold Gini. min All samples belong to the same category.

[0031] (4) Leaf node labeling: Label the leaf node with the label of the majority of categories in that node.

[0032] (5) Model output and state judgment: Input the real-time collected feature vectors into the decision tree starting from the root node, and traverse downwards according to the comparison results of feature values ​​and split points until the leaf node is reached, and output the category label.

[0033] Step 4, Feature Recognition: Input the real-time collected and preprocessed data into the constructed decision tree model to determine whether the student's current state is normal or napping.

[0034] Step 5, Wake-up Reminder: When the student is detected to be napping (preferably 3 times) consecutively, the desk micro-vibration module is triggered to remind the student to pay attention through micro-vibration.

[0035] This patent represents a classroom listening feature recognition system, the structure of which includes:

[0036] (1) Sensor module, a pressure sensor matrix deployed in the head contact area and arm contact area of ​​the desk;

[0037] (2) Preprocessing and feature extraction module: used to extract two types of features after preprocessing: head support duration and arm pressure distribution;

[0038] (3) Decision tree classification module: The decision tree classification module is trained by using the extracted features and dividing the features based on the Gini index;

[0039] (4) Micro-vibration execution module: triggers vibration reminder in response to continuous nap state detection results.

[0040] The beneficial effects of this invention are as follows: This invention analyzes head support time and arm pressure distribution data collected by desk pressure sensors using a decision tree algorithm, enabling real-time and automatic identification of students sleeping on their desks. This method overcomes the limitations of low efficiency in traditional manual supervision, significantly reducing the burden of classroom management for teachers. Compared to image recognition technology, it relies solely on pressure data for detection, resulting in low cost and fundamentally avoiding privacy risks, making it more acceptable to students and parents. Simultaneously, the system uses desk micro-vibration for wake-up, a silent reminder that does not disturb other students, protects students' self-esteem and privacy, helps them quickly regain focus, reduces classroom drowsiness, and, with long-term use, can guide students to develop good classroom habits, indirectly improving overall learning concentration and classroom efficiency. Furthermore, the decision tree algorithm has a clear structure, low hardware requirements, and is easy to deploy in ordinary classrooms. The pressure sensor detection method is unaffected by light, obstructions, or other environmental factors, making it suitable for various classroom scenarios. It also has the potential to expand by combining data from more sensors to identify other classroom states, effectively filling the gap in existing technologies for identifying students sleeping on their desks and enriching the intelligent monitoring technology system in the education field. Actual testing showed that this method can achieve an accuracy rate of over 95% in recognizing students taking naps. Attached Figure Description

[0041] Figure 1 Flowchart of the classroom listening feature recognition method of the present invention Detailed Implementation

[0042] Step 1, Data Acquisition: Pressure data of the student's head contact area and pressure distribution data of the arm contact area are collected in real time by pressure sensors installed on the desk.

[0043] Furthermore, the head contact area is located in the upper left corner of the desk, and the arm contact area is located in the armrest areas on both sides of the desk.

[0044] Step 2, Data Preprocessing and Feature Extraction:

[0045] Step 2.1: Preprocess the collected data;

[0046] Preprocessing includes filtering, noise reduction, and cleaning to remove noise and outliers, making the data comparable.

[0047] Step 2.2: Extract two types of features from the preprocessed data: head support duration and arm pressure distribution;

[0048] (1) Let P be the actual pressure collected in the head contact area. head Set the head contact area pressure trigger threshold P th (e.g., 5N represents the minimum pressure required for the head to lightly touch the table). When P head >P thWhen P is determined to be in "head support state"; head ≤P th When the pressure rises from "unsupported state" to "head support state", the timer is started; when the pressure falls back to "unsupported state", the timer is stopped, and the duration of head support is recorded.

[0049] (2) Calculate the symmetry index of the pressure distribution in the arm contact area, and record it as the arm pressure distribution data;

[0050] (3) Let the head support duration be t. i (i = 1, 2, ..., n), arm pressure distribution data is p i (i = 1, 2, ..., n); and normalize the two types of data to obtain the normalized feature t'. i and p' i The details are as follows:

[0051] The head support duration data is normalized using the following formula:

[0052]

[0053] Where t min t represents the minimum value of the head support duration data. max This represents the maximum value of the head support duration data.

[0054] The arm pressure distribution data is normalized using the following formula:

[0055]

[0056] Where p min p represents the minimum value of the arm pressure distribution data. max This represents the maximum value of the arm pressure distribution data.

[0057] Step 3, Decision Tree Model Construction: Utilizing the two types of features t' extracted in Step 2 i and p' i As input X, it is used to train the decision tree model, which performs binary classification and outputs the student's state as either normal or napping.

[0058] The process of constructing the decision tree model includes:

[0059] (1) Calculate the Gini index for each feature: Select the optimal feature and its optimal split point. Let A1(X) and A2(X) represent the head support duration t', respectively. i and arm pressure distribution p' i Two features are used. After sorting each feature, the median value of adjacent values ​​is taken as the candidate split point 'a'.i Calculate the Gini index for each candidate point;

[0060] For dataset D, the Gini index is defined as:

[0061]

[0062] Where C k Represents dataset D i The sample set of class k, where k=1 represents the napping state and k=2 represents the normal state, |C k | is the number of samples in the k-th class, |D i |For dataset D i The total number of samples; D is a set containing the entire data, D1 = {X∈D|A i (X)≤a i} corresponds to the data set where the selected feature is less than or equal to the split point, D2={X∈D|A i (X)>a i} corresponds to the data set where the selected feature is greater than the dividing point.

[0063] Choose feature A that minimizes the Gini index. * and the dividing point a * Perform node splitting, where (A) * ,a * ) = argmin A,a Gini(D,A,a).

[0064] (2) Root node partitioning: Apply the optimal partitioning (A) to the entire training set at the root node. * ,a * ), generate left and right subtrees.

[0065] (3) Subtree recursion: Repeat the above steps for each child node until one of the following termination conditions is met: the number of node samples is less than the threshold N. min The node's Gini index is less than the threshold Gini. min All samples belong to the same category.

[0066] (4) Leaf node labeling: Label the leaf node with the label of the majority of categories in that node.

[0067] (5) Model output and state judgment: Input the real-time collected feature vectors into the decision tree starting from the root node, and traverse downwards according to the comparison results of feature values ​​and split points until the leaf node is reached, and output the category label.

[0068] Step 4, Feature Recognition: Input the real-time collected and preprocessed data into the constructed decision tree model to determine whether the student's current state is normal or napping.

[0069] Step 5, Wake-up Reminder: When the student is detected to be napping (preferably 3 times) consecutively, the desk micro-vibration module is triggered to remind the student to pay attention through micro-vibration.

Claims

1. A method for identifying classroom listening characteristics, characterized in that, The steps are as follows: Step 1, Data Acquisition: Pressure data of the student's head contact area and pressure distribution data of the arm contact area are collected in real time by pressure sensors installed on the desk; Step 2: Data preprocessing and feature extraction; Step 2.1: Preprocess the collected data; Step 2.2: Extract two types of features from the preprocessed data: head support duration and arm pressure distribution; Step 2.2 specifically involves: (1) Let the actual pressure collected in the head contact area be... Set the pressure trigger threshold in the head contact area. ;when When it is determined to be in a "head-supported state"; when When the pressure rises from "non-supported state" to "head support state", the timer is started; when the pressure falls back to "non-supported state", the timer is stopped and the duration of head support is recorded. (2) Calculate the symmetry index of the pressure distribution in the arm contact area and record it as the arm pressure distribution data; (3) Let the head support duration data be... Arm pressure distribution data is The two types of data were then normalized to obtain normalized features. and The details are as follows: The head support duration data is normalized using the following formula: in This represents the minimum value of the head support duration data. This represents the maximum value of the head support duration data; The arm pressure distribution data is normalized using the following formula: in This represents the minimum value of the arm pressure distribution data. This represents the maximum value of the arm pressure distribution data; Step 3: Decision tree model construction: Utilizing the two types of features extracted in Step 2 and As The input is used to train the decision tree model, which performs binary classification and outputs the student's state as either normal or napping. Step 4, Feature Recognition: Input the real-time collected and preprocessed data into the constructed decision tree model to determine whether the student's current state is normal or napping. Step 5, Wake-up Reminder: When a student is continuously detected to be napping, the desk micro-vibration module is triggered to remind the student to pay attention through micro-vibration.

2. The classroom listening feature recognition method according to claim 1, characterized in that, In step 1, the head contact area is located in the upper left corner of the desk, and the arm contact area is located in the armrest area on both sides of the desk.

3. The classroom listening feature recognition method according to claim 1, characterized in that, In step 2.1, the preprocessing includes filtering, noise reduction, and cleaning to remove noise and outliers, making the data comparable.

4. The classroom listening feature recognition method according to claim 1, characterized in that, In step 3, the construction process of the decision tree model includes: (1) Calculate the Gini index of each feature: Select the optimal feature and its optimal split point; respectively using , Indicates head support duration and arm pressure distribution Two features are used. After sorting each feature, the median value of adjacent values ​​is taken as the candidate split point. Calculate the Gini index for each candidate point; For dataset The Gini index is calculated as follows: in Represents the dataset The Middle A sample set of classes In a nap state, This is the normal state. For the first The number of class samples, For dataset The total number of samples; It is a collection that contains the entire data. This corresponds to the data set where the selected feature is less than or equal to the split point. This corresponds to the data set where the selected feature is greater than the dividing point; Choose the feature that minimizes the Gini index. and dividing point Perform node splitting, where, ; (2) Root node partitioning: Apply the optimal partitioning to the entire training set at the root node. Generate left and right subtrees; (3) Subtree recursion: Repeat the above steps for each child node until one of the following termination conditions is met: the number of node samples is less than the threshold. The node Gini index is less than the threshold. All samples belong to the same category; (4) Leaf node labeling: Label the leaf node with the label of the majority of categories in that node; (5) Model output and state judgment: Input the real-time collected feature vectors into the decision tree starting from the root node, and traverse downwards according to the comparison results of feature values ​​and split points until the leaf node is reached, and output the category label.

5. A classroom listening feature recognition system, characterized in that, For implementing the method according to any one of claims 1-4, the classroom attendance feature recognition system includes: (1) Sensor module, a matrix of pressure sensors deployed in the head contact area and arm contact area of ​​the desk; (2) Preprocessing and feature extraction module: used to extract two types of features after preprocessing: head support duration and arm pressure distribution; (3) Decision tree classification module: The decision tree classification module is trained by using the extracted features and dividing the features based on the Gini index; (4) Micro-vibration execution module: triggers vibration reminder in response to continuous nap state detection results.

Images