An electronic screen classroom teaching system based on wireless network connection

Abstract

The application discloses a kind of electronic screen classroom teaching systems based on wireless network connection, comprising: sensing interaction layer: real-time capture classroom behavior data, collect individual learning state data, after local pre-processing, by wireless transmission layer encryption is sent to edge intelligent layer;Edge intelligent layer: running lightweight AI model for real-time analysis: through voice emotion features and micro-expression recognition to judge student participation, generate personalized feedback, optimize group collaboration layout;Local strategy is returned to sensing interaction layer execution;Wireless transmission layer: through dynamic spectrum allocation to ensure the transmission of interactive data packet, use physical layer security encryption to protect data privacy;Cloud service layer: aggregate multi-classroom interactive data lake, based on historical interaction mode to optimize classroom rhythm, and through digital twin rendering cluster to generate virtual classroom mirror image;Optimized strategy is fed back to edge intelligent layer by wireless transmission layer and localized adaptation, to provide cross-campus joint teaching support for teachers.

Description

Technical Field

[0001] This invention relates to the field of electronic screen classroom teaching technology, and in particular to an electronic screen classroom teaching system based on wireless network connection. Background Technology

[0002] With the development of technology, electronic screen classroom teaching has been widely used in our work and study. This is a comprehensive response to efficiency, fairness and health in the process of education modernization. It is both a technology implementation guided by policy and a common demand of schools, families and society for high-quality educational resources.

[0003] At the same time, the problem of poor interactivity in electronic screen classroom teaching has gradually become prominent. Traditional electronic screen classroom teaching systems are characterized by one-way instruction, limited interactive forms, inefficient group collaboration, lack of personalized feedback, and barriers to cross-campus interaction, which seriously affect the efficiency of normal teaching and learning and reduce students' learning initiative. Therefore, an electronic screen classroom teaching system based on wireless network connection is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an electronic screen classroom teaching system based on wireless network connectivity.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An electronic screen classroom teaching system based on wireless network connection includes: Perception and Interaction Layer: Through the teacher's terminal's UWB positioning, gesture recognition, and voice emotion analysis modules, classroom behavior data (such as teacher movement trajectory, gesture commands, and voice characteristics) is captured in real time. At the same time, individual learning status is collected through the pressure-sensing handwriting and micro-expression cameras on the student's terminal. After local preprocessing, this multimodal data is encrypted and sent to the edge intelligence layer through the wireless transmission layer, providing raw materials for enhancing interactivity. For example, the teacher's gestures can instantly trigger courseware animations, and changes in the pressure of the student's handwriting can reflect the degree of confusion. Edge Intelligence Layer: After receiving data from the Perception and Interaction Layer, it runs a lightweight AI model for real-time analysis: it judges student participation by recognizing voice emotion features and micro-expressions, generates personalized feedback using handwriting dynamics parameters, and optimizes group collaboration layout with the help of spatial positioning data; the analysis results are synchronized to the cloud service layer via the wireless transmission layer to train the global model, and at the same time, local strategies (such as adjusting the frequency of questions and pushing supplementary materials) are sent back to the Perception and Interaction Layer for execution. Wireless transmission layer: Adopting a customized Wi-Fi 6E protocol, it prioritizes the transmission of interactive data packets (such as gesture commands and emotional features) through dynamic spectrum allocation, and uses physical layer security encryption to ensure data privacy; This layer serves as a bridge between the perception and interaction layer and the edge / cloud layer, efficiently uploading the raw data collected by the terminal to the edge node for processing, and distributing real-time strategies generated at the edge (such as student group information) and long-term optimization suggestions from the cloud (such as adjustments to the teaching pace) to the terminal; Cloud service layer: Aggregates multi-classroom interactive data lake, runs reinforcement learning models through teaching strategy engine, optimizes classroom pace based on historical interaction patterns (such as automatically adjusting the explanation / practice ratio), and generates virtual classroom images through digital twin rendering cluster; the optimized strategy is fed back to edge intelligence layer through wireless transmission layer for localization adaptation, and at the same time provides teachers with cross-campus joint teaching support (such as sharing AR experiment guidance).

[0006] The above technical solution further includes: Furthermore, the real-time capture of classroom behavior data through the teacher's terminal's UWB positioning, gesture recognition, and voice emotion analysis modules includes the following steps: UWB (Ultra-Wideband) positioning: The teacher's terminal is equipped with a lidar module that scans the classroom space at a frequency of 30Hz to construct a three-dimensional point cloud map. The UWB tag worn by the teacher periodically emits pulse signals (pulse width <2ns), and the student terminal synchronously emits response signals. By measuring the time difference of flight of the signals between the teacher's terminal and the student terminal, and combining it with a multi-point positioning algorithm, sub-meter-level spatial positioning is achieved. Based on the Kalman filter algorithm, the positioning data is smoothed to generate the teacher's movement trajectory (sampling interval 50ms). Gesture recognition algorithm library: The teacher terminal's built-in RGB-D camera captures hand movements at 60FPS. The MediaPipe framework is used to extract the 3D coordinates of 21 hand joints. The spatial coordinate sequence of the joints is matched with 26 pre-trained teaching scene gesture templates (such as "page turning", "zooming", and "annotating") using dynamic time warping (DTW). When the matching degree exceeds the threshold (>0.85), the corresponding courseware control command is triggered and sent to the main screen through the wireless transmission layer. Voice sentiment analysis: A circular microphone array uses beamforming technology to lock the direction of the teacher's voice source (azimuth accuracy ±3°), performs short-time Fourier transform on the 16kHz sampled audio, extracts 12-dimensional MFCC features and fundamental frequency and energy parameters; inputs them into a lightweight Transformer model (parameters compressed to 3.7M), outputs probability values ​​for excitement and confusion emotion dimensions, binds emotion tags to speech segment timestamps, and generates a classroom speech stream with emotion annotations; Data fusion: The timestamps of the UWB positioning, gesture recognition, and voice emotion analysis modules are synchronized via the NTP protocol (synchronization accuracy <10ms). The isolated forest algorithm is used to detect positioning drift points (velocity change >5m / s) and emotion change points (probability jump >0.6). The effective data is encapsulated in JSON format, which includes four parts: device ID, data type, timestamp, and payload. The data packets are encrypted using the AES-256-GCM algorithm and sent to the edge intelligence layer through the wireless transmission layer.

[0007] Furthermore, the process of collecting individual learning status data through pressure-sensitive handwriting and micro-expression cameras on student terminals includes the following steps: Pressure-sensitive handwriting capture: When students use the smart pen to write on the foldable e-paper notebook, the pen tip pressure sensor (512-level resolution) senses the writing pressure in real time, and the six-axis sensor (three-axis accelerometer + three-axis gyroscope) captures the pen tip movement trajectory at a sampling rate of 200Hz, generating a three-dimensional spatial coordinate sequence. The trajectory is optimized by a handwriting smoothing algorithm (such as the Ramer-Douglas-Peucker algorithm) and converted into digital ink data in SVG format. Metadata tags including average pressure value (unit: grams), writing speed (pixels / second), and stroke direction angle (0°-360°) are added to each stroke. Micro-expression camera capture: The student terminal's front-facing camera captures facial images at 30 FPS with a resolution of 640×480. A lightweight face detection model (such as UltraFace) is used to perform real-time detection on the CPU (detection latency <30ms). The coordinates of 68 facial key points are extracted using the MobileFaceNet model, with a focus on annotating the eyes (orbicularis oculi muscle region) and the mouth (orbicularis oris muscle region). Micro-expression coding: Eye area: Calculate eyelid closure degree (EAR value) and blink frequency; Mouth area: Calculate mouth opening (MAR value) and corner of the mouth upward angle; Overall face: Calculate head attitude angles (pitch angle, yaw angle, roll angle); Multimodal data fusion: By synchronizing the timestamps of handwriting data and micro-expression data using the PTP protocol (synchronization accuracy <100ms), a time series correlation is established between handwriting data (such as the problem-solving process) and micro-expression changes (such as confusion / sudden realization); based on the frequency of handwriting pauses (>2 seconds) and the confusion level of micro-expressions (decreasing EAR value), a correlation analysis is conducted using handwriting accuracy (matching degree with standard answer) and smiling frequency (increasing MAR value); Data preprocessing and transmission: An SVG path simplification algorithm is used to reduce the amount of data (compression ratio ≥ 4:1), extract keyframes (1 frame per second), and retain peak points of facial expression changes; the facial images are blurred (Gaussian blur σ=5) to retain only the coordinates of key points; and the handwriting data is desensitized to remove sensitive information (such as name and student ID). The data is encrypted using the AES-128-CBC algorithm and then transmitted to the edge intelligence layer via the wireless transport layer.

[0008] Furthermore, the method of determining student participation through voice emotion features and micro-expression recognition includes the following steps: Multimodal data acquisition: The microphone array of the student terminal captures speech segments at a sampling rate of 48kHz, uses beamforming technology to locate the speaking student, and performs echo cancellation and noise suppression (SNR improvement ≥15dB); endpoint detection (VAD) is performed on the speech, 13-dimensional MFCC features and fundamental frequency (F0) and energy (RMS) parameters are extracted, and a 200ms speech feature frame is constructed; the RGB camera of the student terminal captures facial images at a frame rate of 30FPS, detects 68 facial key points using the MTCNN algorithm, and performs affine transformation for face alignment; ROI images of the eyes (orbicularis oculi muscle region) and mouth (orbicularis oris muscle region) are cropped, scaled to 64×64 pixels, converted to grayscale images, and pixel values ​​are normalized; Feature extraction for emotion classification: The input is fed into a lightweight Transformer model (MobileBERT architecture, with parameters compressed to 12M), which captures temporal dependencies in speech features through a self-attention mechanism; it outputs probability values ​​for the emotion dimensions of arousal and confusion, with dimensional quantization using a 5-level Likert scale; an improved FER2013 model (with added attention mechanism) is used to extract features from eye and mouth ROI images, identify 6 basic emotions (happiness, surprise, confusion, neutrality, annoyance, and frustration), and output the confidence scores for each category; Participation fusion calculation: Synchronize timestamps of speech emotion features and micro-expression features via NTP protocol (synchronization accuracy <50ms); weighted fusion: The emotional weight of the voice is ω1 = 0.6 (excitement level × 0.7 + confusion level × 0.3). Micro-expression weight ω2=0.4 (mean confidence score for positive emotions); Engagement score S = ω1×(vocal emotion score) + ω2×(micro-expression score); A Gaussian kernel function (σ=1.5m) was used to spatially smooth the participation score, generating a classroom participation heatmap (resolution 0.5m / pixel). Real-time feedback optimization: When the mean regional participation rate is <0.3 (5-point scale), a low participation warning is triggered. For low participation areas, the edge intelligence layer sends instructions to the teacher's terminal via the wireless transmission layer, suggesting adjustments to the teaching method (such as adding questions / interactions), pushing personalized learning resources (based on students' historical learning data), and simultaneously generating group collaboration tasks (such as "Please have students in this area discuss case X"). These tasks are then sent to the student's terminal via UWB signal for vibration alerts. Based on real-time feedback from the teacher (such as gesture confirmation / rejection suggestions), the strategy generation logic is updated through a reinforcement learning model (Q-learning).

[0009] Furthermore, the process of generating personalized feedback using handwriting dynamics parameters and optimizing the team collaboration layout with the help of spatial positioning data includes the following steps: Handwriting dynamics parameter extraction and analysis: Dynamic parameters are extracted from handwriting data from student terminals, including average writing pressure (grams), stroke speed (pixels / second), stroke direction change frequency (times / second), and pause duration (milliseconds). These parameters are then input into a lightweight LSTM model (64 hidden layers) in the edge intelligence layer to identify writing patterns, such as "hesitant" (high pause frequency + low speed), "fluent" (low pressure + high speed), and "corrective" (high direction change frequency). Combining the writing pattern with the current learning content, the student's cognitive state is inferred, such as "high cognitive load" (hesitant writing + complex math problems) and "shallow thinking" (fluent writing + open-ended questions). Generate personalized feedback: Based on the cognitive state inference results, intervention measures are matched from the feedback rule base (containing 50+ teaching strategies). For example, for students with "high cognitive load": push step-by-step problem-solving tips + reduce the difficulty of the questions; for students with "shallow thinking": trigger follow-up questions + recommend extended reading. Natural language prompts (such as "Please try to analyze from the perspective of X...") are generated through NLP models, AR annotations (such as red arrows indicating key steps) are generated next to the handwriting, and rhythm prompts (such as "speed up / slow down writing speed") are provided through the vibration motor of the student terminal. The feedback content is sent to the student terminal through the wireless transmission layer, supporting screen display (E Ink panel refresh rate ≤300ms), voice broadcast (TTS speech rate adjustable), and tactile prompts (vibration intensity graded control). Spatial positioning data acquisition: The UWB tag on the student terminal transmits pulses at a frequency of 10Hz. The base station on the teacher terminal calculates the location coordinates (accuracy <0.3m), performs Kalman filtering on the location data, generates the student's movement trajectory (sampling interval 100ms), and constructs an interaction relationship graph between students based on spatial proximity (<1m) and interaction duration (>30 seconds). Team collaboration layout optimization: The algorithm calculates the interaction density within groups (number of edges / number of members), assesses the efficiency of knowledge flow (based on the synchronicity of handwriting dynamics parameters), maximizes the diversity of interactions within groups (cross-gender / ability combinations), and balances the cognitive levels between groups (standard deviation <0.5). Using simulated annealing algorithm with location data as input, it optimizes the group seating arrangement, generates a new layout scheme, and sends a vibration alert (vibration mode encoding adjustment command) to the student terminal via UWB signal. It monitors group interaction data in real time (re-evaluating the layout every 5 minutes), and automatically triggers member rotation when a group is detected to be in a state of "groupthink" (handwriting similarity >0.8).

[0010] Furthermore, the method of protecting data privacy using physical layer security encryption includes the following steps: Channel feature acquisition preprocessing: The wireless modules (Wi-Fi 6E chips) of the teacher and student terminals periodically transmit Orthogonal Frequency Division Multiplexing (OFDM) probe frames to measure the channel impulse response (CIR). Channel characteristic parameters are extracted from the CIR, including: multipath delay spread (Δτ): reflecting the difference in signal propagation paths; channel frequency response (CFR): describing the changes in signal amplitude and phase; received signal strength indication (RSSI): quantizing the degree of signal attenuation; and the continuous channel characteristics are converted into discrete key bits using an equal probability quantization algorithm (quantization level = 256). Generate dynamic key: An initial shared key (K_initial, length = 256 bits) is established between the teacher's and student's terminals using the Diffie-Hellman key exchange protocol. The quantized channel characteristics are XORed with K_initial to generate a dynamic key (K_dynamic = K_initial ⊕ Quantized_CIR). Key synchronization is ensured through the control channel of the wireless transport layer using forward error correction coding (FEC) (synchronization success rate > 99.9%). Data encryption and transmission: The AES-256-GCM algorithm is used, and K_dynamic is used to encrypt the data to generate a 128-bit authentication tag (MAC). The encrypted data is mapped to OFDM subcarriers (subcarrier spacing = 78.125kHz) and 16-QAM modulation is used for high-order modulation efficiency. A directional beam (beamwidth = 15°) is formed through the 8×8 MIMO antenna array of the teacher terminal, with the beam direction aimed at the target student terminal, and the sidelobe suppression ratio >30dB. Enhanced security: Gaussian white noise (signal-to-noise ratio = 0dB) is added to the transmitted signal to interfere with potential eavesdroppers. The noise direction is optimized using zero-notch beamforming to minimize its impact on legitimate receivers. LDPC code (code length = 64800, code rate = 0.8) is employed for error correction performance close to the Shannon limit. The number of iterative decoding iterations is 10 to ensure a bit error rate < 0.5%. When a change in channel characteristics is detected (Δτ change > 10ns) or the amount of transmitted data reaches a threshold (1GB), a key update is triggered. Receiver decryption verification: The student terminal receives signals through a 2×2 MIMO antenna, performs channel estimation and equalization, and uses the minimum mean square error (MMSE) algorithm to suppress multipath interference. It extracts the current channel features, combines them with K_initial, regenerates K_dynamic, verifies key consistency through hash verification (SHA-3 algorithm), uses K_dynamic to decrypt the data, recovers the original information, and verifies the MAC tag to ensure that the data has not been tampered with.

[0011] Furthermore, the process of generating a virtual classroom image through a digital twin rendering cluster includes the following steps: Multi-source data acquisition: The system uses a LiDAR module (30Hz scanning frequency) on the teacher's terminal and UWB tags on the student's terminal to capture the 3D coordinates of all devices in the classroom in real time (accuracy <0.1m); it obtains the dynamic parameters of student handwriting (pressure, speed, and direction change frequency) from the edge intelligence layer, and captures the coordinates of key points on the student's face (68 feature points) through a micro-expression camera; it synchronizes the courseware content (4K HDR video stream) on the teacher's terminal, captures the digital ink data (SVG format handwriting) on ​​the student's terminal, and synchronizes the timestamps of all data streams through the NTP protocol (synchronization accuracy <1ms), binding spatial coordinates, behavioral parameters, and content data to a unified timeline; Building a digital twin model: Based on LiDAR point cloud data, a triangular mesh model of the classroom (number of vertices > 1 million) is generated. Through PBR, the model is given real-world physical properties (reflectivity, roughness, etc.). Based on the collected handwriting dynamics parameters and micro-expression data, the system drives the limb movements (skeletal animation) and facial expressions of the digital human (Blendshape technology). It maps the status data of equipment such as electronic blackboards and experimental equipment (such as courseware page numbers and experimental parameters) to the virtual model. It adds a physics engine (such as NVIDIA PhysX) to the virtual classroom mirror to simulate real-world interactive behaviors (such as book page turning and experimental equipment collision). Based on the teaching scenario knowledge graph (containing 32 subject ontology), it adds semantic tags to virtual objects (such as "Chemistry Experiment Table - Combustible Material Area"). Real-time rendering optimization: The virtual classroom image is divided into multiple rendering blocks (such as teacher area, student area, and device area). Rendering tasks are allocated to rendering nodes (equipped with RTX 6000 Ada graphics cards) through Kubernetes cluster scheduling. Deferred shading technology is used to separate geometry processing from shading processing, reducing the amount of computation. Instantiation rendering technology is used to process similar objects in batches (such as 20 student avatars), improving rendering efficiency. Based on network load, it automatically switches between 4K (3840×2160) and 8K (7680×4320) resolutions, and the polygon count is reduced for distant objects (to 10% of the original model). Interaction and feedback: Users select virtual experimental equipment via gesture recognition, triggering AR annotations (such as "Danger! High Temperature Area"). The annotation information is fixed to the corresponding position in the physical world using spatial anchoring technology (error <5cm). Historical classroom data (hybrid storage of time-series database and graph database) is retrieved from the interactive data lake and the classroom scene is replayed in timeline form, supporting fast forward / slow motion / pause operations (time accuracy <1 second). Through predictive coding technology (such as WebRTC's SVC scalable coding), the virtual classroom image is compressed to low bandwidth requirements (<10Mbps) to achieve spatiotemporal synchronization of multi-campus images (latency <200ms), supporting students across campuses to operate virtual experimental equipment together.

[0012] Furthermore, the provision of cross-campus collaborative teaching support to teachers includes the following steps: Data aggregation and synchronization across multiple campuses: The edge intelligence layer in classrooms of each campus collects classroom data (handwriting dynamics parameters, micro-expressions, spatial positioning, etc.) in real time and sends it to the cloud service layer through the wireless transmission layer with encryption; the interactive data lake of the cloud service layer (hybrid storage of time-series database and graph database) receives and aggregates data from multiple campuses, supporting millions of data points written per second; the timestamps of data from each campus are synchronized through the NTP protocol (synchronization accuracy <1ms) to build a globally unified spatiotemporal coordinate system; Virtual classroom mirroring: The digital twin rendering cluster generates an independent virtual classroom image for each campus (including student avatars, teaching equipment status, and classroom interaction data). Through spatial stitching algorithms, the virtual classroom images from multiple campuses are merged into a unified cross-campus teaching space (supporting synchronization of up to 8 campuses). The layout of the virtual classroom image is automatically adjusted according to the display resolution of the teacher's terminal (such as 21:9 ultrawide screen) (such as split-screen display and picture-in-picture mode). Cross-campus interaction optimization: By using predictive coding techniques (such as WebRTC's SVC scalable coding), the virtual classroom image is compressed to low bandwidth requirements (<10Mbps). Multipath transmission protocol (MPTCP) is adopted, and Wi-Fi 6E and 5G networks are utilized to improve transmission reliability (packet loss rate <0.1%). Timewarp algorithm is used to compensate for network transmission latency (up to 200ms latency can be offset). For voice and handwriting data of students from different campuses, forward error correction coding (FEC) is used. Design the teacher control interface: The unified view provides teachers with a cross-campus collaborative teaching control panel, supporting split-screen display of classroom images across multiple campuses (up to 4×4 layout), one-click switching of the main teaching campus (sent to student terminals via UWB signal with vibration alert), and cross-campus student group management (supporting dynamic adjustment of group members). After a teacher asks a question, the question is displayed simultaneously on the student terminals of each campus. The answers are summarized to the teacher's interface through voice transcription and handwriting recognition. Teachers can push courseware, 3D models and other resources to all student terminals of all campuses with one click (transmission latency <500ms). Teachers can operate experimental equipment across campuses in the virtual classroom, and student terminals receive AR annotations and operation prompts in real time. Security and privacy protection: Data transmitted across campuses is encrypted using AES-256-GCM, combined with physical layer security technologies (such as artificial noise injection) and a role-based access control (RBAC) model. Teachers can only access data from authorized campuses (e.g., lecturers can access data from all campuses, while teaching assistants are limited to their own campus). Sensitive information (such as student facial images and handwriting content) is automatically detected and filtered, in compliance with the Personal Information Protection Law.

[0013] The present invention has the following beneficial effects: In this invention, multimodal data acquisition is used to capture classroom dynamics in real time. After analysis by a lightweight AI model, participation heatmaps, personalized handwriting feedback, and group layout optimization strategies are generated. The results are synchronized to the cloud service layer using a customized Wi-Fi 6E protocol. The cloud service layer aggregates multi-classroom data lakes through a teaching strategy engine, runs a reinforcement learning model to optimize the classroom pace, and finally generates a virtual classroom image shared across campuses through a digital twin rendering cluster. This forms a closed loop of "perception-analysis-feedback-optimization," effectively solving the problem of poor teaching interactivity, greatly improving teaching and learning efficiency, and potentially stimulating students' learning enthusiasm. Attached Figure Description

[0014] Figure 1 This is a system block diagram of an electronic screen classroom teaching system based on wireless network connection proposed in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Please see Figure 1 As shown, the present invention is an electronic screen classroom teaching system based on wireless network connection, comprising: Perception and Interaction Layer: Through the teacher's terminal's UWB positioning, gesture recognition, and voice emotion analysis modules, classroom behavior data (such as teacher movement trajectory, gesture commands, and voice characteristics) is captured in real time. At the same time, individual learning status is collected through the pressure-sensing handwriting and micro-expression cameras on the student's terminal. After local preprocessing, this multimodal data is encrypted and sent to the edge intelligence layer through the wireless transmission layer, providing raw materials for enhancing interactivity. For example, the teacher's gestures can instantly trigger courseware animations, and changes in the pressure of the student's handwriting can reflect the degree of confusion. Edge Intelligence Layer: After receiving data from the Perception and Interaction Layer, it runs a lightweight AI model for real-time analysis: it judges student participation by recognizing voice emotion features and micro-expressions, generates personalized feedback using handwriting dynamics parameters, and optimizes group collaboration layout with the help of spatial positioning data; the analysis results are synchronized to the cloud service layer via the wireless transmission layer to train the global model, and at the same time, local strategies (such as adjusting the frequency of questions and pushing supplementary materials) are sent back to the Perception and Interaction Layer for execution. Wireless transmission layer: Adopting a customized Wi-Fi 6E protocol, it prioritizes the transmission of interactive data packets (such as gesture commands and emotional features) through dynamic spectrum allocation, and uses physical layer security encryption to ensure data privacy; This layer serves as a bridge between the perception and interaction layer and the edge / cloud layer, efficiently uploading the raw data collected by the terminal to the edge node for processing, and distributing real-time strategies generated at the edge (such as student group information) and long-term optimization suggestions from the cloud (such as adjustments to the teaching pace) to the terminal; Cloud service layer: Aggregates multi-classroom interactive data lake, runs reinforcement learning models through teaching strategy engine, optimizes classroom pace based on historical interaction patterns (such as automatically adjusting the explanation / practice ratio), and generates virtual classroom images through digital twin rendering cluster; the optimized strategy is fed back to edge intelligence layer through wireless transmission layer for localization adaptation, and at the same time provides teachers with cross-campus joint teaching support (such as sharing AR experiment guidance).

[0017] In one embodiment, the real-time capture of classroom behavior data via the teacher's terminal's UWB positioning, gesture recognition, and voice emotion analysis modules includes the following steps: UWB (Ultra-Wideband) positioning: The teacher's terminal is equipped with a lidar module that scans the classroom space at a frequency of 30Hz to construct a three-dimensional point cloud map. The UWB tag worn by the teacher periodically emits pulse signals (pulse width <2ns), and the student terminal synchronously emits response signals. By measuring the time difference of flight of the signals between the teacher's terminal and the student terminal, and combining it with a multi-point positioning algorithm, sub-meter-level spatial positioning is achieved. Based on the Kalman filter algorithm, the positioning data is smoothed to generate the teacher's movement trajectory (sampling interval 50ms). Gesture recognition algorithm library: The teacher terminal's built-in RGB-D camera captures hand movements at 60FPS. The MediaPipe framework is used to extract the 3D coordinates of 21 hand joints. The spatial coordinate sequence of the joints is matched with 26 pre-trained teaching scene gesture templates (such as "page turning", "zooming", and "annotating") using dynamic time warping (DTW). When the matching degree exceeds the threshold (>0.85), the corresponding courseware control command is triggered and sent to the main screen through the wireless transmission layer. Voice sentiment analysis: A circular microphone array uses beamforming technology to lock the direction of the teacher's voice source (azimuth accuracy ±3°), performs short-time Fourier transform on the 16kHz sampled audio, extracts 12-dimensional MFCC features and fundamental frequency and energy parameters; inputs them into a lightweight Transformer model (parameters compressed to 3.7M), outputs probability values ​​for excitement and confusion emotion dimensions, binds emotion tags to speech segment timestamps, and generates a classroom speech stream with emotion annotations; Data fusion: The timestamps of the UWB positioning, gesture recognition, and voice emotion analysis modules are synchronized via the NTP protocol (synchronization accuracy <10ms). The isolated forest algorithm is used to detect positioning drift points (velocity change >5m / s) and emotion change points (probability jump >0.6). The effective data is encapsulated in JSON format, which includes four parts: device ID, data type, timestamp, and payload. The data packets are encrypted using the AES-256-GCM algorithm and sent to the edge intelligence layer through the wireless transmission layer.

[0018] In one embodiment, the process of collecting individual learning status data via a pressure-sensitive handwriting and micro-expression camera on a student terminal includes the following steps: Pressure-sensitive handwriting capture: When students use the smart pen to write on the foldable e-paper notebook, the pen tip pressure sensor (512-level resolution) senses the writing pressure in real time, and the six-axis sensor (three-axis accelerometer + three-axis gyroscope) captures the pen tip movement trajectory at a sampling rate of 200Hz, generating a three-dimensional spatial coordinate sequence. The trajectory is optimized by a handwriting smoothing algorithm (such as the Ramer-Douglas-Peucker algorithm) and converted into digital ink data in SVG format. Metadata tags including average pressure value (unit: grams), writing speed (pixels / second), and stroke direction angle (0°-360°) are added to each stroke. Micro-expression camera capture: The student terminal's front-facing camera captures facial images at 30 FPS with a resolution of 640×480. A lightweight face detection model (such as UltraFace) is used to perform real-time detection on the CPU (detection latency <30ms). The coordinates of 68 facial key points are extracted using the MobileFaceNet model, with a focus on annotating the eyes (orbicularis oculi muscle region) and the mouth (orbicularis oris muscle region). Micro-expression coding: Eye area: Calculate eyelid closure degree (EAR value) and blink frequency; Mouth area: Calculate mouth opening (MAR value) and corner of the mouth upward angle; Overall facial features: Calculate head attitude angles (pitch angle, yaw angle, roll angle); Multimodal data fusion: By synchronizing the timestamps of handwriting data and micro-expression data using the PTP protocol (synchronization accuracy <100ms), a time series correlation is established between handwriting data (such as the problem-solving process) and micro-expression changes (such as confusion / sudden realization); based on the frequency of handwriting pauses (>2 seconds) and the confusion level of micro-expressions (decreasing EAR value), a correlation analysis is conducted using handwriting accuracy (matching degree with standard answer) and smiling frequency (increasing MAR value); Data preprocessing and transmission: An SVG path simplification algorithm is used to reduce the amount of data (compression ratio ≥ 4:1), extract keyframes (1 frame per second), and retain peak points of facial expression changes; the facial images are blurred (Gaussian blur σ=5) to retain only the coordinates of key points; and the handwriting data is desensitized to remove sensitive information (such as name and student ID). The data is encrypted using the AES-128-CBC algorithm and then transmitted to the edge intelligence layer via the wireless transmission layer.

[0019] In one embodiment, determining student participation through voice emotion features and micro-expression recognition includes the following steps: Multimodal data acquisition: The microphone array of the student terminal captures speech segments at a sampling rate of 48kHz, uses beamforming technology to locate the speaking student, and performs echo cancellation and noise suppression (SNR improvement ≥15dB); endpoint detection (VAD) is performed on the speech, 13-dimensional MFCC features and fundamental frequency (F0) and energy (RMS) parameters are extracted, and a 200ms speech feature frame is constructed; the RGB camera of the student terminal captures facial images at a frame rate of 30FPS, detects 68 facial key points using the MTCNN algorithm, and performs affine transformation for face alignment; ROI images of the eyes (orbicularis oculi muscle region) and mouth (orbicularis oris muscle region) are cropped, scaled to 64×64 pixels, converted to grayscale images, and pixel values ​​are normalized; Feature extraction for emotion classification: The input is fed into a lightweight Transformer model (MobileBERT architecture, with parameters compressed to 12M), which captures temporal dependencies in speech features through a self-attention mechanism; it outputs probability values ​​for the emotion dimensions of arousal and confusion, with dimensional quantization using a 5-level Likert scale; an improved FER2013 model (with added attention mechanism) is used to extract features from eye and mouth ROI images, identify 6 basic emotions (happiness, surprise, confusion, neutrality, annoyance, and frustration), and output the confidence scores for each category; Participation fusion calculation: Synchronize timestamps of speech emotion features and micro-expression features via NTP protocol (synchronization accuracy <50ms); weighted fusion: The emotional weight of the voice is ω1 = 0.6 (excitement level × 0.7 + confusion level × 0.3). Micro-expression weight ω2=0.4 (mean confidence score for positive emotions); Engagement score S = ω1×(vocal emotion score) + ω2×(micro-expression score); A Gaussian kernel function (σ=1.5m) was used to spatially smooth the participation score, generating a classroom participation heatmap (resolution 0.5m / pixel). Real-time feedback optimization: When the mean regional participation rate is <0.3 (5-point scale), a low participation warning is triggered. For low participation areas, the edge intelligence layer sends instructions to the teacher's terminal via the wireless transmission layer, suggesting adjustments to the teaching method (such as adding questions / interactions), pushing personalized learning resources (based on students' historical learning data), and simultaneously generating group collaboration tasks (such as "Please have students in this area discuss case X"). These tasks are then sent to the student's terminal via UWB signal for vibration alerts. Based on real-time feedback from the teacher (such as gesture confirmation / rejection suggestions), the strategy generation logic is updated through a reinforcement learning model (Q-learning).

[0020] In one embodiment, generating personalized feedback using handwriting dynamics parameters and optimizing the team collaboration layout using spatial positioning data includes the following steps: Handwriting dynamics parameter extraction and analysis: Dynamic parameters are extracted from handwriting data from student terminals, including average writing pressure (grams), stroke speed (pixels / second), stroke direction change frequency (times / second), and pause duration (milliseconds). These parameters are then input into a lightweight LSTM model (64 hidden layers) in the edge intelligence layer to identify writing patterns, such as "hesitant" (high pause frequency + low speed), "fluent" (low pressure + high speed), and "corrective" (high direction change frequency). Combining the writing pattern with the current learning content, the student's cognitive state is inferred, such as "high cognitive load" (hesitant writing + complex math problems) and "shallow thinking" (fluent writing + open-ended questions). Generate personalized feedback: Based on the cognitive state inference results, intervention measures are matched from the feedback rule base (containing 50+ teaching strategies). For example, for students with "high cognitive load": push step-by-step problem-solving tips + reduce the difficulty of the questions; for students with "shallow thinking": trigger follow-up questions + recommend extended reading. Natural language prompts (such as "Please try to analyze from the perspective of X...") are generated through NLP models, AR annotations (such as red arrows indicating key steps) are generated next to the handwriting, and rhythm prompts (such as "speed up / slow down writing speed") are provided through the vibration motor of the student terminal. The feedback content is sent to the student terminal through the wireless transmission layer, supporting screen display (E Ink panel refresh rate ≤300ms), voice broadcast (TTS speech rate adjustable), and tactile prompts (vibration intensity graded control). Spatial positioning data acquisition: The UWB tag on the student terminal transmits pulses at a frequency of 10Hz. The base station on the teacher terminal calculates the location coordinates (accuracy <0.3m), performs Kalman filtering on the location data, generates the student's movement trajectory (sampling interval 100ms), and constructs an interaction relationship graph between students based on spatial proximity (<1m) and interaction duration (>30 seconds). Team collaboration layout optimization: The algorithm calculates the interaction density within groups (number of edges / number of members), assesses the efficiency of knowledge flow (based on the synchronicity of handwriting dynamics parameters), maximizes the diversity of interactions within groups (cross-gender / ability combinations), and balances the cognitive levels between groups (standard deviation <0.5). Using simulated annealing algorithm with location data as input, it optimizes the group seating arrangement, generates a new layout scheme, and sends a vibration alert (vibration mode encoding adjustment command) to the student terminal via UWB signal. It monitors group interaction data in real time (re-evaluating the layout every 5 minutes), and automatically triggers member rotation when a group is detected to be in a state of "groupthink" (handwriting similarity >0.8).

[0021] In one embodiment, the method of protecting data privacy using physical layer security encryption includes the following steps: Channel feature acquisition preprocessing: The wireless modules (Wi-Fi 6E chips) of the teacher and student terminals periodically transmit Orthogonal Frequency Division Multiplexing (OFDM) probe frames to measure the channel impulse response (CIR). Channel characteristic parameters are extracted from the CIR, including: multipath delay spread (Δτ): reflecting the difference in signal propagation paths; channel frequency response (CFR): describing the amplitude and phase changes of the signal; received signal strength indication (RSSI): quantizing the degree of signal attenuation; continuous channel characteristics are converted into discrete key bits, and an equal probability quantization algorithm (quantization level = 256) is used. Generate dynamic key: An initial shared key (K_initial, length = 256 bits) is established between the teacher's and student's terminals using the Diffie-Hellman key exchange protocol. The quantized channel characteristics are XORed with K_initial to generate a dynamic key (K_dynamic = K_initial ⊕ Quantized_CIR). Key synchronization is ensured through the control channel of the wireless transport layer using forward error correction coding (FEC) (synchronization success rate > 99.9%). Data encryption and transmission: The AES-256-GCM algorithm is used, and K_dynamic is used to encrypt the data to generate a 128-bit authentication tag (MAC). The encrypted data is mapped to OFDM subcarriers (subcarrier spacing = 78.125kHz) and 16-QAM modulation is used for high-order modulation efficiency. A directional beam (beamwidth = 15°) is formed through the 8×8 MIMO antenna array of the teacher terminal, with the beam direction aimed at the target student terminal, and the sidelobe suppression ratio >30dB. Enhanced security: Gaussian white noise (signal-to-noise ratio = 0dB) is added to the transmitted signal to interfere with potential eavesdroppers. The noise direction is optimized using zero-notch beamforming to minimize its impact on legitimate receivers. LDPC code (code length = 64800, code rate = 0.8) is employed for error correction performance close to the Shannon limit. The number of iterative decoding iterations is 10 to ensure a bit error rate < 0.5%. When a change in channel characteristics is detected (Δτ change > 10ns) or the amount of transmitted data reaches a threshold (1GB), a key update is triggered. Receiver decryption verification: The student terminal receives signals through a 2×2 MIMO antenna, performs channel estimation and equalization, and uses the minimum mean square error (MMSE) algorithm to suppress multipath interference. It extracts the current channel features, combines them with K_initial, regenerates K_dynamic, verifies key consistency through hash verification (SHA-3 algorithm), uses K_dynamic to decrypt the data, recovers the original information, and verifies the MAC tag to ensure that the data has not been tampered with.

[0022] In one embodiment, generating a virtual classroom image via a digital twin rendering cluster includes the following steps: Multi-source data acquisition: The system uses a LiDAR module (30Hz scanning frequency) on the teacher's terminal and UWB tags on the student's terminal to capture the 3D coordinates of all devices in the classroom in real time (accuracy <0.1m); it obtains the dynamic parameters of student handwriting (pressure, speed, and direction change frequency) from the edge intelligence layer, and captures the coordinates of key points on the student's face (68 feature points) through a micro-expression camera; it synchronizes the courseware content (4K HDR video stream) on the teacher's terminal, captures the digital ink data (SVG format handwriting) on ​​the student's terminal, and synchronizes the timestamps of all data streams through the NTP protocol (synchronization accuracy <1ms), binding spatial coordinates, behavioral parameters, and content data to a unified timeline; Building a digital twin model: Based on LiDAR point cloud data, a triangular mesh model of the classroom (number of vertices > 1 million) is generated. Through PBR, the model is given real-world physical properties (reflectivity, roughness, etc.). Based on the collected handwriting dynamics parameters and micro-expression data, the system drives the limb movements (skeletal animation) and facial expressions of the digital human (Blendshape technology). It maps the status data of equipment such as electronic blackboards and experimental equipment (such as courseware page numbers and experimental parameters) to the virtual model. It adds a physics engine (such as NVIDIA PhysX) to the virtual classroom mirror to simulate real-world interactive behaviors (such as book page turning and experimental equipment collision). Based on the teaching scenario knowledge graph (containing 32 subject ontology), it adds semantic tags to virtual objects (such as "Chemistry Experiment Table - Combustible Material Area"). Real-time rendering optimization: The virtual classroom image is divided into multiple rendering blocks (such as teacher area, student area, and device area). Rendering tasks are allocated to rendering nodes (equipped with RTX 6000 Ada graphics cards) through Kubernetes cluster scheduling. Deferred shading technology is used to separate geometry processing from shading processing, reducing the amount of computation. Instantiation rendering technology is used to process similar objects in batches (such as 20 student avatars), improving rendering efficiency. Based on network load, it automatically switches between 4K (3840×2160) and 8K (7680×4320) resolutions, and the polygon count is reduced for distant objects (to 10% of the original model). Interaction and feedback: Users select virtual experimental equipment via gesture recognition, triggering AR annotations (such as "Danger! High Temperature Area"). The annotation information is fixed to the corresponding position in the physical world using spatial anchoring technology (error <5cm). Historical classroom data (hybrid storage of time-series database and graph database) is retrieved from the interactive data lake and the classroom scene is replayed in timeline form, supporting fast forward / slow motion / pause operations (time accuracy <1 second). Through predictive coding technology (such as WebRTC's SVC scalable coding), the virtual classroom image is compressed to low bandwidth requirements (<10Mbps) to achieve spatiotemporal synchronization of multi-campus images (latency <200ms), supporting students across campuses to operate virtual experimental equipment together.

[0023] In one embodiment, providing cross-campus collaborative teaching support to teachers includes the following steps: Data aggregation and synchronization across multiple campuses: The edge intelligence layer in classrooms of each campus collects classroom data (handwriting dynamics parameters, micro-expressions, spatial positioning, etc.) in real time and sends it to the cloud service layer through the wireless transmission layer with encryption; the interactive data lake of the cloud service layer (hybrid storage of time-series database and graph database) receives and aggregates data from multiple campuses, supporting millions of data points written per second; the timestamps of data from each campus are synchronized through the NTP protocol (synchronization accuracy <1ms) to build a globally unified spatiotemporal coordinate system; Virtual classroom mirroring: The digital twin rendering cluster generates an independent virtual classroom image for each campus (including student avatars, teaching equipment status, and classroom interaction data). Through spatial stitching algorithms, the virtual classroom images from multiple campuses are merged into a unified cross-campus teaching space (supporting synchronization of up to 8 campuses). The layout of the virtual classroom image is automatically adjusted according to the display resolution of the teacher's terminal (such as 21:9 ultrawide screen) (such as split-screen display and picture-in-picture mode). Cross-campus interaction optimization: By using predictive coding techniques (such as WebRTC's SVC scalable coding), the virtual classroom image is compressed to low bandwidth requirements (<10Mbps). Multipath transmission protocol (MPTCP) is adopted, and Wi-Fi 6E and 5G networks are utilized to improve transmission reliability (packet loss rate <0.1%). Timewarp algorithm is used to compensate for network transmission latency (up to 200ms latency can be offset). For voice and handwriting data of students from different campuses, forward error correction coding (FEC) is used. Design the teacher control interface: The unified view provides teachers with a cross-campus collaborative teaching control panel, supporting split-screen display of classroom images across multiple campuses (up to 4×4 layout), one-click switching of the main teaching campus (sent to student terminals via UWB signal with vibration alert), and cross-campus student group management (supporting dynamic adjustment of group members). After a teacher asks a question, the question is displayed simultaneously on the student terminals of each campus. The answers are summarized to the teacher's interface through voice transcription and handwriting recognition. Teachers can push courseware, 3D models and other resources to all student terminals of all campuses with one click (transmission latency <500ms). Teachers can operate experimental equipment across campuses in the virtual classroom, and student terminals receive AR annotations and operation prompts in real time. Security and privacy protection: Data transmitted across campuses is encrypted using AES-256-GCM, combined with physical layer security technologies (such as artificial noise injection) and a role-based access control (RBAC) model. Teachers can only access data from authorized campuses (e.g., lecturers can access data from all campuses, while teaching assistants are limited to their own campus). Sensitive information (such as student facial images and handwriting content) is automatically detected and filtered, in compliance with the Personal Information Protection Law.

[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An electronic screen classroom teaching system based on wireless network connection, characterized in that, include: Perception and Interaction Layer: Classroom behavior data is captured in real time through the teacher terminal's UWB positioning, gesture recognition, and voice emotion analysis modules. At the same time, individual learning status data is collected through the student terminal's pressure-sensing handwriting and micro-expression camera. The data is pre-processed locally and then encrypted and sent to the edge intelligence layer through the wireless transmission layer. Edge Intelligence Layer: After receiving data from the Perception and Interaction Layer, it runs a lightweight AI model for real-time analysis: it judges student participation by recognizing voice emotion features and micro-expressions, generates personalized feedback using handwriting dynamics parameters, and optimizes the group collaboration layout with the help of spatial positioning data; the analysis results are synchronized to the cloud service layer via the wireless transmission layer to train the global model, and the local strategy is sent back to the Perception and Interaction Layer for execution. Wireless transmission layer: It adopts a customized Wi-Fi 6E protocol, prioritizes the transmission of interactive data packets through dynamic spectrum allocation, and uses physical layer security encryption to protect data privacy; Cloud service layer: Aggregates multi-classroom interactive data lake, runs reinforcement learning models through teaching strategy engine, optimizes classroom pace based on historical interaction patterns, and generates virtual classroom images through digital twin rendering cluster; The optimized strategy is fed back to the edge intelligence layer via the wireless transmission layer for localization adaptation, while also providing teachers with cross-campus collaborative teaching support.

2. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The process of capturing classroom behavior data in real time through the teacher's terminal's UWB positioning, gesture recognition, and voice emotion analysis modules includes the following steps: UWB (Ultra-Wideband) positioning: The teacher's terminal is equipped with a lidar module that scans the classroom space at a frequency of 30Hz to construct a three-dimensional point cloud map; the teacher wears a UWB tag and periodically emits pulse signals, and the student terminal simultaneously emits response signals; spatial positioning is performed by measuring the time difference of flight of the signals between the teacher's terminal and the student terminal, combined with a multi-point positioning algorithm; the positioning data is smoothed based on the Kalman filter algorithm to generate the teacher's movement trajectory. Gesture recognition algorithm library: The teacher terminal's built-in RGB-D camera captures hand movements at 60FPS. The MediaPipe framework is used to extract the three-dimensional coordinates of 21 hand joints. The spatial coordinate sequence of the joints is dynamically time-warped and matched with 26 pre-trained gesture templates for teaching scenarios. When the matching degree exceeds the threshold, the corresponding courseware control command is triggered and sent to the main screen through the wireless transmission layer. Voice sentiment analysis: A circular microphone array uses beamforming technology to lock the direction of the teacher's voice source, performs a short-time Fourier transform on the 16kHz sampled audio, extracts 12-dimensional MFCC features and fundamental frequency and energy parameters; inputs them into a lightweight Transformer model, outputs probability values ​​of excitement and confusion emotion dimensions, binds emotion tags with speech segment timestamps, and generates a classroom speech stream with emotion annotations; Data fusion: The timestamps of the UWB positioning, gesture recognition, and voice emotion analysis modules are synchronized via the NTP protocol. The isolated forest algorithm is used to detect positioning drift points and emotion mutation points. The effective data is encapsulated in JSON format, which includes four parts: device ID, data type, timestamp, and payload. The data packets are encrypted using the AES-256-GCM algorithm and sent to the edge intelligence layer through the wireless transmission layer.

3. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The process of collecting individual learning status data through pressure-sensitive handwriting and micro-expression cameras on student terminals includes the following steps: Pressure-sensitive handwriting capture: When students use the smart pen to write on the foldable e-paper notebook, the pen tip pressure sensor senses the writing pressure in real time, the six-axis sensor captures the pen tip movement trajectory at a sampling rate of 200Hz, generates a three-dimensional spatial coordinate sequence, optimizes the trajectory through a handwriting smoothing algorithm, converts it into digital ink data in SVG format, and adds metadata tags including average pressure value, writing speed, and stroke direction angle to each stroke. Micro-expression camera capture: The student terminal's front-facing camera captures facial images at 30 FPS with a resolution of 640×480. A lightweight face detection model is adopted to perform real-time detection on the CPU. The coordinates of 68 facial key points are extracted through the MobileFaceNet model, with special emphasis on annotating the eyes and mouth. Micro-expression coding: Eye area: Calculate eyelid closure degree and blink frequency; Mouth area: Calculate mouth opening and corner of mouth upward angle; Overall facial features: Calculate head posture angles; Multimodal data fusion: The timestamps of handwriting data and micro-expression data are synchronized using the PTP protocol to establish a time series correlation between handwriting data and micro-expression changes; a comprehensive judgment is made based on the frequency of handwriting pauses and the confusion level of micro-expressions, and a correlation analysis is performed between handwriting accuracy and smile frequency. Data preprocessing and transmission: The SVG path simplification algorithm is used to reduce the amount of data, extract key frames, and retain the peak points of facial expression changes; the facial images are blurred, and only the coordinates of key points are retained; the handwriting data is desensitized to remove sensitive information; the data is encrypted using the AES-128-CBC algorithm and sent to the edge intelligence layer through the wireless transmission layer.

4. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The method of determining student participation through voice emotion features and micro-expression recognition includes the following steps: Multimodal data acquisition: The student terminal's microphone array captures speech segments at a 48kHz sampling rate, uses beamforming technology to locate the speaking student, and performs echo cancellation and noise suppression. Endpoint detection is performed on the speech, extracting 13-dimensional MFCC features, fundamental frequency, and energy parameters to construct a 200ms-long speech feature frame. The student terminal's RGB camera captures facial images at 30FPS, detects 68 facial keypoints using the MTCNN algorithm, and performs affine transformation for face alignment. Regions of Interest (ROI) images of the eyes and mouth are cropped, pixels are scaled, converted to grayscale, and pixel values ​​are normalized. Feature extraction for emotion classification: The input is fed into a lightweight Transformer model (MobileBERT architecture, parameter compression, capturing temporal dependencies in speech features through a self-attention mechanism; outputting probability values ​​for excitement, confusion, and emotion dimensions, with dimensional quantization using a 5-level Likert scale; using an improved FER2013 model, features are extracted from eye and mouth ROI images to identify 6 basic emotions and output confidence scores for each category). Participation fusion calculation: Synchronize timestamps of voice emotion features and micro-expression features via NTP protocol; weighted fusion: Voice emotion weighting ; Micro-expression weight ; Participation score ; Where a is b is ; A Gaussian kernel function was used to spatially smooth the participation score, generating a classroom participation heatmap. Real-time feedback optimization: When the average regional participation rate is less than 0.3, a low participation warning is triggered. For low participation areas, the edge intelligence layer sends instructions to the teacher's terminal through the wireless transmission layer, suggesting adjustments to the teaching method, pushing personalized learning resources, and simultaneously generating group collaboration tasks. These tasks are then sent to the student's terminal via UWB signal for vibration alerts. Based on the teacher's real-time feedback, the strategy generation logic is updated through the reinforcement learning model.

5. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The method of generating personalized feedback using handwriting dynamics parameters and optimizing team collaboration layout using spatial positioning data includes the following steps: Handwriting dynamics parameter extraction and analysis: Dynamic parameters are extracted from the handwriting data of the student terminal, including average writing pressure, stroke speed, frequency of stroke direction change, and pause duration. The parameters are then input into a lightweight LSTM model in the edge intelligence layer to identify writing patterns. By combining the writing patterns with the current learning content, the student's cognitive state is inferred. Generate personalized feedback: Based on the inference results of cognitive state, intervention measures are matched from the feedback rule base. Step-by-step problem-solving prompts are pushed to students with "high cognitive load" and the difficulty of the questions is reduced. Follow-up questions are triggered for students with "shallow thinking" and extended reading is recommended. Natural language prompts are generated through NLP models, AR annotations are generated next to the handwriting, and rhythm prompts are provided through the vibration motor of the student terminal. The feedback content is sent to the student terminal through the wireless transmission layer, supporting screen display, voice broadcast and tactile prompts. Spatial positioning data acquisition: The UWB tag on the student terminal transmits pulses at a frequency of 10Hz. The base station on the teacher terminal calculates the location coordinates, performs Kalman filtering on the location data, generates the student's movement trajectory, and constructs an interaction relationship graph among students based on spatial proximity and interaction duration. Team collaboration layout optimization: The algorithm calculates the interaction density within groups, assesses the efficiency of knowledge flow, maximizes the diversity of interactions within groups, and balances the cognitive levels between groups. Using simulated annealing algorithm with location data as input, it optimizes the seating arrangement of groups, generates a new layout scheme, and sends it to the student terminal via UWB signal for vibration alerts. It monitors group interaction data in real time, and automatically triggers member rotation when it detects that a group has fallen into "groupthink".

6. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The use of physical layer security encryption to protect data privacy Includes the following steps: Channel feature acquisition and preprocessing: The wireless modules of the teacher's terminal and the student's terminal periodically transmit orthogonal frequency division multiplexing probe frames to measure the channel impulse response (CIR). Channel characteristic parameters are extracted from the CIR, including: multipath delay spread (reflecting differences in signal propagation paths); channel frequency response (describing signal amplitude and phase changes); received signal strength indication (quantizing the degree of signal attenuation); and continuous channel characteristics are converted into discrete key bits using an equal probability quantization algorithm. Generate dynamic key: An initial shared key is established between the teacher's terminal and the student's terminal using the Diffie-Hellman key exchange protocol; the quantized channel characteristics are XORed with K_initial to generate a dynamic key; and forward error correction coding is used to adjust key synchronization through the control channel of the wireless transport layer. Data encryption and transmission: The AES-256-GCM algorithm is used, and K_dynamic is used to encrypt the data to generate a 128-bit authentication tag. The encrypted data is mapped to OFDM subcarriers and 16-QAM modulation is used to achieve high-order modulation efficiency. A directional beam is formed through the 8×8 MIMO antenna array of the teacher terminal, and the beam direction is aimed at the target student terminal. Enhanced security: Gaussian white noise is added to the transmitted signal to interfere with potential eavesdroppers. The noise direction is reduced to minimize its impact on legitimate receivers using zero-contour beamforming technology. LDPC codes are used to achieve error correction performance close to the Shannon limit. The number of iterative decodings is 10 to reduce the bit error rate. When a change in channel characteristics is detected or the amount of transmitted data reaches a threshold, a key update is triggered. Receiver decryption verification: The student terminal receives signals through a 2×2 MIMO antenna, performs channel estimation and equalization, and uses the minimum mean square error algorithm to suppress multipath interference. It extracts the current channel features, combines them with K_initial, regenerates K_dynamic, verifies key consistency through hash verification, uses K_dynamic to decrypt the data, recovers the original information, and verifies the MAC tag to ensure that the data has not been tampered with.

7. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The process of generating a virtual classroom image using a digital twin rendering cluster includes the following steps: Multi-source data acquisition: The system captures the 3D coordinates of all devices in the classroom in real time through the lidar module on the teacher's terminal and the UWB tag on the student's terminal; it obtains the dynamic parameters of the student's handwriting from the edge intelligence layer and captures the coordinates of key points on the student's face through the micro-expression camera; it synchronizes the courseware content on the teacher's terminal, captures the digital ink data on the student's terminal, and synchronizes the timestamps of all data streams through the NTP protocol, binding spatial coordinates, behavioral parameters, and content data to a unified timeline. Building a digital twin model: Based on LiDAR point cloud data, a triangular mesh model of the classroom is generated. Through PBR, the model is given real-world physical properties. Based on the collected handwriting dynamics parameters and micro-expression data, the body movements and facial expressions of the digital human are driven. The status data of the electronic blackboard and experimental equipment are mapped to the virtual model. A physics engine is added to the virtual classroom mirror to simulate real-world interactive behavior. Based on the knowledge graph of teaching scenarios, semantic tags are added to virtual objects. Real-time rendering optimization: The virtual classroom image is divided into multiple rendering blocks. Rendering tasks are allocated to rendering nodes through Kubernetes cluster scheduling. Deferred shading technology is used to separate geometry processing from shading processing, reducing the amount of computation. Instantiated rendering technology is used to process similar objects in batches, improving rendering efficiency. The system automatically switches between 4K and 8K based on network load and reduces the number of polygons for distant objects. Interaction and feedback: Users select virtual experimental equipment through gesture recognition, triggering AR annotation. The annotation information is fixed to the corresponding position in the physical world through spatial anchoring technology. Historical classroom data is retrieved from the interactive data lake, and classroom scenes are replayed in timeline form, supporting fast forward / slow motion / pause operations. Through predictive coding technology, the virtual classroom image is compressed to low bandwidth requirements, and spatiotemporal synchronization of multi-campus images is achieved, supporting students across campuses to operate virtual experimental equipment together.

8. The electronic screen classroom teaching system based on wireless network connection according to claim 1, characterized in that, The provision of cross-campus collaborative teaching support to teachers includes the following steps: Data aggregation and synchronization across multiple campuses: The edge intelligence layer in each campus classroom collects classroom data in real time and transmits it to the cloud service layer via a wireless transmission layer with encryption; the interactive data lake in the cloud service layer receives and aggregates data from multiple campuses; and the timestamps of data from each campus are synchronized through the NTP protocol to construct a globally unified spatiotemporal coordinate system. Virtual classroom mirroring: The digital twin rendering cluster generates an independent virtual classroom image for each campus. Through a spatial stitching algorithm, the virtual classroom images from multiple campuses are merged into a unified cross-campus teaching space. The layout of the virtual classroom image is automatically adjusted according to the display resolution of the teacher's terminal. Cross-campus interaction optimization: By using predictive coding technology, the virtual classroom image is compressed to low bandwidth requirements. A multipath transmission protocol is adopted, and Wi-Fi 6E and 5G networks are used to improve transmission reliability. A time warp algorithm is used to compensate for network transmission latency. Forward error correction coding is used for the voice and handwriting data of students across campuses. Design the teacher control interface: The unified view provides teachers with a cross-campus collaborative teaching control panel, supporting split-screen display of classroom images across multiple campuses and one-click switching of the main teaching campus; cross-campus student group management, after the teacher asks a question, the question is displayed synchronously on the student terminals of each campus, and the answers are summarized to the teacher's interface through voice transcription and handwriting recognition; the teacher can push courseware and 3D model resources to the student terminals of all campuses with one click; the teacher operates cross-campus experimental equipment in the virtual classroom, and the student terminals receive AR annotations and operation prompts in real time; Security and privacy protection: Data transmitted across campuses is encrypted using AES-256-GCM, combined with physical layer security technology and a role-based access control model, ensuring that teachers can only access data from authorized campuses; it also automatically detects and filters sensitive information, complying with the requirements of the Personal Information Protection Law.

Images