A method of monitoring movement of an individual sampling device

Abstract

This invention discloses a method for monitoring the movement of individual sampling devices, belonging to the field of individual sampling device monitoring technology. The method collects acceleration and angular velocity data using a 6-axis attitude sensor, separates pump vibration and human motion signals through singular spectrum decomposition combined with fading Kalman filtering, extracts three-dimensional features in the time domain, frequency domain, and correlation domain, determines the device wearing status using a random forest algorithm, and divides valid / invalid sampling periods using time accumulation verification. Simultaneously, it adapts to different working conditions through adaptive benchmark updates and employs hash chain encryption to achieve data tamper-proofing and traceability. This invention has strong anti-vibration interference capability, high anti-cheating accuracy, and wide scenario adaptability, effectively ensuring the authenticity and validity of individual sampling data, and is suitable for individual gas sampling scenarios for occupational hazard factor monitoring.

Description

Technical Field

[0001] This invention relates to the field of individual sampling device monitoring technology, and specifically to a method for monitoring the movement of individual sampling devices. Background Technology

[0002] In the field of occupational disease prevention and control, individual sampling devices need to be worn by workers and move with them to collect the composition and concentration of gases in the working environment in order to assess the probability of occupational diseases.

[0003] In existing technologies, motion detection solutions mostly use a single accelerometer or PIR sensor, which has the following drawbacks: weak anti-interference capability, diaphragm pump vibration can cause serious interference to the sensor signal, and traditional filtering cannot effectively separate pump vibration from human motion signals; insufficient anti-cheating capability, relying solely on simple acceleration threshold judgment, which cannot distinguish between normal walking and manual shaking cheating.

[0004] Based on this, the present invention proposes a method for monitoring the movement of individual sampling devices, thereby solving the aforementioned problems existing in the prior art. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for monitoring the movement of individual sampling devices. Through data acquisition, adaptive vibration compensation, and multi-dimensional feature fusion judgment, the method can achieve accurate monitoring of the wearing status of individual sampling devices and effectively solve the technical problems of pump vibration interference and inaccurate cheating judgment.

[0006] To address the aforementioned technical deficiencies, in a first aspect, the present invention proposes a method for monitoring the movement of an individual sampling device, comprising the following steps:

[0007] Acquire motion data from an individual sampling device, the motion data including acceleration data and angular velocity data;

[0008] The motion data is subjected to vibration interference separation processing to eliminate interference signals generated by the vibration of the pump body inside the individual sampling device, thereby obtaining processed motion data;

[0009] Extract multidimensional features from the processed motion data, including time-domain features, frequency-domain features, and correlation features;

[0010] Based on the multidimensional features, a preset classification model is used to determine the wearing status of the individual sampling device, which includes effective wearing status and ineffective wearing status.

[0011] Based on the determination of the wearing status, the sampling data of the individual sampling device is marked or processed.

[0012] As a further improvement to the technical solution of the present invention, the vibration interference separation processing of the motion data includes:

[0013] Singular spectral decomposition was used to extract the characteristic frequency bands of the pump body vibration, and a vibration signal model was constructed.

[0014] A fading Kalman filter is introduced, and the filter coefficients are dynamically adjusted based on the vibration signal model to counteract the interference of the pump body vibration on the motion data.

[0015] The data processed by the fading Kalman filter is then subjected to low-pass filtering and moving average filtering to obtain low-frequency characteristic signals representing human motion.

[0016] As a further improvement to the technical solution of this invention, the fading Kalman filter achieves dynamic compensation by introducing a fading factor, and its optimal state estimation formula is as follows:

[0017] ;

[0018] in, Let be the optimal posterior state estimate of the system at time k. Let k be the predicted prior state of the system at time k. Let K be the Kalman filter gain at time k. Let k be the observation value at time k. To measure residuals;

[0019] The formula for the fading factor is:

[0020] ;

[0021] in, For residual theory covariance, It is a gradually diminishing factor.

[0022] As a further improvement to the technical solution of the present invention, the extraction of multidimensional features from the processed motion data includes:

[0023] Extract time-domain features, including at least the root mean square of acceleration, peak factor, and gait period stability parameters;

[0024] Frequency domain features are extracted using Fast Fourier Transform, including at least the energy percentage of the frequency bands characteristic of human walking.

[0025] The correlation characteristics between acceleration data and angular velocity data are calculated and used as the correlation characteristics.

[0026] As a further improvement to the technical solution of this invention, the extraction of frequency domain features through Fast Fourier Transform (FFT) is based on the energy proportion of the characteristic frequency band, calculated using the following formula:

[0027]

[0028] In the formula, P represents the proportion of energy in the characteristic frequency band of human walking. The total energy of the characteristic frequency band, Let k be the signal energy at the k-th frequency point. To sum the energy of all frequency points within the 1~3Hz characteristic frequency band; Total energy across the entire frequency band. This is to sum the energy over all M frequency points.

[0029] As a further improvement to the technical solution of the present invention, the step of determining the wearing status of the individual sampling device based on the multidimensional features using a preset classification model includes:

[0030] The multidimensional features are input into the random forest model, and the probability of the individual sampling device being in one of four states is normal walking, cheating shaking, stationary or not being worn.

[0031] The initial wearing status at the current moment is determined based on the probability stated therein;

[0032] A time-cumulative verification condition is set. If the duration of continuously meeting the normal walking characteristics reaches a first threshold, it is determined to be a valid sampling period; if the duration of continuously not meeting the normal walking characteristics reaches a second threshold, it is determined to be an invalid sampling period.

[0033] As a further improvement to the technical solution of the present invention, an adaptive benchmark update step is also included:

[0034] Motion data of the individual sampling device in a static state are collected at predetermined time intervals;

[0035] Based on the motion data under the static state, the acceleration reference value and angular velocity reference value are dynamically updated to offset the effects of sensor zero drift and temperature changes.

[0036] As a further improvement to the technical solution of the present invention, a working condition calibration step is also included:

[0037] Receives external input parameters for different work conditions;

[0038] The determination threshold of the multidimensional features is adaptively adjusted based on the operating parameters.

[0039] As a further improvement to the technical solution of this invention, it also includes anti-tampering and full-process traceability steps:

[0040] The acquired motion data, the processed motion data, the multidimensional features, and the determination result of the wearing status are stored in a chain-like encrypted manner.

[0041] Records event logs of invalid wearing states and supports cross-validation with the backend system.

[0042] Secondly, the present invention provides an individual sampling device, comprising:

[0043] One or more processors;

[0044] Memory, used to store one or more programs;

[0045] A six-axis attitude sensor is used to collect motion data;

[0046] When the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] This invention acquires acceleration and angular velocity motion data of individual sampling devices, performs vibration interference separation processing on the motion data to eliminate pump vibration interference, extracts time-domain, frequency-domain, and correlation multi-dimensional features, and uses a classification model to accurately determine the device wearing status. Finally, based on the determination results, the sampling data is marked and processed. This effectively avoids monitoring misjudgments caused by pump vibration, accurately distinguishes between valid and invalid wearing status, ensures that the sampling data truly matches the actual wearing situation of the operator, and significantly improves the accuracy and data validity of individual sampling mobile monitoring. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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.

[0052] like Figure 1As shown, this invention uses a 6-axis attitude sensor and a main control MCU as the core hardware, and adopts the SSA+FKF+FFT+Random Forest core algorithm. Through five core modules—sensor data acquisition and initialization, adaptive vibration interference separation and compensation, multi-dimensional feature fusion and discrimination, adaptive benchmark update and working condition calibration, and tamper-proof full-process traceability—it achieves closed-loop monitoring of the wearing status and sampling effectiveness of individual sampling devices. The specific implementation steps and parameters of each module are as follows:

[0053] 1. Sensor selection and data acquisition initialization

[0054] This step completes hardware adaptation, raw data acquisition, and initial benchmark calibration, providing standardized input data for subsequent signal processing. Specific implementation details are as follows:

[0055] Sensor hardware selection

[0056] A 6-axis attitude sensor integrating a 3-axis accelerometer and a 3-axis gyroscope is selected. The accelerometer range is set to ±8g, and the gyroscope range is set to ±250° / s. The sensor sampling frequency is set to 50-100Hz. A data transmission connection is established with the main control MCU of the individual sampling device through an I2C or SPI communication interface to ensure the real-time and continuous acquisition of motion data.

[0057] Synchronous acquisition trigger

[0058] After the device is powered on, the main control MCU performs two actions simultaneously: first, it starts the diaphragm pump built into the individual sampling device to begin gas sampling; second, it triggers the 6-axis attitude sensor to start motion data acquisition. At the same time, a pump working status flag is added to the data frame to distinguish the vibration signal of the pump body in working and non-working states.

[0059] Initial baseline value establishment

[0060] After the device is powered on, it is kept stationary for 10 seconds to continuously collect raw motion data. The acceleration and the initial reference value of the gyroscope are calculated using the mean method to eliminate the sensor's factory zero drift error.

[0061] Initial reference value for acceleration (taking the X-axis as an example):

[0062] ;

[0063] Where N is the number of static sampling points, The original acceleration data is for the X-axis. The initial reference values ​​for acceleration on the Y-axis and Z-axis are calculated in the same way as those on the X-axis.

[0064] Initial reference value of gyroscope (taking the X-axis as an example):

[0065] ;

[0066] Original data calibration formula (applied to both triaxial accelerometer and gyroscope):

[0067] ;

[0068] The initialization and calibration described above eliminate initial sensor errors and provide standardized data for subsequent signal processing.

[0069] Initial vibration characteristics of the pump body

[0070] During the initial baseline acquisition phase, the initial vibration characteristics of the diaphragm pump after startup (10-30Hz) were recorded simultaneously as the basic model parameters for subsequent vibration signal separation.

[0071] 2. Adaptive vibration disturbance separation and compensation

[0072] This step, one of the core invention points, employs a four-stage combined processing strategy: singular spectrum decomposition for coarse separation, fading Kalman filter for dynamic compensation, 5Hz low-pass filter, and moving average fine filter, to accurately separate the diaphragm pump vibration signal from the human motion signal. Specific implementation details are as follows:

[0073] Singular Spectral Analysis (SSA) coarse separation

[0074] The 10-30Hz characteristic frequency band of pump vibration is extracted using the singular spectrum decomposition algorithm, and a pump vibration signal model is constructed. This initially separates the pump vibration component from the human motion component in the original calibration data. The core formula for singular spectrum decomposition is as follows:

[0075] ;

[0076] Where Y is the trajectory matrix, U is the left singular matrix, Σ is the diagonal singular value matrix, and VT is the right singular matrix; the singular components corresponding to the pump body vibration are screened out through singular value decomposition, thus completing the coarse separation of the vibration signal.

[0077] Decreasing Kalman Filter (FKF) Dynamic Compensation

[0078] Based on the coarse separation of SSA, a fading Kalman filter is introduced to dynamically adjust the filter coefficients and fading factor, thereby canceling the residual interference of pump vibration on human motion signals in real time and realizing dynamic compensation of vibration signals.

[0079] The formula for optimal state estimation is:

[0080] ;

[0081] in, The optimal posterior state estimate of the system at time k is the optimal estimate of the actual human motion state of the individual sampling device at time k. Specifically, it is the motion state quantities such as triaxial acceleration and triaxial angular velocity after filtering and compensation to completely eliminate the vibration interference of the diaphragm pump. It is the core basic data for subsequent multidimensional feature extraction and wearing status determination.

[0082] The prior state prediction of the system at time k is the motion state prediction of the individual sampling device at time k based on the motion state at time k-1. It includes the uncorrected diaphragm pump vibration interference component, reflects the trend benchmark of human motion, and is the initial prediction input for filtering.

[0083] The Kalman filter gain at time k is the dynamic correction weight for pump vibration interference at time k. It is adaptively adjusted by the fading factor λk: when the diaphragm pump vibration interference is strong, the filter gain is increased to enhance the correction effect of the observation data and offset the influence of vibration on the motion state; when the interference is weak, the gain is decreased to ensure the stability of the state estimation. This achieves dynamic adaptive compensation for vibration interference and is the core adjustment parameter for anti-vibration interference in this invention.

[0084] The observation value at time k is the raw motion observation data of the individual sampling device collected by the 6-axis attitude sensor at time k. Specifically, it is the raw sampled value of the three-axis acceleration and three-axis angular velocity. It is also superimposed with the continuous vibration interference signal of the diaphragm pump and the environmental noise, and is the raw input data source for filtering.

[0085] The measurement residual, which is the difference between the sensor observation data and the state prediction value at time k, is the combined residual term of the diaphragm pump vibration interference and environmental noise. It is the core correction quantity used by the fading Kalman filter to cancel the pump body vibration interference: state prediction value. A benchmark reflecting the trend of human movement, observed values It includes real motion plus pump vibration interference. The difference between the two is the pump vibration interference and noise. The predicted state is corrected by the residual, and finally the real human motion state without vibration interference is obtained.

[0086] The formula for the fading factor is:

[0087] ;

[0088] in, For residual theory covariance, The fading factor is used to dynamically adjust the filter coefficient and counteract pump vibration interference in real time.

[0089] Multi-stage fine filtration

[0090] The data after FKF dynamic compensation is then subjected to 5Hz low-pass filtering and moving average filtering in sequence to remove high-frequency environmental noise and instantaneous fluctuation noise, retaining only the 1-3Hz low-frequency characteristic signal of human walking, thus obtaining pure motion data that can be used for feature extraction.

[0091] 3. Walking state discrimination based on multi-dimensional feature fusion

[0092] This step constructs a three-dimensional feature system encompassing the time domain, frequency domain, and correlation, and combines this with the random forest algorithm to accurately determine the device wearing status, distinguishing between four states: normal walking, cheating / shaking, stationary, and not wearing the device. Specific implementation details are as follows:

[0093] Temporal feature extraction

[0094] Temporal features are extracted from clean motion data to comprehensively reflect the temporal variation patterns of human motion. Core features include:

[0095] Calculate the root mean square acceleration (X-axis) after filtering:

[0096] ;

[0097] In the formula, Let be the pure X-axis acceleration value of the i-th sampling point, and M be the total number of sampling points.

[0098] Simultaneously, peak factor and gait period stability parameters are extracted to characterize the temporal variation features of motion data.

[0099] Correlation feature extraction: Based on the Pearson coefficient, the cross-correlation of X / Y axis acceleration is calculated to capture the coordinated pattern of posture changes during human walking, distinguishing between genuine walking and manual swaying. The formula for calculating the X / Y axis acceleration correlation coefficient is as follows:

[0100] ;

[0101] In the formula, M is the total number of sampled data points, and i is the index of the sampled point (i=1,2,...,M). Let be the pure X-axis acceleration value at the i-th sampling point. Let be the pure Y-axis acceleration value at the i-th sampling point. Let X be the sample mean of the pure acceleration along the X-axis. Let Y be the sample mean of the pure acceleration along the Y-axis. The centered ripple component of the X-axis acceleration. The centered fluctuation component of the Y-axis acceleration. Let X be the sum of squares of the deviations of the acceleration from the mean. Let Y be the sum of squares of the deviations of acceleration from the mean. The standard deviation (normalization factor) of the X-axis acceleration samples. The standard deviation (normalization factor) of the Y-axis acceleration samples.

[0102] The invention uses the Pearson correlation coefficient formula to distinguish between genuine walking and cheating behavior from the dimension of motion posture coordination, and improves the accuracy of judgment by fusing time-domain and frequency-domain features:

[0103] Normal wearing and actual walking: Human gait has natural regularity, and the X / Y axis acceleration fluctuates synchronously with the gait, showing a strong linear correlation. The absolute value is usually ≥0.7, which is judged as an effective wearing state;

[0104] Disadvantages of manual shaking: Human-induced shaking is irregular; there is no coordinated fluctuation in X / Y axis acceleration; and the linear correlation is extremely weak. An absolute value of ≤0.3 indicates an invalid wearing status.

[0105] At rest / not worn: There is no effective fluctuation in acceleration and no effective value in the correlation coefficient, so it is directly judged as an invalid wearing state;

[0106] This feature fundamentally eliminates manual shaking and cheating, ensuring the authenticity and compliance of individual sampling data.

[0107] Frequency domain feature extraction: By analyzing the frequency domain characteristics of the data through Fast Fourier Transform (FFT), the characteristic frequency band of human walking in the 1-3Hz range is identified, and the energy proportion of this frequency band is calculated. The core formula is:

[0108] ;

[0109] In the formula, P represents the proportion of energy in the characteristic frequency band of human walking. The total energy of the characteristic frequency band, Let k be the signal energy at the k-th frequency point. To accumulate the signal energy of all frequency points within the 1~3Hz human walking characteristic frequency band, the total energy of the characteristic frequency band is obtained. Total energy across the entire frequency band. This is to sum the energy over all M frequency points.

[0110] This invention quantifies the energy percentage of the 1-3Hz human walking characteristic frequency band and combines it with the working condition thresholds of different jobs (e.g., workshop inspection ≥55% 1-2Hz, outdoor work ≥60% 1.5-3Hz, mining ≥50% 0.8-1.5Hz) to accurately distinguish four states: normal walking, cheating swaying, stationary, and not wearing a mask.

[0111] Normal walking: The energy is concentrated in 1~3Hz, and the P value meets the corresponding working condition threshold, which is considered to be effective wearing;

[0112] Cheating shaking: Energy is concentrated in the high-frequency band >3Hz, and the P value is far below the threshold, which is judged as invalid wearing;

[0113] When stationary / not worn: the energy in the 1~3Hz frequency band is extremely low, and the P value is close to 0, which is directly judged as invalid wearing;

[0114] The frequency domain dimension enables accurate determination of the wearing status, which greatly improves the accuracy and reliability of anti-cheating identification.

[0115] The three-dimensional features of time domain, frequency domain, and correlation are input into a pre-defined random forest classification model. The model outputs probability values ​​for four states, and the initial state is determined by a voting method, using the following formula:

[0116] ;

[0117] Wherein, P1−P4 are the probabilities of four states: normal walking, cheating and shaking, stationary, and not wearing a mask, respectively.

[0118] Time accumulation verification: If the normal walking characteristics are met continuously for ≥10 minutes, it is determined as a valid sampling period; if the criteria are not met for 30 consecutive seconds, it is marked as an invalid sampling period.

[0119] 4. Adaptive benchmark update and operating condition calibration

[0120] This step addresses issues related to sensor drift, temperature variations, and compatibility with different work scenarios, improving long-term monitoring stability and accuracy. Specific implementation details are as follows:

[0121] Adaptive reference update: Every 30 minutes, the main control MCU controls the sensor to collect short-term static data. The acceleration and gyroscope reference values ​​are fine-tuned through reference increments to compensate for zero drift from long-term sensor operation and the influence of ambient temperature. The acceleration reference increment update formula is as follows:

[0122] ;

[0123] in, The gyroscope reference value is updated in the same way as the acceleration.

[0124] Operating condition calibration: Supports the upper computer to preset operating condition parameters for different types of work and automatically adapts to characteristic thresholds: ≥55% energy ratio in the 1~2Hz frequency band for workshop inspection, ≥60% energy ratio in the 1.5~3Hz frequency band for outdoor operations, and ≥50% energy ratio in the 0.8~1.5Hz frequency band for mining, improving the accuracy of judgment in different scenarios.

[0125] 5. Tamper-proof and full-process traceability

[0126] This step involves encrypted storage of monitoring data, marking invalid data, and full-process traceability to ensure data compliance and authenticity. Specific implementation details are as follows:

[0127] Chain-based encrypted storage: SHA-256 hash chain encryption is used. The core formula is:

[0128] ;

[0129] in, The hash value of the k-th data block. For the content of the k-th data block, bind a timestamp and a unique device identifier to ensure that the data cannot be tampered with.

[0130] Data cross-validation: If a sampling period is determined to be invalid, the main control MCU will mark the gas sampling data of the corresponding period as "Datainvalid". This part of the data will not be included in the occupational disease risk assessment. At the same time, the event logs such as the trigger time, duration and reason for the invalid status are recorded. The system supports the background management system to retrieve historical attitude data and sampling data for cross-validation, so as to achieve full traceability.

[0131] This invention utilizes a combination of singular spectrum decomposition and fading Kalman filtering algorithms to efficiently separate vibration signals from diaphragm pump vibrations in individual sampling devices and human motion signals, significantly improving anti-vibration interference capabilities. By fusing time-domain, frequency-domain, and correlation-based multi-dimensional features and combining them with a random forest model, it can accurately determine the device wearing status, effectively distinguishing between normal walking and cheating shaking, and greatly improving the accuracy of anti-cheating identification. Simultaneously, through adaptive benchmark updates and multi-task calibration, it achieves flexible adaptation to different work scenarios, ensuring long-term monitoring stability. Finally, with the addition of hash chain-based encrypted storage and an automatic marking mechanism for invalid sampling data, it achieves full-process data traceability and tamper-proofness, significantly improving the accuracy, authenticity, and compliance of individual sampling monitoring.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0133] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method of monitoring movement of an individual sampling device, characterized by, Includes the following steps: Acquire motion data from an individual sampling device, the motion data including acceleration data and angular velocity data; The motion data is subjected to vibration interference separation processing to obtain processed motion data; Extract multidimensional features from the processed motion data, including time-domain features, frequency-domain features, and correlation features; Based on the multidimensional features, a preset classification model is used to determine the wearing status of the individual sampling device, which includes effective wearing status and ineffective wearing status. Based on the determination of the wearing status, the sampling data of the individual sampling device is marked or processed.

2. The method of claim 1, wherein, The vibration interference separation process for the motion data includes: Singular spectral decomposition was used to extract the characteristic frequency bands of the pump body vibration, and a vibration signal model was constructed. A fading Kalman filter is introduced, and the filter coefficients are dynamically adjusted based on the vibration signal model to counteract the interference of the pump body vibration on the motion data. The data processed by the fading Kalman filter is then subjected to low-pass filtering and moving average filtering to obtain low-frequency characteristic signals representing human motion.

3. The method of claim 1, wherein, The fading Kalman filter achieves dynamic compensation by introducing a fading factor, and its optimal state estimation formula is as follows: ； wherein, is the optimal estimation value of the posterior state of the system at time k, is the prior state prediction value of the system at time k, is the Kalman filter gain at time k, is the observation value at time k, is the measurement residual; The formula for the fading factor is: ； in, For residual theory covariance, It is a gradually diminishing factor.

4. The method for monitoring the movement of an individual sampling device according to claim 1, characterized in that, The extraction of multidimensional features from the processed motion data includes: Extract time-domain features, including at least the root mean square of acceleration, peak factor, and gait period stability parameters; Frequency domain features are extracted using Fast Fourier Transform, including at least the energy percentage of the frequency bands characteristic of human walking. The correlation characteristics between acceleration data and angular velocity data are calculated and used as the correlation characteristics.

5. The method for monitoring the movement of an individual sampling device according to claim 4, characterized in that, The core of extracting frequency domain features using Fast Fourier Transform is the energy proportion of the characteristic frequency band, calculated using the following formula: ； In the formula, P represents the proportion of energy in the characteristic frequency band of human walking. The total energy of the characteristic frequency band, Let k be the signal energy at the k-th frequency point. To sum the energy of all frequency points within the 1~3Hz characteristic frequency band; Total energy across the entire frequency band. This is to sum the energy over all M frequency points.

6. The method for monitoring the movement of an individual sampling device according to claim 1, characterized in that, The step of determining the wearing status of the individual sampling device based on the multidimensional features and using a preset classification model includes: The multidimensional features are input into the random forest model, and the probability of the individual sampling device being in one of four states is normal walking, cheating shaking, stationary or not being worn. The initial wearing status at the current moment is determined based on the probability stated therein; A time-cumulative verification condition is set. If the duration of continuously meeting the normal walking characteristics reaches a first threshold, it is determined to be a valid sampling period; if the duration of continuously not meeting the normal walking characteristics reaches a second threshold, it is determined to be an invalid sampling period.

7. The method for monitoring the movement of an individual sampling device according to claim 1, characterized in that, It also includes an adaptive benchmark update step: Motion data of the individual sampling device in a static state are collected at predetermined time intervals; Based on the motion data under the static state, the acceleration reference value and angular velocity reference value are dynamically updated to offset the effects of sensor zero drift and temperature changes.

8. The method for monitoring the movement of an individual sampling device according to claim 1, characterized in that, It also includes operating condition calibration steps: Receives external input parameters for different work conditions; The determination threshold of the multidimensional features is adaptively adjusted based on the operating parameters.

9. The method for monitoring the movement of an individual sampling device according to claim 1, characterized in that, It also includes anti-tampering and full-process traceability steps: The acquired motion data, the processed motion data, the multidimensional features, and the determination result of the wearing status are stored in a chain-like encrypted manner. Records event logs of invalid wearing states and supports cross-validation with the backend system.

10. An individual sampling device, characterized in that, include: One or more processors; Memory, used to store one or more programs; A six-axis attitude sensor is used to collect motion data; When the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method as described in any one of claims 1-9.

Images