Intelligent monitoring device with anti-disassembly function

Abstract

The invention relates to an intelligent monitoring device with an anti-disassembly function. A built-in optical fiber sign sensor accurately monitors parameters such as heart rate, respiration, snoring, apnea, body movement, bed on / off, sudden heartbeat arrest and the like of a user; the MEMS sensor is combined with the sign sensor, so that the snore and apnea monitoring accuracy is improved; and the MEMS sensor and the photoelectric sensor realize illegal disassembly alarm. The intelligent monitoring device adopts a special structural design, can be normally disassembled only by using a special tool, and gives a real-time alarm according to abnormal alarm data reported by the sensor when the intelligent monitoring device is illegally disassembled. Through the synergistic effect of the components, parameters such as the heart rate, respiration, snoring, apnea, sudden heartbeat arrest, body movement and on-bed / off-bed of a user can be detected, potential safety hazards such as sudden death and sudden diseases can be effectively prevented in a supervision scene, and digital and intelligent decision-making assistance is provided for intelligent supervision.

Description

Technical Field

[0001] This invention relates to the field of intelligent sleep monitoring devices, and more particularly to an intelligent monitoring device with anti-disassembly function. Background Technology

[0002] In special supervisory settings such as prisons, there are extremely stringent requirements for monitoring the vital signs and managing the safety of detainees. Currently, prison supervision mainly relies on traditional methods such as manual patrols and video surveillance, as well as some intelligent devices with basic functions. However, these existing methods and technologies have many shortcomings.

[0003] Manual patrols are a traditional method of prison supervision, but they suffer from inefficiency and susceptibility to subjective factors. Patrol personnel need to regularly move between cells, which not only consumes significant manpower and time, but also makes it impossible to monitor inmates' movements in real time during the intervals between patrols. Especially at night or in special circumstances that extend patrol intervals, gaps in supervision may occur, making it difficult to promptly detect and address sudden health issues in inmates, such as sudden death or illness, which could pose safety hazards.

[0004] Video surveillance systems are an important auxiliary tool for prison supervision, but they are essentially two-dimensional records of scenes within cells and cannot directly perceive the vital signs of detainees. For some potential health risks, such as abnormal heart rate or respiratory arrest, video surveillance cannot provide early warnings. Furthermore, video surveillance also raises privacy concerns to some extent, potentially triggering resistance from detainees and disrupting the order of supervision.

[0005] While some existing smart monitoring devices offer basic sleep monitoring capabilities, they suffer from numerous limitations. The monitoring parameters are relatively limited, typically only recording basic sleep data. For vital signs crucial in prison settings, such as heart rate, respiration, snoring, sleep apnea, and cardiac arrest, the monitoring is insufficiently accurate and comprehensive, failing to provide adequate evidence for supervisory decisions. Furthermore, their accuracy needs improvement, as they are susceptible to external environmental interference, leading to misjudgments or omissions—unacceptable for high-security environments like prisons.

[0006] Furthermore, in prison settings, the integrity and security of monitoring equipment are paramount. However, most current smart monitoring devices lack effective anti-tampering protection mechanisms, making them vulnerable to damage or disassembly by inmates. Once disassembled, the equipment not only malfunctions, preventing timely monitoring of inmates' vital signs, but it could also be used for illegal purposes, such as crafting tools from disassembled parts, thus increasing monitoring risks. Moreover, the difficulty for monitoring personnel to detect illegal disassembly in a timely manner further undermines the reliability of monitoring.

[0007] Prison supervision requires a highly reliable, secure, and comprehensive equipment system. However, existing technologies and methods are significantly inadequate in terms of the comprehensiveness and accuracy of vital sign monitoring, as well as the anti-tampering security of the equipment, making it difficult to meet the needs of real-time, accurate, and secure supervision of detainees in prison settings. Supervisory authorities urgently need intelligent monitoring equipment capable of comprehensively monitoring detainees' vital signs, accurately predicting health risks, and effectively preventing unauthorized dismantling. This would enhance the intelligence level and security capabilities of prison supervision, reduce potential security risks, and ensure the safety and stability of prisons. Summary of the Invention

[0008] To address the significant shortcomings of existing regulatory technologies and methods in terms of the comprehensiveness, accuracy, and anti-disassembly security of vital sign monitoring, the present invention aims to provide an intelligent monitoring device with anti-disassembly functionality. Through multimodal heterogeneous sensor fusion technology, it achieves accurate monitoring of user sleep vital signs and real-time alarm function for unauthorized disassembly.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A smart monitoring device with anti-disassembly functionality includes a device body, a monitoring module, and an anti-disassembly module. The monitoring module, located inside the device body, includes: a fiber optic vital signs sensor for monitoring parameters such as heart rate, respiration, snoring, sleep apnea, body movement, in / out of bed status, and cardiac arrest; a MEMS sensor that works in conjunction with the fiber optic vital signs sensor to improve the accuracy of snoring and sleep apnea monitoring; a voiceprint sensor that, combined with the fiber optic vital signs sensor, enhances the reliability of snoring and sleep apnea monitoring; and a wiring harness that transmits the sensor signals to a main controller. The main controller connects to and controls the sensors, processing and analyzing the collected data.

[0010] The anti-disassembly module includes: a mechanical structure comprising: a mounting plate on which the device body is mounted via at least one magnetic switch structure, with magnetic switch structure mounting holes at each of the four corners of the mounting plate; a magnetic switch structure comprising a base, a locking block, a locking block, a locking pin, and a spring; wherein the base is a cuboid structure, including a locking block mounting hole penetrating the top and bottom surfaces, a locking block mounting portion on the side, and a locking pin mounting hole penetrating the locking block mounting hole and the locking block mounting portion; the locking block is a cylinder with an internal cavity, one side of which has a through hole extending into the interior to form a channel, and a first blocking portion is provided within the channel; the locking block is a triangular structure with a blind hole inside, the opening end of which is aligned with the locking pin mounting hole; the locking pin is located within the locking pin mounting hole, with a raised step in the middle, and the spring is loosely fitted around the outer periphery of the locking pin near the side of the locking block, one end of which abuts against the raised step, and the other end of which abuts against the opening end face of the blind hole of the locking block; in the locked state, the locking pin is inserted into the through hole of the locking block under the action of the spring force and abuts against the first blocking portion within the channel.

[0011] The present invention provides an intelligent monitoring device with anti-disassembly function, the advantages of which are: (1) Employing multimodal heterogeneous sensors, the device comprehensively analyzes vibration and other signals to fully capture the user's vital signs, significantly improving monitoring accuracy compared to traditional single-sensor systems. The device integrates data from multiple sensors and processes it using a weighted fusion algorithm, allocating weights based on the sensitivity and reliability of each sensor to significantly improve the accuracy and reliability of vital sign parameters. Simultaneously, a database of the user's personal sleep characteristics is established, comparing real-time and historical data to achieve personalized health monitoring and timely detection of abnormal changes.

[0012] (2) The device’s anti-disassembly structure cleverly balances the ease of disassembly with the protection against unauthorized disassembly. Key components are secured with snap-locking mechanisms, making them difficult to disassemble without specialized tools, thus effectively preventing unauthorized disassembly. The anti-disassembly structure works in conjunction with the internal sensor monitoring system. In the event of unauthorized disassembly, the sensors detect changes in vibration and light intensity and trigger an alarm, promptly stopping the unauthorized activity and protecting the product’s safety.

[0013] (3) The device integrates MEMS sensors and photoelectric sensors to accurately monitor unauthorized disassembly and promptly issue alarms. The alarm mechanism includes both local and remote levels. Local buzzers and warning lights promptly alert on-site personnel, while the remote wireless communication module sends alarm information to the monitoring center and user terminals, facilitating rapid action. This not only addresses unauthorized disassembly but also deters potential illegal activities, enhancing the device's safety performance. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the intelligent monitoring device with anti-disassembly function of the present invention; Figure 2A flowchart illustrating the normal installation and disassembly of the intelligent monitoring device with anti-disassembly function of the present invention; Figure 3 This is a flowchart illustrating the status monitoring and early warning process of the intelligent monitoring device with anti-disassembly function according to the present invention; Figure 4 This is a schematic diagram of the structure of the intelligent monitoring device with anti-disassembly function of the present invention. Figure 1 ; Figure 5 This is a schematic diagram of the structure of the intelligent monitoring device with anti-disassembly function of the present invention. Figure 2 ; Figure 6 This is a partial schematic diagram of the intelligent monitoring device with anti-disassembly function of the present invention; Figure 7 Explosion-proof structure of the intelligent monitoring device with anti-disassembly function of the present invention Figure 1 ; Figure 8 Explosion-proof structure of the intelligent monitoring device with anti-disassembly function of the present invention Figure 2 ; Figure 9 Explosion-proof structure of the intelligent monitoring device with anti-disassembly function of the present invention Figure 3 ; Figure 10 This is a schematic diagram of the locking block of the intelligent monitoring device with anti-disassembly function of the present invention; Figure 11 This is a schematic diagram of the card block of the intelligent monitoring device with anti-disassembly function of the present invention. Detailed Implementation

[0015] To fully understand this invention, detailed steps and structures will be presented in the following description to illustrate the technical solution of this invention. Preferred embodiments of the invention are described in detail below; however, in addition to these detailed descriptions, the invention may have other embodiments.

[0016] Example 1

[0017] A smart monitoring device with anti-disassembly function includes a device body 1, a monitoring module, and an anti-disassembly module. The monitoring module, located inside the device body, includes: a fiber optic vital signs sensor for monitoring parameters such as heart rate, respiration, snoring, sleep apnea, body movement, in / out of bed status, and cardiac arrest; a MEMS sensor that works in conjunction with the fiber optic vital signs sensor to improve the accuracy of snoring and sleep apnea monitoring; a voiceprint sensor that, combined with the fiber optic vital signs sensor, enhances the reliability of snoring and sleep apnea monitoring; and a wiring harness that transmits the sensor signals to a main controller. The main controller connects to and controls the sensors, processing and analyzing the collected data.

[0018] The anti-disassembly module includes: a mechanical structure comprising: a mounting plate 3, on which the device body 1 is mounted via at least one magnetic switch structure 3, and magnetic switch structure mounting holes are provided at each of the four corners of the mounting plate; a magnetic switch structure 4, comprising a base 5, a locking block 6, a locking block 7, a locking pin 8, and a spring 9; wherein the base 5 is a cuboid structure, including a locking block mounting hole 10 penetrating the top and bottom surfaces, a locking block mounting part 11 provided on the side, and a locking pin mounting hole 12 penetrating the locking block mounting hole and the locking block mounting part; the locking block 7 is a circular block with an internal cavity. The cylinder has a through hole 13 on one side, which extends into the interior to form a channel. A first blocking part 14 is provided in the channel. The locking block 6 has a triangular structure with a blind hole 15 inside. The opening end of the blind hole is aligned with the mounting hole of the locking pin. The locking pin is located in the mounting hole of the locking pin and has a raised step in the middle. A spring is loosely fitted on the outer periphery of the locking pin near the side of the locking block. One end of the spring abuts against the raised step, and the other end of the spring abuts against the opening end face of the blind hole of the locking block. In the locked state, the locking pin is inserted into the through hole of the locking block under the action of the spring force and abuts against the first blocking part in the channel.

[0019] In a specific embodiment, a magnetic material is provided on the side of the locking pin near the locking block. The internal cavity of the locking block is connected to the top surface, and an annular boss is provided on the outer periphery of the top surface, which is fixed to the device body. A connecting plate 2 is also included, disposed between at least two mounting plates, to fix the at least two mounting plates together. A magnetic switch key is also included, comprising a strong magnet. A sensing structure is also included, including a photoelectric sensor, which works in conjunction with a MEMS sensor to realize an unauthorized disassembly alarm function. At least two photoelectric sensors are included: a first photoelectric sensor located within the mounting hole of the locking block and a second photoelectric sensor located on the surface of the connecting plate between the connecting plate and the device body.

[0020] The main controller is equipped with a status monitoring and early warning unit, which includes a signal acquisition and preprocessing module, a feature extraction module, a data fusion and feature-level fusion module, a model training and classification recognition module, and a classification recognition and alarm decision module.

[0021] Signal Acquisition and Preprocessing Module: A high-precision clock chip is configured on the main control board of the intelligent monitoring device to generate a clock signal as a synchronous sampling reference signal for the fiber optic vital signs sensor, MEMS sensor, and voiceprint sensor. The sampling control terminals of each sensor are connected to this clock signal source to achieve synchronous sampling. A low-pass filter is used for the heart rate and respiration signals acquired by the fiber optic vital signs sensor, with the cutoff frequency set at around 10Hz to filter out high-frequency noise. A band-pass filter is used for the snoring and apnea signals acquired by the MEMS sensor, with the passband frequency range set from 20Hz to 1000Hz to filter out low-frequency interference and high-frequency noise. An audio signal adapted to audio signal processing is used for the audio signal acquired by the voiceprint sensor to retain components that match the frequency characteristics of snoring and filter out background environmental noise and other irrelevant sound signals.

[0022] Feature extraction module: Parallel processing of three types of sensors; the fiber optic vital signs sensor outputs a mixed signal, which, after bandpass filtering, detrending, and normalization, is decomposed using wavelet decomposition to obtain four sub-bands: respiration, heart rate, body movement, and bed exit; independent component analysis is then performed on the respiration and heart rate sub-bands to suppress crosstalk; subsequently, adaptive peak detection is used in the heart rate sub-band to obtain real-time heart rate, heart rate variability, and amplitude; peak and trough detection is used in the respiration sub-band to obtain respiratory frequency, amplitude, and smoothness; the body movement sub-band uses sliding window RMS to detect turning events and their intensity and duration; the bed exit sub-band determines bed exit based on amplitude drop and records the switching time; the MEMS sensor extracts the dominant frequency, amplitude, phase, and duration of snoring vibrations and monitors the vibration energy attenuation during apnea; the voiceprint sensor obtains the snoring tone through spectral analysis, extracts the timbre using Mel-frequency cepstral coefficients, calculates the sound pressure level to obtain loudness, and statistically analyzes the snoring interval and frequency; all features are output to the fusion module after being aligned by timestamps.

[0023] The data fusion and feature-level fusion module determines weight coefficients based on the dependence and reliability of each monitoring parameter on sensor features. It then multiplies the corresponding feature values ​​of each sensor by the weight coefficients and sums them to obtain comprehensive snoring feature parameters. The module sequentially combines various feature parameters extracted from fiber optic vital signs sensors, MEMS sensors, and voiceprint sensors into feature vectors and normalizes each feature parameter within the vectors. The model training and classification module organizes the collected fused feature vector data samples and labels them with physiological states and behavioral categories. It performs Z-score standardization on each feature in the feature vectors and divides the dataset into training, validation, and test sets with a ratio of 70% for training, 15% for validation, and 15% for test. Based on the feature vector dimension, it determines the shape of the LSTM network input layer, the number of neurons in the LSTM hidden layer, automatically learns the time dependencies and long-term memory features in the time series data, adds a Dropout layer after the LSTM hidden layer to prevent overfitting, determines the number of neurons in the output layer based on the number of categories in the classification task, and uses the softmax activation function. It defines the cross-entropy loss function and selects Adam. The optimizer sets the appropriate parameters, inputs the training set data into the LSTM network in batches for forward and backward propagation training, monitors performance metrics, and uses k-fold cross-validation to evaluate model performance and adjust hyperparameters.

[0024] Classification and Alarm Decision Module: When the intelligent monitoring device is in operation, it will fuse feature vectors in real time and input them into the trained classification model to classify and identify the current user's physiological state and behavior category. When the identification result indicates the presence of abnormalities such as snoring, breathing apnea, or cardiac arrest, the alarm mechanism will be triggered, activating the pillow's local sound and light alarm device, and sending the alarm information to the remote monitoring center via the wireless communication module.

[0025] Example 2

[0026] Correspondingly, the present invention also provides a method for using an intelligent monitoring device with anti-disassembly function, including: S1, normal installation and disassembly; S11, assemble at least two mounting plates via connecting plates and install them to the headboard position of the bed board using fasteners; S12, fix the annular boss of the locking block to the four corners of the device body; S13, the magnetic switch structure is installed at the four corners of each mounting plate, and magnetic switch structure mounting holes are provided; at this time, the locking pin is inserted into the locking block mounting hole under the action of spring force; S14, when the magnetic switch key is placed near any card block, the magnetic material on the side of the locking pin near the card block is attracted by the magnetic force, and the locking pin moves away from the mounting hole of the locking block. S15, Install the locking blocks fixed at the four corners of the device body into the corresponding locking block mounting holes. After installation, the magnetic switch key moves away from the card block, and the locking pin is inserted into the through hole of the locking block under the action of the spring force and abuts against the first blocking part in the channel. S16. Using the above method, install the locking blocks at the four corners of the mounting plate in sequence; S17, during normal replacement and removal, place the magnetic switch key near any of the card blocks. Due to the magnetic force, the magnetic material on the side of the locking pin closest to the card block is attracted, and the locking pin moves away from the through hole of the locking block. S18, disengage the locking block corresponding to the locking block mounting hole, and remove the locking blocks at the four corners of the mounting plate in sequence using the above method.

[0027] S2, unauthorized dismantling of surveillance equipment; S21, Vibration Characteristic Data Analysis: Collect vibration velocity and acceleration data from normal disassembly experiments. Use data analysis software to calculate the statistical characteristic values ​​of the vibration data, including the mean, standard deviation, maximum, and minimum values, to determine the typical range of vibration characteristics during normal disassembly. Simultaneously, record the frequency components and duration of the vibration to provide a basis for setting the vibration range.

[0028] S22, Analysis of Light Change Patterns from Photoelectric Sensors Extract light intensity change data detected by photoelectric sensors during normal disassembly, and analyze the sequence, amplitude, and duration of these changes. Calculate the statistical characteristic values ​​of the light intensity changes to determine the typical range of light change patterns during normal disassembly.

[0029] S23, Set threshold values ​​for normal disassembly vibration range: Based on the vibration characteristic data analysis results, set threshold values ​​for the vibration rate and acceleration ranges during normal disassembly. Specifically, this includes setting the peak acceleration range, root mean square acceleration range, and specific frequency component range. The set threshold values ​​should cover the vibration characteristics generated by most normal disassembly operations.

[0030] S24, Set the threshold for normal disassembly light change mode: Based on the analysis results of the light change mode of the photoelectric sensor, set the threshold for the light intensity change of the photoelectric sensor during normal disassembly. Specifically, this includes setting the amplitude range, duration range, and sequence of light intensity changes.

[0031] S25, Real-time data monitoring and preliminary judgment: The main controller continuously receives monitoring data from the MEMS sensor and photoelectric sensor in real time. It performs a preliminary assessment of the vibration data, checking if it falls within the set threshold range for normal disassembly vibration. Simultaneously, it performs a preliminary assessment of the light change data from the photoelectric sensor, checking if it conforms to the set threshold range for normal disassembly light change patterns.

[0032] S26, Comprehensive Judgment Logic Operation: Boolean logic operators are used to design a comprehensive judgment logic that integrates the preliminary judgment results from vibration data and photoelectric sensor signals. Only when the vibration data is within the set range and the photoelectric sensor signal conforms to the set mode is the comprehensive judgment result considered normal disassembly. Otherwise, it is judged as illegal disassembly.

[0033] S27, Result Processing: Based on the comprehensive judgment logic calculation result, if the disassembly is determined to be normal, the normal disassembly process will proceed without triggering an alarm. If the disassembly is determined to be illegal, the alarm device will be triggered immediately, and the alarm information will be sent to the remote monitoring center and the user terminal.

[0034] Specifically, in the steps of the S26 comprehensive judgment logic operation, the comprehensive judgment of vibration data and photoelectric sensor signals includes: S261 defines two flag variables to record whether the vibration data meets the set range and whether the light change of the photoelectric sensor meets the set mode, respectively. The vibration data flag is used to indicate whether the vibration is normal, and the photoelectric sensor flag is used to indicate whether the light change is normal.

[0035] S262, the judgment conditions are set as follows: Vibration data judgment: When the peak acceleration detected by the MEMS sensor is within the range of [X1, X2] m / s², the root mean square acceleration is within the range of [Y1, Y2] m / s², and the frequency range of a specific frequency component in the vibration is [Z1, Z2] Hz, the vibration data is judged to be valid; otherwise, it is invalid.

[0036] Photoelectric sensor signal judgment: When the light intensity change detected by the photoelectric sensor is within the range of [A1, A2] lux, the change duration is [B1, B2] seconds, and the change sequence conforms to the preset sequence, the photoelectric sensor is judged to be in a valid state; otherwise, it is in an invalid state. During normal disassembly, the first photoelectric sensor located in the locking block mounting hole and the second photoelectric sensor on the connecting plate surface between the connecting plate and the device body will both be activated, with the first photoelectric sensor activating slightly later than the second photoelectric sensor, conforming to the preset sequence. When the device body is illegally disassembled by brute force to separate it from the connecting plate, the first photoelectric sensor is activated, but because the locking block is engaged in the locking block mounting hole, the second photoelectric sensor is not activated, which does not conform to the preset sequence.

[0037] S263, the comprehensive judgment process is as follows: The system determines the current disassembly operation as normal and does not trigger an alarm mechanism only when both the vibration data flag and the photoelectric sensor flag are valid. If at least one of the vibration data flag or the photoelectric sensor flag is invalid, the system determines the current disassembly operation as illegal and immediately triggers an alarm mechanism.

[0038] S3, Status Monitoring and Early Warning; S31, Signal Acquisition and Preprocessing S311, synchronous sampling Set up a unified sampling clock signal source: Configure a high-precision clock chip on the main control board of the intelligent monitoring device. This chip generates a stable clock signal, which serves as the synchronous sampling reference signal for the fiber optic vital signs sensor, MEMS sensor, and acoustic signature sensor.

[0039] Connect and synchronize the sensors: Connect the sampling control terminals of the fiber optic vital signs sensor, MEMS sensor, and acoustic signature sensor to a unified sampling clock signal source. On the trigger edge of each clock signal, such as the rising edge, the three sensors synchronously initiate a data acquisition operation, ensuring that the acquired data is strictly aligned on the time axis, providing temporal consistency for subsequent fusion analysis.

[0040] S312, Filtering and Noise Reduction Fiber Optic Vital Signs Sensor Filtering: For heart rate and respiratory signals acquired by fiber optic vital signs sensors, these signals are typically concentrated in a low-frequency range, while high-frequency noise may originate from external electromagnetic interference and other factors. A low-pass filter is used for filtering, setting an appropriate cutoff frequency, for example, around 10Hz based on the normal range of human heart rate and respiratory rate. This filters out high-frequency noise above the cutoff frequency, retaining the valid heart rate and respiratory signal waveforms.

[0041] MEMS sensor filtering: For snoring and apnea signals acquired by MEMS sensors, the vibration frequencies corresponding to snoring and the characteristic frequencies of apnea are usually located in a specific mid-frequency range. A bandpass filter is used to filter the MEMS sensor signals, setting the passband frequency range to 20Hz to 1000Hz. Only effective vibration signals related to snoring and apnea within this range are retained, while low-frequency interference below the lower limit frequency, such as the slight shaking of the pillow, and high-frequency noise above the upper limit frequency, such as high-frequency electromagnetic interference from surrounding electronic devices, are filtered out.

[0042] Voiceprint sensor filtering: The audio signal collected by the voiceprint sensor contains rich frequency components, but the features related to snoring are mainly concentrated in a specific audio range. Filtering the voiceprint sensor signal using suitable audio signal processing algorithms, such as digital filters, preserves audio components that match the frequency characteristics of snoring, filters out background noise and other irrelevant high- or low-frequency sound signals, improves the quality of the voiceprint signal, and enhances the accuracy of subsequent snoring feature extraction.

[0043] S32, Feature Extraction S321, Feature Extraction from Fiber Optic Vital Sensor The fiber optic vital signs sensor outputs a continuous analog signal, which simultaneously superimposes the heart rate pulse, respiratory rhythm, body movement impact, and microbending loss changes caused by the subject's in-bed / out-of-bed state; the above four types of physiological / behavioral features are extracted from the single signal using a frequency domain-time domain joint processing flow. S3211, Preprocessing The single signal is sequentially subjected to bandpass filtering, sliding baseline correction, and amplitude normalization to eliminate power frequency drift and individual amplitude differences, thereby obtaining a standardized waveform.

[0044] S3212, Multi-resolution decomposition The standardized waveform is input into the wavelet decomposition unit, and after multi-level decomposition, the respiratory sub-band, heart rate sub-band, body motion sub-band, and in-bed / out-of-bed sub-band are obtained; then, independent component analysis is performed on the respiratory sub-band and heart rate sub-band to suppress inter-band crosstalk. S3213, Heart Rate Feature Extraction Adaptive peak detection is performed on the heart rate sub-bands, and the real-time heart rate value is calculated based on the time interval between adjacent peaks; the dispersion of continuous peak-to-peak intervals is used to characterize heart rate variability; and the amplitude changes of peak and trough values ​​are recorded for subsequent heart rate amplitude analysis.

[0045] S3214, Respiratory Feature Extraction Peak-valley detection is performed on the respiratory subband, and the respiratory rate is calculated based on the time interval between adjacent peaks; the respiratory amplitude is characterized by the peak-valley difference; and the waveform similarity between adjacent respiratory cycles is calculated to evaluate the smoothness of the respiratory waveform.

[0046] S3215, Body Motion Feature Extraction Submodule Calculate the root mean square value of the sliding window in the motion sub-band; when the root mean square value exceeds the statistical threshold of the resting segment, it is marked as a motion event; record the maximum intensity and duration of the event.

[0047] S3216, In / Out-of-bed Feature Extraction Submodule The average amplitude is calculated for the in-bed / out-of-bed sub-band according to a preset time period. If the amplitude decrease and duration both meet the preset conditions, the state is determined to be out of bed; if the amplitude recovers to above the baseline ratio, the state is determined to be in bed; the state switching timestamp is recorded.

[0048] S3217, Feature Integration Submodule The above heart rate, respiration, body movement, and in-bed / out-of-bed features are aligned by timestamps to form a continuous feature vector, which is then output to the data fusion and feature-level fusion module for subsequent classification, recognition, and alarm decision-making. S322, MEMS sensor feature extraction Snoring vibration feature extraction: The micro-vibration components related to snoring are analyzed from acceleration and angular velocity signals acquired by MEMS sensors. By calculating the spectrum of the vibration signals, the dominant frequency range of snoring vibrations is identified, and characteristic parameters such as vibration amplitude and phase within this frequency range are extracted. Simultaneously, the duration of snoring vibrations is statistically analyzed to distinguish between continuous and intermittent snoring events, providing a basis for assessing the severity of snoring.

[0049] Vibration feature extraction during apnea: During apnea, the body's breathing stops, and the corresponding minute vibrations in the chest and abdomen disappear or significantly weaken. Monitoring the vibration energy changes of MEMS sensor signals during suspected apnea periods, when the vibration energy falls below a set apnea threshold and persists for a certain apnea detection duration, features such as the start and end times of that period and the degree of vibration energy attenuation are extracted for the identification and analysis of apnea events.

[0050] S323, Voiceprint Sensor Feature Extraction Snoring tone feature extraction: Spectral analysis is performed on the audio signal collected by the voiceprint sensor to calculate the fundamental frequency of the audio signal and determine the pitch of the snoring sound. Different types of snoring, such as nasal obstruction type and pharyngeal relaxation type, often have different tone features. Extracting tone parameters helps to classify snoring types.

[0051] Snoring timbre feature extraction: Feature extraction methods from audio signal processing are used to extract the timbre features of snoring. Timbre features reflect the sound quality characteristics of snoring. Even snoring sounds with similar pitches may have different timbres due to factors such as the vocalization location and airflow channel. Timbre features can be used to distinguish snoring sounds caused by different reasons. Specifically, Mel frequency cepstral coefficients are used for feature extraction: First, the audio signal collected by the voiceprint sensor is preprocessed, including windowing and framing; a fast Fourier transform is performed on each frame to obtain its power spectrum; the power spectrum is filtered using a Mel filter bank to obtain the spectral features in the Mel frequency domain; the logarithm of the filtered spectral features is taken, and a discrete cosine transform is performed to extract its cepstral coefficients; several low-frequency coefficients in the cepstral coefficients are selected as the main features, and the first and second differences of these coefficients are calculated to characterize the dynamic changes of the audio signal; the above features are combined to form a complete timbre feature vector, and then normalized to eliminate amplitude differences under different environments, finally obtaining snoring timbre features that can be used for subsequent analysis.

[0052] Snoring loudness feature extraction: The sound pressure level of the audio signal from the voiceprint sensor is calculated to obtain the loudness value of the snoring. Loudness directly reflects the intensity of snoring. Excessively high snoring loudness may affect sleep quality or indicate potential breathing problems. Extracting loudness features can be used to assess the severity of snoring.

[0053] Snoring interval time feature extraction: This involves monitoring changes in the time interval between snoring events, calculating the duration of the interval between adjacent snoring events, and counting the number of consecutive snoring occurrences and the frequency of intermittent snoring. Snoring interval time features help analyze the regularity of breathing patterns and are of great significance for the early detection of abnormal breathing conditions such as sleep apnea.

[0054] S33, Data Fusion and Feature-Level Fusion S331, Weighted Fusion Algorithm Determine the feature weights for each sensor: Based on the dependence and reliability of different monitoring parameters on the features of each sensor, assign weight coefficients to the features extracted by the fiber optic vital signs sensor, MEMS sensor, and voiceprint sensor. For example, in a snoring monitoring scenario, through analysis of a large amount of experimental data, determine that the acoustic features of the voiceprint sensor, such as pitch, timbre, and loudness, have a weight of α; the vibration features of the MEMS sensor, such as snoring vibration frequency and amplitude, have a weight of β; and the body movement features of the fiber optic vital signs sensor, such as body movement amplitude and frequency, have a weight of γ.

[0055] S332, Weighted Fusion Calculation: The corresponding feature values ​​extracted by each sensor are multiplied by their corresponding weight coefficients, and then the weighted feature values ​​are summed to obtain a comprehensive snoring feature parameter.

[0056] The comprehensive snoring characteristic value = (acoustic characteristic value of voiceprint sensor × α) + (vibration characteristic value of MEMS sensor × β) + (body motion characteristic value of fiber optic sensor × γ).

[0057] These comprehensive feature parameters more fully and accurately reflect the characteristics of snoring events, providing a more reliable basis for subsequent classification and identification.

[0058] S333, Feature-level fusion Feature vector construction: Features extracted from fiber optic vital signs sensors, such as heart rate value and heart rate variability; respiratory features, such as respiratory rate and apnea duration; body movement features, such as number of body movements and amplitude; in-bed / out-of-bed features, such as in-bed status and out-of-bed time; snoring vibration features extracted from MEMS sensors, such as vibration frequency and amplitude; apnea vibration features, such as vibration energy attenuation; and snoring tone, timbre, loudness, and interval time extracted from voiceprint sensors, are combined in a specific order to form a feature vector. The feature vector is represented as [heart rate value, respiratory rate, in-bed / out-of-bed status, snoring vibration frequency, snoring tone, ...], containing multi-dimensional feature information.

[0059] Feature normalization: Since the dimensions and numerical ranges of different feature parameters may vary significantly, in order to eliminate the influence of dimensions and improve the stability of data processing, the feature parameters in the feature vector are normalized. A linear normalization method is used to map each feature value to the interval [0, 1] or [-1, 1], making the features numerically comparable and facilitating unified processing in subsequent model training and classification algorithms.

[0060] S34, Model Training and Classification S341, Data Preprocessing and Preparation Data processing: The collected data samples, including fused feature vectors from fiber optic vital signs sensors, MEMS sensors, and voiceprint sensors, were arranged and organized in chronological order. Each feature vector has a corresponding timestamp and labeled physiological state and behavioral category, such as "normal sleep," "mild snoring," and "apnea."

[0061] Data standardization: The features in the feature vector are standardized. Given the significant differences in the numerical ranges of different features, the Z-score standardization method is used. The mean and standard deviation of each feature are calculated, and each feature value is subtracted from the mean and divided by the standard deviation to ensure that the standardized feature values ​​have zero mean and unit variance.

[0062] Dataset partitioning: The cleaned and standardized dataset is divided into training, validation, and test sets. The partition ratio is 70% for training, 15% for validation, and 15% for test.

[0063] S342, LSTM Network Construction Input Layer: The dimension of the input layer is determined by the dimension of the feature vector. Each input sample is a time series, and its sequence length is determined based on the temporal correlation of the sleep monitoring data. The shape of the input layer is set to (t, n), where t is the sequence length and n is the number of features. The number of features n includes multimodal features such as heart rate, respiratory rate, body movement intensity, in-bed / out-of-bed status, and snoring characteristics. These features are generated through preprocessing and feature extraction modules to ensure that the input data comprehensively reflects the physiological and behavioral changes during sleep. The length t of the time series is sufficient to cover multiple respiratory cycles and heart rate cycles, thereby capturing the periodic changes during sleep.

[0064] Hidden Layers: The number of neurons in the LSTM hidden layers is determined by a combination of factors, including data complexity, time series length, and model computational resources. The LSTM hidden layers, through the synergistic effect of their input gate, forget gate, and output gate, handle short-term and long-term dependencies in time series data.

[0065] The input gate controls the inflow of new information. It receives the input feature vector from the current time step and the hidden state from the previous time step, generating a control signal through an activation function. This control signal determines how much new information from the current time step is written into the cell state. The input gate's weight matrix focuses on sensitivity to changes in heart rate and respiratory rate, enabling the model to promptly capture subtle changes in physiological signals and write them into the cell state. The bias term is adjusted to ensure a smooth inflow of new information into the cell state during normal sleep, avoiding over-suppression or over-activation.

[0066] The forget gate is responsible for determining which information in the cell state needs to be forgotten. It receives the input feature vector from the current time step and the hidden state from the previous time step, and generates a control signal for the forget gate through an activation function. This control signal determines which information in the cell state needs to be forgotten. The weight matrix of the forget gate focuses on sensitivity to abnormal events such as apnea and cardiac arrest, enabling the model to promptly forget previous normal state information when these abnormal events are detected, thus better focusing on the current abnormal state. The bias term is adjusted to ensure that important information in the cell state is not prematurely forgotten during normal sleep, while irrelevant information is quickly forgotten when abnormal events occur.

[0067] The output gate determines how much of the hidden state at the current time step is output. It receives the input feature vector from the current time step and the hidden state from the previous time step, and generates a control signal for the output gate through an activation function. This control signal determines how much of the hidden state at the current time step is output. The weight matrix of the output gate focuses on the sensitivity to body movement intensity and in / out-of-bed status, ensuring that the model outputs the current hidden state promptly when it detects body movement or out-of-bed events, allowing the subsequent classification and recognition module to react quickly. The bias term is adjusted to ensure that the output of the hidden state remains stable during normal sleep, avoiding misjudgments caused by noise or minor fluctuations.

[0068] Dropout Layer: A Dropout layer is added after the LSTM hidden layers to prevent overfitting. During training, the Dropout layer randomly discards some neurons' outputs; this mechanism is tuned to adapt to the periodicity and suddenness of physiological signals. The dropout ratio is dynamically adjusted based on the model's performance on a validation set containing different sleep stages and individual differences. When processing sleep monitoring data, the Dropout layer dynamically selects neurons to drop based on the characteristics of the input signals to ensure stable performance across different sleep stages. In light sleep, heart rate and respiratory rate fluctuate more frequently, and the Dropout layer adjusts its dropout strategy based on these characteristics; while in deep sleep, physiological signals are more stable, and the dropout strategy is adjusted accordingly. When handling sudden events such as apnea and cardiac arrest, the Dropout layer dynamically adjusts its dropout strategy based on the event's characteristics to ensure the model responds quickly and outputs the correct signal. The Dropout layer effectively enhances the model's robustness in practical applications, adapting to different sleep stages and individual physiological differences.

[0069] Output Layer: The number of neurons in the output layer is determined by the number of categories in the classification task. The output layer uses the softmax activation function, mapping the neuron outputs to the interval [0, 1], so that the sum of all output values ​​is 1, representing the probability distribution of the sample belonging to each category. In sleep monitoring, the output categories include normal sleep, mild snoring, sleep apnea, cardiac arrest, and other abnormal conditions. The softmax activation function converts the neuron outputs into a probability distribution, allowing the model to output the confidence score for each category, facilitating subsequent classification, identification, and alarm decisions.

[0070] S343, Model Training Define the loss function and optimizer: For multi-class classification problems, the cross-entropy loss function is used to measure the difference between the model's predicted probability distribution and the true label distribution. The Adam optimizer is selected, with parameters set to a learning rate of 0.001, β1 = 0.9, and β2 = 0.999.

[0071] Training Process: The training set data is input into the constructed LSTM network for training. In each training epoch, the training set data is input in batches, with each batch containing 32 or 64 samples. During forward propagation, the network calculates the probability distribution of each sample belonging to each class based on the input feature sequence and calculates the error between the predicted and true labels using a loss function. During backpropagation, the error signal is propagated back layer by layer, updating the network weights and bias parameters. Performance metrics such as the loss values ​​of the training and validation sets and the classification accuracy are monitored in real time during training.

[0072] Cross-validation and hyperparameter tuning: k-fold cross-validation is used to evaluate model performance. The training set is divided into k subsets, with k set to 5. One subset is used for validation each time, while the rest are used for training. This process is repeated k times, and the average performance metric is taken. Based on the cross-validation results, the LSTM network hyperparameters are adjusted, including the number of hidden layer neurons, sequence length, Dropout ratio, and learning rate, to optimize model performance and avoid overfitting or underfitting. The loss and accuracy curves of the training and validation sets are observed. If the training loss decreases while the validation loss increases, training is stopped early or the hyperparameters are further adjusted.

[0073] S35, Classification and Alarm Decision Real-time feature vector input: During the actual operation of the intelligent monitoring device, the real-time fused feature vector obtained through signal acquisition and preprocessing, feature extraction and data fusion steps is continuously input into the trained classification model.

[0074] Model classification and recognition: The classification model quickly classifies and recognizes the input real-time feature vector based on the learned feature patterns and classification boundaries, and determines which category the current user's physiological state and behavior belongs to, such as normal sleep, abnormal snoring, sleep apnea, etc.

[0075] Alarm Decision: When the classification and identification results indicate the presence of abnormalities such as snoring, breathing apnea, or cardiac arrest, the corresponding alarm mechanism is immediately triggered. On one hand, the pillow's local audible and visual alarm device is activated, emitting an alarm sound and flashing lights to alert the user or nearby personnel. On the other hand, the alarm information is promptly transmitted to a remote monitoring center via wireless communication modules such as Wi-Fi, Bluetooth, and 4G / 5G, ensuring rapid rescue measures can be taken in emergencies to protect the user's health and safety.

[0076] The preferred embodiments of the present invention have been described above. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An intelligent monitoring device with anti-disassembly function, characterized in that, Includes the device body, monitoring module, and anti-disassembly module; The monitoring module is located inside the device body and includes: Fiber optic vital signs sensors are used to monitor user parameters such as heart rate, respiration, snoring, sleep apnea, body movement, in / out of bed, and cardiac arrest. MEMS sensors work in conjunction with fiber optic vital signs sensors; A voiceprint sensor works in conjunction with a fiber optic vital signs sensor. The wiring harness transmits the aforementioned sensor signals to the main controller; The main controller connects to and controls the aforementioned sensors, and processes and analyzes the collected data. The anti-disassembly module includes: Mechanical structures, including: Mounting plate, on which the device body is mounted by at least one magnetic switch structure, and mounting holes for magnetic switch structures are provided at all four corners of the mounting plate; The magnetic switch structure includes a base, a locking block, a locking pin, and a spring. The base has a cuboid structure, including a locking block mounting hole that runs through the top and bottom surfaces, a latching block mounting part on the side, and a locking pin mounting hole that runs through the locking block mounting hole and the latching block mounting part. The locking block is a cylinder with an internal cavity. A through hole is opened on one side of the cylinder, and the through hole extends into the interior to form a channel. A first blocking part is provided in the channel. The internal cavity of the locking block is connected to the top surface. An annular boss is provided on the outer periphery of the top surface. The annular boss is fixed to the device body. The card block has a triangular structure and a blind hole inside, with the opening end of the blind hole aligned with the mounting hole of the locking pin. The locking pin is located in the locking pin mounting hole, with a raised step in the middle. The spring is loosely fitted on the outer periphery of the locking pin near the block. One end of the spring abuts against the raised step, and the other end of the spring abuts against the blind hole opening end face of the block. In the locked state, the locking pin is inserted into the through hole of the locking block under the action of the spring force and abuts against the first blocking part in the channel.

2. The intelligent monitoring device with anti-disassembly function as described in claim 1, characterized in that: The main controller is equipped with a status monitoring and early warning unit, which includes a signal acquisition and preprocessing module, a feature extraction module, a data fusion and feature-level fusion module, a model training and classification recognition module, and a classification recognition and alarm decision module. Signal acquisition and preprocessing module: A high-precision clock chip is configured on the main control board of the intelligent monitoring device to generate a clock signal as a synchronous sampling reference signal for the fiber optic vital signs sensor, MEMS sensor and acoustic signature sensor. The sampling control terminals of each sensor are connected to this clock signal source to achieve synchronous sampling. Low-pass filters are used for the heart rate and respiratory signals acquired by the fiber optic vital signs sensor; band-pass filters are used for the snoring and apnea signals acquired by the MEMS sensor. The audio signal collected by the voiceprint sensor is processed using a filtering algorithm adapted to audio signal processing, retaining the components that match the frequency characteristics of snoring, and filtering out background noise and other irrelevant sound signals. Feature extraction module: processes three types of sensors in parallel; the fiber optic vital signs sensor outputs a mixed signal, which is then bandpass filtered, detrended, and normalized, and wavelet decomposition is used to obtain four sub-bands: respiration, heart rate, body movement, and bed exit. Independent component analysis was performed on the respiration and heart rate subbands to suppress crosstalk; subsequently, adaptive peak detection was performed in the heart rate subband to obtain real-time heart rate, heart rate variability and amplitude. The respiratory subband detects the peaks and troughs of breathing to obtain respiratory rate, amplitude, and smoothness; the body movement subband uses a sliding window RMS to detect turning events and their intensity and duration; the bed-off subband determines bed-off based on amplitude drop and records the switching time; the MEMS sensor extracts the dominant frequency, amplitude, phase, and duration of snoring vibrations and monitors the vibration energy attenuation during apnea. The voiceprint sensor obtains the snoring tone through spectrum analysis, extracts the timbre using Mel-frequency cepstral coefficients, calculates the loudness of the sound pressure level, and counts the snoring interval and frequency of occurrence. All features are output to the fusion process after being aligned by timestamps; Data fusion and feature-level fusion module: The weighting coefficients are determined based on the dependence and reliability of each monitoring parameter on the sensor features. The feature values ​​of each sensor are multiplied by the weighting coefficients and then summed to obtain the comprehensive snoring feature parameters. The various feature parameters extracted by the fiber optic vital signs sensor, MEMS sensor and acoustic fingerprint sensor are combined into a feature vector in sequence, and the feature parameters in the feature vector are normalized. Model training and classification module: Organize the collected fusion feature vector data samples and label them with physiological state and behavior category labels. Perform Z-score standardization on each feature in the feature vector and divide the dataset into training set, validation set and test set. The shape of the LSTM network input layer is determined based on the feature vector dimension, and the number of neurons in the LSTM hidden layer is determined accordingly. The system automatically learns the temporal dependencies and long-term memory features in time-series data. A Dropout layer is added after the LSTM hidden layer to prevent overfitting. The number of neurons in the output layer is determined based on the number of categories in the classification task, and a softmax activation function is used. A cross-entropy loss function is defined, and the Adam optimizer is selected. Corresponding parameters are set, and the training set data is input into the LSTM network in batches for forward and backward propagation training. Performance metrics are monitored, and k-fold cross-validation is used to evaluate model performance and adjust hyperparameters. Classification and Alarm Decision Module: When the intelligent monitoring device is in operation, it will fuse feature vectors in real time and input them into the trained classification model to classify and identify the current user's physiological state and behavior category. When the identification result indicates the presence of abnormalities such as snoring, breathing apnea, or cardiac arrest, the alarm mechanism will be triggered, activating the pillow's local sound and light alarm device, and sending the alarm information to the remote monitoring center via the wireless communication module.

3. The intelligent monitoring device with anti-disassembly function as described in claim 1, characterized in that: The anti-disassembly module also includes a sensing structure, which includes a photoelectric sensor that works in conjunction with a MEMS sensor to enable an alarm function for unauthorized disassembly. The photoelectric sensor includes at least two sensors: a first photoelectric sensor located in the mounting hole of the locking block and a second photoelectric sensor located on the surface of the connecting plate between the connecting plate and the device body.

4. The intelligent monitoring device with anti-disassembly function as described in claim 1, characterized in that: It also includes a magnetic switch key, which includes a strong magnet; correspondingly, the locking pin is provided with magnetic material on the side near the card block.

5. The intelligent monitoring device with anti-disassembly function as described in claim 1, characterized in that: It also includes a connecting plate, which is disposed between at least two mounting plates to securely connect the at least two mounting plates.

6. A method of using an intelligent monitoring device with anti-disassembly function as described in any one of claims 1-5, characterized in that, Includes the following steps: S1, normal installation and disassembly; S11, assemble at least two mounting plates via connecting plates and install them to the headboard position of the bed board using fasteners; S12, fix the annular boss of the locking block to the four corners of the device body; S13, magnetic switch structure mounting holes are provided at all four corners of the mounting plate; at this time, the locking pin is inserted into the locking block mounting hole under the action of spring force; S14, when the magnetic switch key is placed near any card block, the magnetic material on the side of the locking pin near the card block is attracted by the magnetic force, and the locking pin moves away from the mounting hole of the locking block. S15, Install the locking blocks fixed at the four corners of the device body into the corresponding locking block mounting holes. After installation, the magnetic switch key moves away from the card block, and the locking pin is inserted into the through hole of the locking block under the action of the spring force and abuts against the first blocking part in the channel. S16. Using the above method, install the locking blocks at the four corners of the mounting plate in sequence; S17, during normal replacement and removal, place the magnetic switch key near any of the card blocks. Due to the magnetic force, the magnetic material on the side of the locking pin closest to the card block is attracted, and the locking pin moves away from the through hole of the locking block. S18, disengage the locking block corresponding to the locking block mounting hole, and remove the locking blocks at the four corners of the mounting plate in sequence using the above method.

7. The method of using the intelligent monitoring device with anti-disassembly function as described in claim 6, characterized in that, It also includes the following steps: S2, unauthorized removal of surveillance equipment; S21, Vibration characteristic data analysis: Collect vibration rate and acceleration data during normal disassembly experiments, and use data analysis software to calculate the statistical characteristic values ​​of the vibration data, including the mean, standard deviation, maximum and minimum values, to determine the typical range of vibration characteristics during normal disassembly; at the same time, record the frequency components and duration information of the vibration to provide a basis for setting the vibration range; S22, Analysis of Light Change Patterns from Photoelectric Sensors Extract light intensity change data detected by photoelectric sensors during normal disassembly, and analyze the sequence, amplitude, and duration of light intensity changes; Calculate the statistical characteristic values ​​of light intensity changes to determine the typical range of light change patterns during normal disassembly; S23, Set the vibration range threshold for normal disassembly: Based on the vibration characteristic data analysis results, set the vibration rate and acceleration range thresholds for normal disassembly; specifically, this includes setting the peak acceleration range, root mean square acceleration range, and specific frequency component range; the set thresholds should cover the vibration characteristics generated by most normal disassembly operations. S24, Set the threshold for normal disassembly light change mode: Based on the analysis results of the light change mode of the photoelectric sensor, set the threshold for the light intensity change of the photoelectric sensor during normal disassembly; specifically, this includes setting the amplitude range, duration range, and change sequence of the light intensity change. S25, Real-time data monitoring and preliminary judgment: The main controller continuously receives monitoring data from MEMS sensors and photoelectric sensors in real time; it makes a preliminary judgment on the vibration data and checks whether it is within the set threshold of normal disassembly vibration range; at the same time, it makes a preliminary judgment on the light change data of the photoelectric sensor and checks whether it meets the set threshold of normal disassembly light change mode. S26, Comprehensive Judgment Logic Operation: The comprehensive judgment logic is designed using Boolean logic operators to integrate the preliminary judgment results of vibration data and photoelectric sensor signals; only when the vibration data is within the set range and the photoelectric sensor signal meets the set mode, the comprehensive judgment result is normal disassembly; Otherwise, it will be considered illegal dismantling; S27, Judgment Result Processing: Based on the comprehensive judgment logic operation result, if it is determined to be normal disassembly, the normal disassembly process will be entered and no alarm will be triggered; if it is determined to be illegal disassembly, the alarm device will be triggered immediately, and the alarm information will be sent to the remote monitoring center and the user terminal.

8. The method of using the intelligent monitoring device with anti-disassembly function as described in claim 7, characterized in that, It also includes the following steps: In the steps of the S26 comprehensive judgment logic operation, the comprehensive judgment of vibration data and photoelectric sensor signals includes: S261 defines two flag variables to record whether the vibration data meets the set range and whether the light change of the photoelectric sensor meets the set mode, respectively; the vibration data flag is used to indicate whether the vibration is normal and the photoelectric sensor flag is used to indicate whether the light change is normal. S262, the judgment conditions are set as follows: Vibration data judgment: When the peak acceleration detected by the MEMS sensor is within the range of [X1, X2] m / s², the root mean square acceleration is within the range of [Y1, Y2] m / s², and the frequency range of a specific frequency component in the vibration is [Z1, Z2] Hz, the vibration data is judged as valid; otherwise, it is invalid. Photoelectric sensor signal judgment: When the light intensity change detected by the photoelectric sensor is within the range of [A1, A2] lux, the change duration is [B1, B2] seconds, and the change sequence conforms to the preset sequence, the photoelectric sensor is judged to be in a valid state; otherwise, it is in an invalid state. During normal disassembly, the first photoelectric sensor located in the locking block mounting hole and the second photoelectric sensor on the connecting plate surface between the connecting plate and the device body will both be turned on, and the first photoelectric sensor will be turned on slightly later than the second photoelectric sensor, which conforms to the preset sequence. When the device body is illegally disassembled by using brute force to separate it from the connecting plate, the first photoelectric sensor will be turned on, but because the locking block is stuck in the locking block mounting hole, the second photoelectric sensor will not be turned on, which does not conform to the preset sequence. S263, the comprehensive judgment process is as follows: The system determines that the current disassembly operation is normal and does not trigger an alarm mechanism only when both the vibration data flag and the photoelectric sensor flag are in a valid state; if at least one of the vibration data flag or the photoelectric sensor flag is in an invalid state, the system determines that the current disassembly operation is illegal and immediately triggers an alarm mechanism.

9. The method of using the intelligent monitoring device with anti-disassembly function as described in claim 8, characterized in that, It also includes the following steps: S3, Status Monitoring and Early Warning; S31, Signal Acquisition and Preprocessing: A high-precision clock chip is configured on the main control board of the intelligent monitoring device to generate a clock signal as a synchronous sampling reference signal for the fiber optic vital signs sensor, MEMS sensor, and acoustic signature sensor. The sampling control terminals of each sensor are connected to this clock signal source to achieve synchronous sampling. A low-pass filter is used for the heart rate and respiratory signals acquired by the fiber optic vital signs sensor; a band-pass filter is used for the snoring and apnea signals acquired by the MEMS sensor. The audio signal collected by the voiceprint sensor is processed using a filtering algorithm adapted to audio signal processing, retaining the components that match the frequency characteristics of snoring, and filtering out background noise and other irrelevant sound signals. S32, Feature Extraction: Parallel processing of three types of sensors; the fiber optic vital signs sensor outputs a mixed signal, which is then bandpass filtered, detrended and normalized, and wavelet decomposition is used to obtain four sub-bands: respiration, heart rate, body movement and bed exit. Independent component analysis was performed on the respiration and heart rate subbands to suppress crosstalk; subsequently, adaptive peak detection was performed in the heart rate subband to obtain real-time heart rate, heart rate variability and amplitude. The respiratory subband detects the peaks and troughs of breathing to obtain respiratory rate, amplitude, and smoothness; the body movement subband uses a sliding window RMS to detect turning events and their intensity and duration; the bed-off subband determines bed-off based on amplitude drop and records the switching time; the MEMS sensor extracts the dominant frequency, amplitude, phase, and duration of snoring vibrations and monitors the vibration energy attenuation during apnea. The voiceprint sensor obtains the snoring tone through spectrum analysis, extracts the timbre using Mel-frequency cepstral coefficients, calculates the loudness of the sound pressure level, and counts the snoring interval and frequency of occurrence. All features are output to the fusion process after being aligned by timestamps; S33, Data fusion and feature-level fusion: The weighting coefficients are determined based on the dependence and reliability of each monitoring parameter on the sensor features. The feature values ​​of each sensor are multiplied by the weighting coefficients and then summed to obtain the comprehensive snoring feature parameters. The various feature parameters extracted by the fiber optic vital signs sensor, MEMS sensor and acoustic fingerprint sensor are combined into a feature vector in sequence, and the feature parameters in the feature vector are normalized. S34, Model Training and Classification: Organize the collected fusion feature vector data samples and label them with physiological state and behavior category labels. Perform Z-score standardization on each feature in the feature vector and divide the dataset into training set, validation set and test set. The shape of the LSTM network input layer is determined based on the feature vector dimension, and the number of neurons in the LSTM hidden layer is determined accordingly. The system automatically learns the temporal dependencies and long-term memory features in time-series data. A Dropout layer is added after the LSTM hidden layer to prevent overfitting. The number of neurons in the output layer is determined based on the number of categories in the classification task, and a softmax activation function is used. A cross-entropy loss function is defined, and the Adam optimizer is selected. Corresponding parameters are set, and the training set data is input into the LSTM network in batches for forward and backward propagation training. Performance metrics are monitored, and k-fold cross-validation is used to evaluate model performance and adjust hyperparameters. S35, Classification and Alarm Decision: When the intelligent monitoring device is in actual operation, the feature vector will be fused in real time and input into the trained classification model to classify and identify the current user's physiological state and behavior category; when the identification result indicates the presence of abnormal conditions such as snoring, breathing apnea, or cardiac arrest, the alarm mechanism will be triggered, the local sound and light alarm device on the pillow will be activated, and the alarm information will be sent to the remote monitoring center through the wireless communication module.

Images