A smart sleep posture monitoring anti-bedsore nursing system and method based on a pressure sensing array

Abstract

The application discloses a kind of intelligent sleeping posture monitoring anti-bed sore nursing system and method based on pressure sensing array, including pressure data acquisition module, sleeping posture identification module, individual calibration module, space-time pressure risk assessment module and nursing task closed loop verification module.Pressure data acquisition module is formed by the sensing array of the pressure distribution data of human body that a plurality of piezoresistive film pressure sensor splicing gathers;Sleeping posture identification module inputs the pressure data after pre-processing into lightweight convolutional neural network and outputs sleeping posture identification result;Individual calibration module fuses and improves individual adaptability with the identification result and personal calibration archives;Space-time pressure risk assessment module is weighted according to the risk score of each body area according to pressure duration, pressure intensity and pressure concentration degree;Nursing task closed loop verification module continuously reads multiple frames of pressure data to verify pressure relief effect after intervention.The application is suitable for bed sore prevention and nursing in pension institutions and home care scenarios.

Description

Technical Field

[0001] This invention relates to the field of intelligent health monitoring technology, specifically to an intelligent sleep posture monitoring and pressure ulcer prevention nursing system and method based on a pressure sensor array. The system collects human body pressure distribution data through a pressure sensor array, combines a lightweight convolutional neural network and individualized calibration to achieve sleep posture recognition, and performs pressure ulcer risk assessment and nursing closed-loop verification based on spatiotemporal pressure distribution. Background Technology

[0002] Pressure ulcers are a common complication caused by prolonged pressure on local tissues, leading to persistent ischemia, hypoxia, and tissue necrosis. They primarily occur in elderly people who are bedridden for extended periods and patients with limited mobility. Clinical statistics show that the incidence of pressure ulcers in long-term bedridden patients can reach 8% to 15%, increasing patient suffering and treatment costs, and in severe cases, leading to infection and endangering life. Currently, pressure ulcer prevention mainly relies on caregivers turning patients at fixed intervals (usually every 2 hours), but this approach has the following technical problems.

[0003] First, traditional timed turning procedures are performed at fixed intervals, making it impossible to determine which body areas require priority intervention based on the actual pressure levels. The degree of pressure on different body areas varies significantly under different sleeping positions, and fixed-interval turning may result in low-risk areas being frequently disturbed while high-risk areas are not addressed in a timely manner, leading to low nursing efficiency.

[0004] Second, existing pressure sensor-based sleep posture monitoring systems mostly use generic recognition models that do not consider the effects of individual body shape differences, curled-up side-lying postures, and sensor placement conditions. Subjects of different body shapes exhibit significantly different pressure distribution patterns in the same sleeping position, and generic models have a high misjudgment rate for side-lying postures, affecting the accuracy of subsequent risk assessments.

[0005] Third, the existing system cannot automatically verify whether the intervention has truly relieved pressure in high-risk areas after nursing staff perform the turning intervention. There are cases where nursing staff perform the turning operation but the turning is not done properly, and the pressure in high-risk areas is not effectively relieved, and the system cannot detect or track these situations.

[0006] Fourth, existing solutions are mostly geared towards single-person monitoring and lack the ability to manage multiple subjects concurrently and coordinate nursing tasks uniformly. In multi-bed care scenarios in elderly care institutions, nursing staff need to monitor the status of multiple individuals simultaneously. Existing solutions cannot provide centralized risk warnings and task management, making them unsuitable for the needs of multi-bed care. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides an intelligent sleep posture monitoring and pressure ulcer prevention care system and method based on a pressure sensor array, aiming to achieve the following technical objectives: (1) Provides personalized sleeping posture recognition capability by fusing the output of a general convolutional neural network model with a personal calibration profile to reduce recognition errors caused by different body types and sleeping posture habits; (2) Provide risk assessment based on spatiotemporal pressure distribution, calculate risk scores in multiple dimensions such as pressure duration, pressure intensity and pressure concentration, and accurately locate the body areas that need intervention; (3) Provide an automatic verification mechanism for the effect of nursing intervention, continuously read multiple frames of pressure data after intervention to verify the pressure relief effect, and form a traceable nursing closed loop; (4) Supports concurrent monitoring of multiple objects and management of nursing tasks, isolates data by unique identifier, and adapts to multi-bed care scenarios.

[0008] To achieve the above technical objectives, the present invention adopts the following technical solution: A smart sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array includes a pressure data acquisition module (1), a sleep posture recognition module (2), an individualized calibration module (3), a spatiotemporal pressure risk assessment module (4), and a nursing task closed-loop verification module (5).

[0009] The pressure data acquisition module (1) acquires two-dimensional pressure distribution data of a person lying down through a pressure sensor array (8). The pressure sensor array (8) consists of multiple piezoresistive thin-film pressure sensors arranged in rows and columns to form a sensing matrix. The pressure values ​​of each sensing unit are scanned and read through an analog multiplexer and a microcontroller (9) in a row-column multiplexing manner. The microcontroller (9) uploads the pressure data to the server (10) via a wireless network.

[0010] The sleeping posture recognition module (2) preprocesses the two-dimensional pressure distribution data and inputs it into a lightweight convolutional neural network model, outputting the recognition results of three sleeping postures: supine, left lateral, and right lateral. The preprocessing includes five steps: clearing edge column crosstalk data, threshold clearing, quantile scaling, square root transformation, and interpolation scaling.

[0011] The individualized calibration module (3) collects stable pressure frames of various sleeping positions for each monitored subject to generate a personal calibration profile. During inference, it fuses the output of the lightweight convolutional neural network model with the individual posture feature prototype to obtain the final sleeping position recognition result. When the personal calibration profile is missing or incomplete, it automatically reverts to the output of the lightweight convolutional neural network model.

[0012] The spatiotemporal pressure risk assessment module (4) maintains continuous pressure exposure data for each sensing unit in the pressure sensing array, calculates the risk score of each sensing unit based on the pressure duration, pressure intensity and pressure concentration, and maps high-risk sensing units to the corresponding body regions.

[0013] The nursing task closed-loop verification module (5) creates a nursing task when the risk score of the body area reaches the preset threshold. After the nursing intervention is completed, it continuously reads multiple frames of pressure data to determine whether the original high-risk area has been effectively relieved. If it is effective, the task is archived; if it is ineffective, the task is re-triggered or upgraded.

[0014] The system also includes an out-of-bed detection module (6) and a multi-object management module (7). The out-of-bed detection module (6) continuously monitors the total pressure value of the pressure sensor array. When the total pressure value is lower than a preset pressure threshold and the duration exceeds a preset time threshold, it is determined to be an out-of-bed state, and an alarm signal is triggered during a preset nighttime period. The multi-object management module (7) assigns a unique identifier to each monitored subject, stores various data in isolation according to the unique identifier, summarizes and displays the status of multiple monitored subjects through a web interface, and pushes data updates in real time through the WebSocket protocol.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) By using an individualized calibration fusion mechanism, the output of the general convolutional neural network model is weighted and fused with the individual posture feature prototype to make a decision. This effectively solves the problem of misjudgment of side lying caused by different body types, curled-up side lying and sensor placement differences, and improves the individual adaptability of sleeping posture recognition. (2) By using the spatiotemporal pressure risk assessment module, a composite risk score is calculated by weighting the duration of pressure, pressure intensity and pressure concentration. This can accurately locate the body area that needs intervention, avoid the blindness of the traditional timed turning program, and improve the efficiency of nursing resource allocation. (3) Through the nursing task closed-loop verification mechanism, the pressure relief effect is automatically verified by continuously reading multiple frames of pressure data after the intervention, ensuring that the nursing intervention is truly effective, forming a traceable nursing closed loop of risk detection, task assignment, intervention execution, effect verification and result archiving, avoiding the problem of inadequate turning over but the system is unaware of it; (4) Data is isolated by a unique identifier through the multi-object management module, supporting concurrent monitoring of multiple beds and unified scheduling of nursing tasks, adapting to the multi-bed care scenario of elderly care institutions; (5) The system hardware adopts a modular sensor splicing design and a lightweight convolutional neural network. The model has only 15,451 parameters and the model file is about 60KB. It can run efficiently on resource-constrained embedded platforms and servers, and is low in cost, making it suitable for deployment in grassroots elderly care institutions and home care scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2This is a flowchart of the preprocessing process for the sleeping posture recognition module of the present invention; Figure 3 This is a schematic diagram of the lightweight convolutional neural network structure of the present invention; Figure 4 This is a flowchart illustrating the workflow of the individualized calibration module of the present invention. Figure 5 This is a flowchart of the spatiotemporal pressure risk assessment module of the present invention; Figure 6 This is the state transition diagram of the nursing task closed-loop verification module of the present invention; Figure 7 This is a flowchart illustrating the judgment process of the bed-off detection module of the present invention. Figure 8 This is a schematic diagram of the web interface of the multi-object management module of the present invention; Figure 9 This is a schematic diagram of the hardware connection of the pressure sensing array of the present invention; Figure 10 This is an overall flowchart of the method of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0018] like Figure 1 As shown, this embodiment provides an intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array, including a pressure data acquisition module (1), a sleep posture recognition module (2), an individualized calibration module (3), a spatiotemporal pressure risk assessment module (4), a nursing task closed-loop verification module (5), an exit detection module (6), and a multi-object management module (7). The pressure sensor array (8), microcontroller (9), and server (10) constitute the hardware foundation of the system.

[0019] Pressure data acquisition module (1): In this embodiment, four 16×16 piezoresistive thin-film pressure sensors of 25cm×55cm are used, spliced ​​in a 2×2 manner to form a sensing matrix covering approximately 110cm×55cm with a resolution of 32×32. Each sensor contains 256 pressure sensing units, and each sensing unit corresponds to a physical area of ​​approximately 3.4cm×1.7cm. Eight CD4067 16-channel analog multiplexers are used to realize row and column multiplexing scanning, of which four are used for row selection and four are used for column selection. The ESP32-S3 microcontroller (9) sequentially enables a pair of multiplexers corresponding to the target sensor and selects the row and column intersection points, and reads the analog values ​​in the range of 0 to 4095 through the built-in 12-bit analog-to-digital converter. The microcontroller uploads 1024 comma-separated integer sample values ​​to the server (10) via 2.4GHz Wi-Fi using HTTP POST.

[0020] Sleep posture recognition module (2): After receiving the data, the server reshapes it into a 32×32 matrix, such as Figure 2 The following preprocessing steps are performed: First, remove crosstalk data from columns 0 and 31, as these columns produce invalid readings due to sensor stitching edge effects; Second, zero out sampling points below the effective pressure threshold to eliminate environmental noise; Third, scale and limit the effective pressure of the current frame to the 99th percentile value and the range to 0 to 1 to eliminate inter-frame amplitude differences; Fourth, perform a square root transformation y=√x to compress the dynamic range, making the low-pressure region features more prominent; Fifth, adjust the 32×32 matrix to a 64×32 size using bilinear interpolation as the model input.

[0021] like Figure 3 As shown, the preprocessed 64×32 stress matrix is ​​input into the MiniPostureNet lightweight convolutional neural network. This network contains four convolutional modules: the first module uses a 5×5 convolutional kernel with 8 channels, followed by a 2×2 max-pooling layer, outputting a 32×16 feature map; the second module uses a 3×3 convolutional kernel with 16 channels, followed by a 2×2 max-pooling layer, outputting a 16×8 feature map; the third module uses a 3×3 convolutional kernel with 32 channels, followed by a 2×2 max-pooling layer, outputting an 8×4 feature map; the fourth module uses a 3×3 convolutional kernel with 32 channels, without a pooling layer, outputting an 8×4 feature map. Each convolutional layer is followed by a batch normalization layer and a ReLU activation function. A global average pooling layer is applied after the fourth convolutional module to compress the 32 8×4 feature maps into a 32-dimensional feature vector, which is then passed through a fully connected layer to output scores for three sleeping postures, normalized to probability values ​​by the Softmax function. The model has a total of 15,451 parameters, a model file size of approximately 60KB, and a computational cost of approximately 1.88M MACs.

[0022] Individualized calibration module (3): such as Figure 4 As shown, stable pressure frames were collected for each subject in three sleeping positions: supine, left lateral, and right lateral. During data collection, the system prompted the subject to maintain the target posture. A stable frame was defined as one where the pressure change between consecutive frames was less than a preset stability threshold. After collecting 15 stable pressure frames for each sleeping position, the feature prototypes for each posture were calculated, and a personal calibration profile was generated and saved to the server in JSON format. During the inference phase, the three posterior probabilities output by MiniPostureNet were weighted and fused with the personal posture feature prototypes. The category corresponding to the maximum value after fusion was taken as the final sleeping position recognition result. When the personal calibration profile was missing or incomplete, the system automatically reverted to the output of MiniPostureNet to ensure that the system could still operate in an uncalibrated state.

[0023] Spatiotemporal pressure risk assessment module (4): such as Figure 5 As shown, 1024 sensor units (32×32) maintain continuous pressure exposure data. For each sensor unit, a risk score is calculated based on three dimensions: pressure duration (48%), reflecting the cumulative duration of continuous pressure on the unit; pressure intensity (32%), reflecting the absolute pressure level experienced by the unit; and pressure concentration (20%), reflecting the degree of pressure concentration in the surrounding area. The weighted sum of these three dimensions outputs an engineering risk score from 0 to 100. Risk levels are divided into four intervals: 0-29 for low risk, 30-59 for concern, 60-79 for higher risk, and 80-100 for high risk. High-risk sensor units are mapped to body regions such as the shoulder and back, hip, sacrum and coccyx, or heel, outputting the main pressure areas and corresponding risk levels.

[0024] Nursing task closed-loop verification module (5): such as Figure 6 As shown, when the risk score for the same major body region reaches 30 points and appears consecutively for 3 frames, the system creates a nursing task and records it as pending. After the caregiver accepts the task through the web interface, the status changes to accepted. After performing the turn, the caregiver clicks the "complete" button to enter the intervention state. The system then enters the verification phase, continuously reading 5 frames of pressure data to determine whether the pressure value of the original high-risk area is lower than the second preset threshold. If it is lower, it is determined to be effective pressure relief, and the nursing task status is changed to archived, completing the closed loop; if it is not lower, it is determined to be ineffective pressure relief, and the system recreates the nursing task or upgrades the task priority, notifying the caregiver to re-execute the intervention.

[0025] Bed exit detection module (6): such as Figure 7As shown, the system continuously monitors the total pressure value of the pressure sensor array. When the total pressure value is below 5000 and the duration exceeds 30 seconds, it is determined that the person has left the bed. The system further determines whether it is currently nighttime (22:00 to 07:00). If the leaving of bed occurs during nighttime, the system automatically triggers a full-screen flashing alarm and an audio alert to notify the caregiver. If the leaving of bed occurs during daytime, the system only records the event and does not trigger an alarm.

[0026] Multi-object management module (7): such as Figure 8 As shown, each monitored individual is assigned a unique patient_id identifier, and stress data, personal calibration files, risk assessment data, and nursing task data are stored in isolation by patient_id. The caregiver's workbench displays the sleeping position, bedside status, risk level, and nursing tasks of multiple monitored individuals in a card-based format via a web interface, supporting viewing details, navigating to individual monitoring pages, and handling nursing tasks. The front-end and back-end push data updates in real time via the Socket.IO protocol to ensure that the interface status is synchronized with actual monitoring.

[0027] like Figure 9 As shown, the hardware connection relationship of the pressure sensing array (8) is as follows: four 16×16 piezoresistive thin-film pressure sensors A, B, C, and D are spliced ​​in a 2×2 manner, and the row lines and column lines of each sensor are connected to the corresponding CD4067 analog multiplexer. The four row multiplexers are responsible for selecting the currently scanned row, and the four column multiplexers are responsible for selecting the currently scanned column. The ESP32-S3 microcontroller (9) controls the channel selection port of the multiplexer through the GPIO pin, and after selecting the row and column intersection, reads the analog value of the sensing unit through the ADC pin. After scanning 1024 sensing units, the microcontroller packages the data and uploads it to the Flask inference server (10) via Wi-Fi.

[0028] like Figure 10 As shown, this embodiment provides a method for preventing bedsores by monitoring sleep posture based on a pressure sensor array, including the following steps: S1 collects two-dimensional pressure distribution data of a person lying down using a 32×32 pressure sensor array. The ESP32-S3 microcontroller scans 1024 sensing units in a row-column multiplexing manner and uploads the 1024 sampled values ​​to the server via Wi-Fi at a frequency of one frame per second.

[0029] S2, the server preprocesses the pressure distribution data: removes crosstalk data in columns 0 and 31; zeroes out sampling points below the effective pressure threshold; scales to the 0-1 range using the 99th percentile; performs square root transformation to compress the dynamic range; and adjusts the 32×32 matrix to 64×32 size using bilinear interpolation. The preprocessed 64×32 pressure matrix is ​​then input into the MiniPostureNet lightweight convolutional neural network, which outputs the probabilities of three sleeping positions: supine, left lateral, and right lateral.

[0030] S3: Read the subject's personal calibration file, and perform a weighted fusion of the three posterior probabilities of sleeping positions output in step S2 with the subject's personal posture feature prototype. Take the category corresponding to the maximum value after fusion as the final sleeping position recognition result. If the personal calibration file is missing or incomplete, the output of step S2 is used directly as the final sleeping position recognition result.

[0031] S4 maintains continuous pressure exposure data for each sensor unit, and calculates a risk score from 0 to 100 using a weighted summation method with 48% for pressure duration, 32% for pressure intensity, and 20% for pressure concentration. Sensor units with risk scores exceeding the preset level are mapped to body regions such as the shoulder and back, hip, sacrum and coccyx, or heel, and the main pressure areas and risk levels are output.

[0032] S5: When the risk score for the same body area reaches 30 points for 3 consecutive frames, a nursing task is created. After the caregiver accepts the task and performs the turning intervention, the system continuously reads 5 frames of pressure data to verify the pressure relief effect. If the pressure value of the original high-risk area is lower than the second preset threshold, it is determined to be effective pressure relief, and the nursing task is archived; if it is not lower, it is determined to be ineffective pressure relief, and the nursing task is recreated or the task priority is upgraded.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any person skilled in the art can make several variations or modifications to the disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and modifications 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 scope of the present invention.

Claims

1. A smart sleep posture monitoring and pressure ulcer prevention system based on a pressure sensor array, characterized in that, include: The pressure data acquisition module (1) is configured to acquire two-dimensional pressure distribution data of a person lying down through a pressure sensing array (8), wherein the pressure sensing array is formed by splicing multiple piezoresistive thin-film pressure sensors to form a sensing matrix; the sleeping posture recognition module (2) is configured to preprocess the two-dimensional pressure distribution data and input it into a lightweight convolutional neural network model, and output the recognition results of three types of sleeping postures: supine, left lateral, and right lateral; the individualized calibration module (3) is configured to collect stable pressure frames of multiple sleeping postures for each monitored person to generate a personal calibration profile, and fuse the output of the lightweight convolutional neural network model with the personal posture feature prototype during inference to obtain the final sleeping posture recognition result. When the personal calibration file is missing or incomplete, it automatically falls back to the output of the lightweight convolutional neural network model; the spatiotemporal pressure risk assessment module (4) is configured to maintain continuous pressure exposure data for each sensing unit in the pressure sensing array, calculate the risk score of each sensing unit based on the pressure duration, pressure intensity and pressure concentration, and map the high-risk sensing unit to the corresponding body area; the nursing task closed-loop verification module (5) is configured to create a nursing task when the risk score of the body area reaches a preset threshold, continuously read multiple frames of pressure data after the nursing intervention is completed, determine whether the original high-risk area has been effectively depressurized, archive the task if it is effective, and re-trigger or upgrade the task if it is ineffective.

2. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The individualized calibration module (3) is specifically configured as follows: for each monitored person, stable pressure frames of three sleeping positions are collected, namely supine, left lateral, and right lateral. After a preset number of frames are collected for each sleeping position, a personal calibration file is generated and saved in the form of a data file. During inference, the three types of posterior probabilities output by the lightweight convolutional neural network model are weighted and fused with the personal posture feature prototype, and the category corresponding to the maximum value after fusion is taken as the final sleeping position recognition result.

3. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The nursing task closed-loop verification module (5) is specifically configured as follows: when the risk score of the same body area reaches the first preset threshold and a preset number of frames appear consecutively, a nursing task is created and recorded as a pending order; when the caregiver accepts the order and performs the turning intervention, the verification stage is entered, and a preset number of pressure data frames are read consecutively to determine whether the pressure value of the original high-risk area is lower than the second preset threshold. If the pressure is lower than the threshold, the pressure relief is considered valid and the nursing task is archived. If the pressure is not lower than the threshold, the pressure relief is considered invalid and the nursing task is recreated or the task priority is upgraded.

4. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The spatiotemporal pressure risk assessment module (4) is specifically configured as follows: calculate the risk score of each sensing unit by weighted summation of the proportion of pressure duration, pressure intensity, and pressure concentration. The range of the risk score is a preset interval, and the sensing units with risk scores exceeding the preset level are mapped to the body areas of the shoulder, back, hip, sacrum, coccyx, or heel.

5. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The preprocessing of the sleeping posture recognition module (2) includes: clearing crosstalk data from the edge columns of the pressure sensing array; clearing sampling points below the effective pressure threshold to zero; scaling and limiting to a preset range using the preset quantile value of the effective pressure of the current frame; performing a square root transformation on the scaled data to compress the dynamic range; and adjusting the pressure matrix to the input size of the lightweight convolutional neural network using an interpolation method.

6. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The lightweight convolutional neural network includes: four convolutional modules with 8, 16, 32 and 32 channels respectively; each convolutional layer is followed by a batch normalization layer and a ReLU activation function; the first three convolutional modules are followed by a max pooling layer, and the fourth convolutional module is followed by a global average pooling layer; after global average pooling, a fully connected layer is connected to output the scores of the three sleeping postures, which are then converted into posture probabilities by Softmax normalization.

7. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, It also includes a bed-off detection module (6), configured to: continuously monitor the total pressure value of the pressure sensor array, and determine the bed-off state when the total pressure value is lower than a preset pressure threshold and the duration exceeds a preset time threshold. An alarm signal is triggered when the person leaves the bed during a preset nighttime period.

8. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, It also includes a multi-object management module (7), configured to: assign a unique identifier to each monitored person, and store the stress data, personal calibration file, risk assessment data and nursing task data in isolation according to the unique identifier; The web interface displays the sleeping posture, risk level, and care tasks of multiple monitored individuals in a card format, and pushes data updates in real time via the WebSocket protocol.

9. The intelligent sleep posture monitoring and pressure ulcer prevention nursing system based on a pressure sensor array according to claim 1, characterized in that, The pressure sensing array (8) is composed of multiple piezoresistive thin-film pressure sensors arranged in rows and columns. Each sensor contains a preset number of pressure sensing units. The pressure values ​​of each sensing unit are scanned and read in a row and column multiplexing manner by an analog multiplexer and a microcontroller (9). The microcontroller uploads the pressure data to the server (10) via a wireless network.

10. A method for preventing bedsores by monitoring sleep posture based on a pressure sensor array, characterized in that, Includes the following steps: S1. Collect two-dimensional pressure distribution data of a person lying down using a pressure sensor array; S2. After preprocessing the pressure distribution data, input it into a lightweight convolutional neural network model and output the sleeping posture recognition result; S3. Fuse the sleeping posture recognition result with the monitored person's personal calibration file to obtain the final sleeping posture recognition result; S4. Calculate the risk score based on the pressure duration, pressure intensity, and pressure concentration of each sensor unit and map it to a body area; S5. When the risk score of the body area reaches a preset threshold, create a nursing task. After the nursing intervention is completed, continuously read multiple frames of pressure data to verify the pressure relief effect, and archive or upgrade the task based on the verification results.

Images