Pressure sore prevention mattress regulation and control system based on multi-modal perception and dynamic risk zoning

Abstract

The invention relates to the technical field of intelligent medical care equipment, in particular to a pressure sore prevention mattress regulation and control system based on multi-modal sensing and dynamic risk zoning, which comprises an intelligent mattress module, a multi-modal sensing module, a regulation and control module and a mattress driving module, each sub-airbag is provided with a pressure feedback sensor, and the preset areas are in one-to-one correspondence with the sub-airbags; the multi-modal sensing module integrates pressure, temperature and humidity sensing arrays, and acquires multi-dimensional data of a corresponding preset area in a spatial alignment manner; the regulation and control module generates a coupling data field through time alignment, analyzes and outputs a dynamic risk partition map through a convolutional neural network model, and then generates a regulation and control instruction in combination with the current pressure of the sub-airbags; the mattress driving module performs local pressure regulation, and the system further comprises a feedback optimization module for dynamically adjusting parameters. Precise identification and directional pressure reduction of high-risk areas of pressure sores are realized, and the problem of low efficiency of global intervention is avoided.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical care equipment technology, specifically to an anti-pressure ulcer mattress control system based on multimodal perception and dynamic risk zoning. Background Technology

[0002] Pressure injuries (also known as pressure ulcers) are a common and serious complication in long-term bedridden patients. They are mainly caused by prolonged pressure on local tissues, leading to obstructed blood circulation, tissue ischemia, hypoxia, and necrosis. Traditional methods for preventing pressure ulcers mainly rely on nursing staff to manually turn the patient at regular intervals and use static pressure-reducing mats. These methods are inefficient, lack timely intervention, and are not personalized enough to meet the continuous care needs of long-term bedridden patients.

[0003] In the prior art, invention patent CN118806270B discloses a nursing monitoring system based on an anti-pressure ulcer smart mattress. This system monitors the patient's body movement to determine if the turning frequency is insufficient and issues a reminder. However, it lacks the ability to directly monitor and proactively intervene in key factors contributing to pressure ulcer formation. Invention patent CN120436905A proposes an anti-pressure ulcer pad based on deep Q-learning and attention mechanisms. It primarily relies on pressure distribution and posture data to optimize airbag adjustment strategies. However, the reinforcement learning model is complex to train and does not consider key early physiological indicators of pressure ulcers, such as abnormal skin temperature changes caused by local tissue ischemia and hypoxia, resulting in insufficient universality in clinical scenarios. Furthermore, most existing technologies employ global intervention strategies, such as circulating inflation and deflation of all mattress airbags, which cannot provide precise and targeted pressure relief intervention for high-risk areas, leading to low intervention efficiency and potential impact on patient comfort. Therefore, there is an urgent need for an intelligent anti-pressure ulcer system that can integrate multi-dimensional physiological information, accurately identify high-risk areas in real time, and provide proactive and refined local intervention accordingly. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention aims to provide a pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning. This system can improve pressure ulcer prevention effectiveness and patient comfort by enabling early and accurate identification of high-risk areas for pressure ulcers and allowing for refined local intervention.

[0005] The basic solution provided by this invention is a pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning, characterized in that it includes an intelligent mattress module, a multimodal perception module, a control module, and a mattress drive module. The smart mattress module includes an airbag execution array, which includes multiple sub-airbags that can be independently controlled to inflate and deflate. Each sub-airbag in the airbag execution array is equipped with an independent pressure feedback sensor to collect the current pressure value of the sub-airbag in real time. The array is divided into preset areas corresponding to the number of sub-airbags according to the distribution of the sub-airbags, and the preset areas correspond one-to-one with the sub-airbags. The multimodal sensing module includes a pressure sensing array, a temperature sensing array, and a humidity sensing array. The pressure sensing array includes multiple pressure sensing units for real-time acquisition of body pressure data at preset points on the contact surface between the bedridden person and the mattress. The temperature sensing array includes multiple temperature sensing units for contact acquisition of temperature data at preset points on the contact surface between the bedridden person's body and the mattress. The humidity sensing array includes multiple humidity sensing units for contact acquisition of humidity data at preset points on the contact surface between the bedridden person's body and the mattress. The pressure sensing array, temperature sensing array, and humidity sensing array are arranged in a spatially aligned manner. The control module includes: a data fusion unit, used to perform time alignment on the body pressure, temperature, and humidity data at each preset point; based on preset region division, to perform regional aggregation on the time-aligned body pressure, temperature, and humidity data at each preset point within each preset region to obtain a coupled three-dimensional vector for each preset region; and to construct a body pressure, temperature, and humidity coupled data field based on the coupled three-dimensional vectors of all preset regions; a risk identification unit, with a built-in risk identification model, used to analyze the body pressure, temperature, and humidity coupled data field and output a dynamic risk zoning map, which identifies high-risk areas with risk probabilities exceeding a preset risk threshold and their corresponding risk probability values; and a control strategy unit, used to generate control instructions for one or more specific sub-airbags in the airbag execution array based on the dynamic risk zoning map and the current pressure value of each sub-airbag, wherein the control instructions include a target pressure value and an adjustment speed. The mattress drive module is used to control the air pumps and valves of the corresponding sub-airbags according to the control instructions, and to perform local pressure regulation operations.

[0006] The principle of this invention is as follows: The pressure, temperature, and humidity sensing arrays of the multimodal sensing module adopt a spatially aligned design to collect body pressure, temperature, and humidity data of the contact surface between the bedridden person and the mattress, realizing the corresponding collection of multi-dimensional information in the same physical area and obtaining risk-related data of each preset area of ​​the mattress; The data fusion unit of the control module uses the collection timestamp as a reference and eliminates the collection delay of different sensors through time alignment, integrating the scattered single-dimensional data into a unified body pressure, temperature, and humidity coupled data field, providing a standardized and integrated data carrier for risk analysis; The risk identification model performs in-depth analysis of the coupled data field, explores the synergistic effect of pressure, temperature, and humidity on pressure ulcer formation, accurately identifies high-risk areas with excessive risk levels, and outputs a dynamic risk zoning map containing the location, range, and risk probability value of the area; The control strategy unit, based on the risk zoning map, locks the sub-airbags corresponding to the high-risk areas, and generates control instructions containing target pressure values ​​and adjustment speeds based on the current pressure value of the corresponding sub-airbags. The mattress drive module responds to the instructions to control the inflation and deflation of the corresponding sub-airbags, realizing precise and localized decompression intervention in high-risk areas.

[0007] The beneficial effects of this invention are as follows: It centrally integrates the three core influencing factors of pressure ulcers—pressure, temperature, and humidity—and the array spatial alignment design ensures that data are associated with the same area, avoiding misjudgment of risks caused by multimodal data misalignment, and can accurately capture early signs of pressure ulcers; it uses risk zoning maps to directionally adjust corresponding sub-airbags, replacing the traditional global circulation decompression mode, which not only improves the targeting of decompression in high-risk areas, but also reduces unnecessary pressure fluctuations in non-risk areas, balancing the effect of pressure ulcer prevention with patient comfort; from data collection and risk identification to decompression intervention, no manual intervention is required throughout the entire process, and it can continuously adapt to changes in the body position and fluctuations in the physiological state of bedridden patients, meeting the continuous care needs of long-term bedridden patients and reducing the workload of nursing staff.

[0008] Furthermore, the risk identification model includes a convolutional neural network model, whose input is a coupled data field of body pressure, temperature, and humidity, and whose output is a risk probability vector containing risk probability values ​​of each preset region. The dynamic risk zoning map is obtained by threshold segmentation and connected component clustering of the risk probability vector. The threshold segmentation logic includes: setting a preset risk threshold, and selecting preset regions with risk probabilities not lower than the risk threshold based on the risk probability vector, and marking them as candidate high-risk regions. The connected component clustering logic includes: judging the adjacency relationship between candidate high-risk regions based on preset neighborhood rules, clustering adjacent candidate high-risk regions into a complete high-risk region, whose risk probability value is the weighted average of the risk probability values ​​of each preset region that makes up the complete high-risk region; non-adjacent candidate high-risk regions are treated as independent high-risk regions, and their risk probability value is the risk probability value of the corresponding preset region.

[0009] Multi-dimensional feature extraction is performed using convolutional neural networks to uncover deep correlations between body pressure, temperature, and humidity data. The output risk probability vector directly corresponds to the physical area of ​​the mattress. A dynamic risk zoning map generated through threshold segmentation and connected component clustering provides a clear spatial basis for generating control instructions. The threshold segmentation logic ensures the objectivity of candidate high-risk area selection, while the connected component clustering logic supported by preset neighborhood rules avoids fragmented division of high-risk areas, ensuring that the same physically continuous high-risk areas are integrated into a complete area. For a complete high-risk area, its risk probability value is obtained through weighted averaging, and the risk probability values ​​of each preset area that makes up the complete risk area are averaged to ensure uniformity during adjustment. This avoids different adjustments between adjacent preset areas that may cause patient discomfort or even aggravate pressure ulcers, thus improving patient comfort.

[0010] Furthermore, the logic for generating control commands by the control strategy unit includes: mapping the location of high-risk areas in the dynamic risk zoning map to the corresponding target sub-airbags; calculating the expected pressure value of the target sub-airbags based on the risk probability value of the high-risk areas using a preset mapping function, wherein the preset mapping function is a linear mapping function and the expected pressure value is negatively correlated with the risk probability value; calculating the pressure difference between the current pressure of the target sub-airbags and the expected pressure value; and determining the inflation / deflation rate according to a preset range and speed correspondence rule based on the preset difference range to which the pressure difference belongs.

[0011] The linear mapping function makes the expected pressure value negatively correlated with the risk probability value, that is, the higher the risk, the lower the expected pressure value. The pressure difference (decompression amplitude) is calculated in combination with the current pressure value. Under normal scenarios where the current pressure of the sub-balloon is within the dynamic stability range of the system (near the default reference pressure), the higher the risk, the larger the pressure difference, that is, the larger the decompression amplitude, thus achieving a precise match between risk level and decompression intensity. At the same time, the inflation and deflation speed is divided according to the pressure difference, which ensures a rapid response when the difference is high and avoids frequent fine-tuning when the difference is low, thereby improving patient comfort. High-risk areas can be mapped to the corresponding target sub-balloon, ensuring that the control commands are accurately directed to the corresponding execution unit.

[0012] Furthermore, the method for time alignment by the data fusion unit includes: acquiring the collection timestamps of body pressure data, temperature data, and humidity data from all preset points, using the earliest collection timestamp as a benchmark; and using linear interpolation to synchronize the time of other data, ensuring that the time deviation of each data collection after time synchronization does not exceed a preset time threshold.

[0013] Using the earliest acquisition timestamp as a benchmark, and combining linear interpolation, time calibration is performed on data of different modalities to effectively eliminate time differences caused by sensor acquisition frequency and transmission delay, ensuring that multi-dimensional data at the same moment correspond and match; the time deviation is controlled within a preset threshold to avoid distortion of coupled data fields caused by time asynchrony, thereby ensuring the accuracy of input data for the risk identification model and reducing the risk of misjudgment.

[0014] Furthermore, it also includes a feedback optimization module, which is used to re-collect data through the pressure sensing array and multimodal sensing module within a preset time after the mattress drive module performs local pressure adjustment operation, and calculate the risk probability value reduction rate of the original high-risk area; if the reduction rate is lower than the preset reduction rate threshold, the mapping function parameters in the control strategy unit are gradually adjusted according to the preset parameter adjustment range until the reduction rate reaches the preset reduction rate threshold.

[0015] The feedback optimization module enables closed-loop optimization of the control strategy, which can dynamically adjust the mapping function parameters according to the actual intervention effect, adapt to the physical condition and postural characteristics of different patients, and avoid intervention failure caused by fixed strategies. If the risk reduction in the original high-risk area does not meet expectations, the parameters are adjusted in time to enhance the decompression intensity and ensure that the risk continues to decrease. If the expected results are achieved, the current strategy is maintained, realizing a continuous optimization cycle of effect evaluation, parameter adjustment and effect re-evaluation. Attached Figure Description

[0016] Figure 1 This is a system module diagram of an embodiment of the pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning of the present invention. Detailed Implementation

[0017] The following detailed description illustrates the specific implementation method: The basic implementation examples are as follows: Figure 1 As shown: A pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning, characterized in that: it includes an intelligent mattress module, a multimodal perception module, a control module, and a mattress drive module; The smart mattress module includes an airbag execution array, which includes multiple sub-airbags that can be independently controlled to inflate and deflate. Each sub-airbag in the airbag execution array is equipped with an independent pressure feedback sensor to collect the current pressure value of the sub-airbag in real time. The array is divided into preset areas corresponding to the number of sub-airbags according to the distribution of the sub-airbags, and the preset areas correspond one-to-one with the sub-airbags. The multimodal sensing module includes a pressure sensing array, a temperature sensing array, and a humidity sensing array. The pressure sensing array includes multiple pressure sensing units for real-time acquisition of body pressure data at preset points on the contact surface between the bedridden person and the mattress. The temperature sensing array includes multiple temperature sensing units for contact acquisition of temperature data at preset points on the contact surface between the bedridden person's body and the mattress. The humidity sensing array includes multiple humidity sensing units for contact acquisition of humidity data at preset points on the contact surface between the bedridden person's body and the mattress. The pressure sensing array, temperature sensing array, and humidity sensing array are arranged in a spatially aligned manner. The control module includes: a data fusion unit, used to perform time alignment on the body pressure, temperature, and humidity data at each preset point; based on preset region division, to perform regional aggregation on the time-aligned body pressure, temperature, and humidity data at each preset point within each preset region to obtain a coupled three-dimensional vector for each preset region; and to construct a body pressure, temperature, and humidity coupled data field based on the coupled three-dimensional vectors of all preset regions; a risk identification unit, with a built-in risk identification model, used to analyze the body pressure, temperature, and humidity coupled data field and output a dynamic risk zoning map, which identifies high-risk areas with risk probabilities exceeding a preset risk threshold and their corresponding risk probability values; and a control strategy unit, used to generate control instructions for one or more specific sub-airbags in the airbag execution array based on the dynamic risk zoning map and the current pressure value of each sub-airbag, wherein the control instructions include a target pressure value and an adjustment speed. The mattress drive module is used to control the air pumps and valves of the corresponding sub-airbags according to the control instructions, and to perform local pressure regulation operations.

[0018] The pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning of the present invention also includes a feedback optimization module, which is used to re-collect data through the pressure sensing array and the multimodal perception module within a preset time after the mattress drive module performs local pressure adjustment operation, and calculate the risk probability value reduction rate of the original high-risk area; if the reduction rate is lower than the preset reduction rate threshold, the mapping function parameters in the control strategy unit are gradually adjusted according to the preset parameter adjustment range until the reduction rate reaches the preset reduction rate threshold.

[0019] In this embodiment, the smart mattress module has the following dimensions: mattress size 1.8m (length, X-axis) * 0.9m (width, Y-axis); airbag execution array: sub-airbags are configured as 16*8 independent micro sub-airbags; the size of a single sub-airbag is 11.25cm*11.25cm; the sub-airbag material is medical-grade TPU, the inflation pressure range is 0-15kPa, and it has good flexibility and sealing; each sub-airbag is equipped with an independent air circuit interface, which is connected to the air pump and valve of each independent air circuit controlled by the mattress drive module through a silicone air tube, with 64*32 preset points, which are evenly distributed on the mattress surface in a grid pattern.

[0020] Multimodal sensing module: The pressure sensing array is set with 64*32 flexible thin film pressure sensing units (corresponding to the length-to-width ratio of the mattress) in a 2:1 row-to-column ratio, and is evenly distributed on the mattress surface in a grid pattern, corresponding to each preset point; the spacing between each flexible thin film pressure sensing unit is uniform, covering the entire area of ​​the mattress; the analog signal output by the flexible thin film pressure sensing unit is converted into a 12-bit digital signal by an ADC and transmitted to the control module; The temperature sensing array uses 64*32 temperature sensing units (digital temperature sensors), which are fully spatially aligned with the pressure sensing array (row-to-column ratio 2:1, sensor spacing consistent), and are deployed on the inner surface of the mattress. They communicate with the control module via a single bus protocol. The humidity sensor array uses 64*32 humidity sensing units (flexible humidity sensors), which are arranged on the same plane as the pressure and temperature sensing arrays and are completely spatially aligned. Preset area division: The mattress is divided into 16*8 non-overlapping preset areas (matching the number of sub-airbags). In this embodiment, rectangular areas of the same size are used. The vertex coordinates of each area are [4*i, 4*j], [4*i, 4*(j+1)], [4*(i+1), 4*j], [4*(i+1), 4*(j+1)], where i ranges from [0, 15] and corresponds to the area number in the long direction, and j ranges from [0, 7] and corresponds to the area number in the wide direction. Each preset area covers 4*4 pressure sensing units, 4*4 temperature sensing units, 4*4 humidity sensing units, and one sub-airbag. The area numbers and sub-airbag numbers are calibrated to establish a spatial mapping relationship.

[0021] Control module: The control module uses an Intel Core i7-12700H industrial PC with 8GB DDR5 memory, a 512GB solid-state drive, and runs Ubuntu 22.04 operating system. It is equipped with Python 3.9 and PyTorch 2.0 environments for algorithm deployment and data processing. The data fusion unit in the control module performs time alignment on the body pressure, temperature, and humidity data at each preset point. Based on the preset region division, it performs regional aggregation on the time-aligned body pressure, temperature, and humidity data of each preset point within each preset region to obtain the coupled three-dimensional vector of each preset region. The process of constructing a coupled data field of body pressure, temperature, and humidity based on the coupled three-dimensional vectors of all preset regions is as follows: 1. Time alignment: Step 1: Obtain the collection timestamps for body pressure, temperature, and humidity data respectively. The timestamp precision is 1ms; Step 2: Determine the minimum timestamp The timestamp is used as the base time. Step 3: Use linear interpolation to synchronize the time of other modal data. The interpolation formula is:

[0022] in, and The data values ​​are those acquired at adjacent times before and after the reference time, satisfying the following conditions: ; The preset threshold is that the time deviation of each data acquisition after time synchronization is less than or equal to 50ms. If the deviation exceeds this threshold, the data set will be discarded and re-acquisition will be triggered.

[0023] 2. Coupled data field generation: Based on a pre-defined region division, time-aligned body pressure, temperature, and humidity data are aggregated by region. The coupled three-dimensional vector for a single pre-defined region is... The calculation method is as follows:

[0024]

[0025]

[0026] in, This is the average value of the values ​​collected by 16 pressure sensing units within the preset area. This is the average value of the values ​​collected by 16 temperature sensing units within the preset area. This is the average value of the humidity data collected by 16 humidity sensing units within the preset area. The first one in the preset area respectively The system collects data from pressure, temperature, and humidity sensors; based on the above process, it obtains coupled three-dimensional vectors for all preset regions; and finally constructs a 16*8*3 dimensional coupled data field of body pressure, temperature, and humidity, with each preset region corresponding to a coupled three-dimensional vector.

[0027] Risk Identification Unit: The risk identification model adopts a convolutional neural network model with an encoder-decoder architecture. Encoder: 4 convolutional layers (Conv1, Conv2, Conv3, Conv4, all with 3x3 kernels, stride 1, padding=1), 2 max-pooling layers (Pool1, Pool2: 2x2 kernels, stride 2, located after Conv2 and Conv4 respectively); Decoder: 4 deconvolutional layers (DeConv1, DeConv2, DeConv3, DeConv4), 2 upsampling layers (Up1, Up2: sampling factor 2, located before DeConv1 and DeConv3 respectively); Activation Function: ReLU function for hidden layers, Sigmoid function for output layer. Training process: Based on multimodal data of 500 clinical bedridden patients (including high-risk areas of pressure ulcers), the data was divided into training, validation and test sets in a 7:2:1 ratio; the cross-entropy loss function and Adam optimizer were used (learning rate 0.001, 100 iterations, batch size 32); input and output: the input is a coupled data field of body pressure, temperature and humidity, and the output is a 16*8-dimensional risk probability vector (each value in the vector is in the range [0,1], representing the risk probability of the corresponding preset area).

[0028] The dynamic risk zoning map generation process is as follows: Threshold segmentation: A preset risk threshold of 0.7 is used to filter out regions with a risk probability greater than or equal to 0.7, marking them as candidate high-risk regions; Connected component clustering: The preset neighborhood rule is an 8-neighbor connected component rule, specifically: if a candidate high-risk region (a,b) and any one or more of its eight neighboring regions (a±1,b), (a,b±1), and (a±1,b±1) are also candidate high-risk regions, then these candidate high-risk regions are grouped into the same connected component, forming a complete high-risk region; if a candidate high-risk region has no neighboring regions... Candidate high-risk areas are treated as independent high-risk areas. Map output: The final generated dynamic risk zoning map includes a unique number for each high-risk area, boundary coordinates (minimum / maximum area numbers a_min and a_max in the long direction, and minimum / maximum area numbers b_min and b_max in the wide direction), risk probability value (if it is a complete high-risk area, the risk probability value is the average risk probability of each candidate high-risk area within that high-risk area), and a list of preset area numbers corresponding to each high-risk area, providing clear spatial positioning basis for the control strategy unit.

[0029] In the control module, the control strategy unit maps the preset area numbers contained in the high-risk area to the corresponding sub-airbag numbers according to the spatial mapping relationship, thereby determining the target sub-airbag; the preset linear mapping function is: Reference parameter: Reference pressure The pressure is set to 5 kPa (the maintenance pressure of the sub-inflator under normal conditions, which can be adjusted according to the individual patient's comfort level), and the adjustment coefficient is [not specified]. Take 3 kPa; if the risk probability value of a high-risk area If the value is 0.8, then the current pressure value is as described above. The calculated result is 2.6 kPa.

[0030] The inflation / deflation rates are determined as follows, and the pressure difference is calculated: ,in The current pressure value of the sub-airbag is collected by the pressure feedback sensor; The preset range and speed correspondence rules are as follows: When At that time, the inflation / deflation rate is taken as 0.5 kPa / s; when At that time, the inflation / deflation rate is taken as 1.0 kPa / s; when At that time, the inflation / deflation rate is taken as 1.5 kPa / s.

[0031] Feedback optimization module: The risk reduction rate is calculated as follows, with a preset duration of 60 seconds: 60 seconds after the mattress drive module performs the intervention, data is re-collected; the reduction rate formula is:

[0032] in The risk probability value for high-risk areas before intervention. This is the risk probability value recalculated after intervention.

[0033] The parameter adjustment rules are as follows: preset decrease rate threshold: 30%; parameter adjustment increment is the adjustment coefficient for each adjustment. Increase by 0.5; Constraints: Adjusted The pressure should be less than or equal to 5 to avoid the patient's comfort being affected by the expected pressure value being lower than 1 kPa; if the rate of decrease still does not meet the target after 3 adjustments, an alarm signal should be sent to prompt manual inspection.

[0034] Mattress drive module: The hardware configuration includes 128 miniature DC air pumps, one for each sub-airbag, operating at 12V, with a maximum inflation pressure of 20kPa and a flow rate of 1L / min, meeting the requirements for rapid inflation and deflation. It also includes 128 normally closed solenoid valves, one for each air pump, operating at 12V, used to control the airflow. The pressure feedback sensor system uses 128 miniature pressure sensors, one built into each sub-airbag, with a range of 0-20kPa. The mattress drive module receives control commands from the control module (target sub-airbag number, target pressure value, and inflation / deflation rate), controlling the corresponding air pumps and solenoid valves to open and inflate / deflate at the set rate.

[0035] The system workflow is as follows: 1. Initialization: After the system starts, each module completes self-check, the multimodal perception module starts to collect data at a preset frequency, and the control module loads preset parameters (model weights, thresholds, etc.). 2. Data Acquisition and Fusion: The multimodal sensing module synchronously acquires body pressure, temperature, and humidity data, and the data fusion unit completes time alignment to generate a 16*8 dimensional coupled data field of body pressure, temperature, and humidity. 3. Risk Identification: The convolutional neural network model analyzes the coupled data field, outputs a risk probability vector, and generates a dynamic risk partition map through threshold segmentation and clustering; 4. Control command generation: The control strategy unit maps the target sub-airbag, calculates the desired pressure value and inflation / deflation rate, and generates control commands; 5. Pressure reduction execution: The mattress drive module responds to commands and controls the inflation and deflation of the target sub-airbags, with pressure feedback sensors providing real-time closed-loop control; 6. Feedback and Optimization: Data is collected again 60 seconds after intervention, the risk reduction rate is calculated, and the adjustment coefficient is adjusted as needed. This forms a closed-loop optimization.

[0036] The above are merely embodiments of the present invention. Commonly known structures and characteristics are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning, characterized in that: It includes a smart mattress module, a multimodal sensing module, a control module, and a mattress drive module; The smart mattress module includes an airbag execution array, which includes multiple sub-airbags that can be independently controlled to inflate and deflate. Each sub-airbag in the airbag execution array is equipped with an independent pressure feedback sensor to collect the current pressure value of the sub-airbag in real time. The array is divided into preset areas corresponding to the number of sub-airbags according to the distribution of the sub-airbags, and the preset areas correspond one-to-one with the sub-airbags. The multimodal sensing module includes a pressure sensing array, a temperature sensing array, and a humidity sensing array. The pressure sensing array includes multiple pressure sensing units for real-time collection of body pressure data at preset points on the contact surface between the bedridden person and the mattress. The temperature sensing array includes multiple temperature sensing units for contact-based collection of temperature data at preset points on the contact surface between the bedridden person's body and the mattress. The humidity sensing array includes multiple humidity sensing units for contact-based collection of humidity data at preset points on the contact surface between the bedridden person's body and the mattress. The pressure sensing array, temperature sensing array, and humidity sensing array are arranged in spatial alignment. The control module includes: a data fusion unit, used to perform time alignment on body pressure, temperature, and humidity data at each preset point; based on preset region division, to perform regional aggregation on the time-aligned body pressure, temperature, and humidity data at each preset point within each preset region to obtain a coupled three-dimensional vector for each preset region; and to construct a body pressure, temperature, and humidity coupled data field based on the coupled three-dimensional vectors of all preset regions; and a risk identification unit, which has a built-in risk identification model, used to analyze the body pressure, temperature, and humidity coupled data field and output a dynamic risk zoning map, which identifies high-risk areas with risk probabilities exceeding a preset risk threshold and their corresponding risk probability values. The control strategy unit is used to generate control instructions for one or more specific sub-airbags in the airbag execution array based on the dynamic risk zoning map and the current pressure value of each sub-airbag. The control instructions include a target pressure value and an adjustment speed. The mattress drive module is used to control the air pumps and valves of the corresponding sub-airbags according to the control instructions, and to perform local pressure regulation operations.

2. The pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning according to claim 1, characterized in that: The risk identification model includes a convolutional neural network model, whose input is a coupled data field of body pressure, temperature and humidity, and whose output is a risk probability vector containing risk probability values ​​of each preset region. The dynamic risk partition map is obtained by threshold segmentation and connected component clustering of the risk probability vector; The threshold segmentation logic includes: setting a preset risk threshold, filtering out preset regions with risk probabilities not lower than the risk threshold based on the risk probability vector, and marking them as candidate high-risk regions; the connected component clustering logic includes: judging the adjacency relationship between candidate high-risk regions based on preset neighborhood rules, clustering adjacent candidate high-risk regions into a complete high-risk region, whose risk probability value is the weighted average of the risk probability values ​​of each preset region that makes up the complete high-risk region; non-adjacent candidate high-risk regions are treated as independent high-risk regions, and their risk probability value is the risk probability value of the corresponding preset region.

3. The pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning according to claim 2, characterized in that: The logic of the control strategy unit generating control commands includes: mapping the location of high-risk areas in the dynamic risk zoning map to the corresponding target sub-airbags; calculating the expected pressure value of the target sub-airbags based on the risk probability value of the high-risk areas using a preset mapping function, wherein the preset mapping function is a linear mapping function and the expected pressure value is negatively correlated with the risk probability value; calculating the pressure difference between the current pressure of the target sub-airbags and the expected pressure value; and determining the inflation / deflation rate according to a preset range and speed correspondence rule based on the preset difference range to which the pressure difference belongs.

4. The pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning according to claim 3, characterized in that: The method for time alignment by the data fusion unit includes: acquiring the collection timestamps of body pressure data, temperature data, and humidity data from all preset points, using the earliest collection timestamp as the benchmark; and using linear interpolation to synchronize the time of other data, ensuring that the time deviation of each data collection after time synchronization does not exceed a preset time threshold.

5. The pressure ulcer prevention mattress control system based on multimodal perception and dynamic risk zoning according to claim 4, characterized in that: It also includes a feedback optimization module, which is used to re-collect data through the pressure sensing array and multimodal sensing module within a preset time after the mattress drive module performs local pressure adjustment operation, and calculate the risk probability value reduction rate of the original high-risk area; if the reduction rate is lower than the preset reduction rate threshold, the mapping function parameters in the control strategy unit are gradually adjusted according to the preset parameter adjustment range until the reduction rate reaches the preset reduction rate threshold.

Images