Dynamic self-adaptive air suspension mattress and control method

Abstract

The invention discloses a dynamic self-adaptive air suspension mattress and a control method, and relates to the technical field of smart home, the mattress comprises: a data acquisition module for acquiring body pressure data of the mattress and a physiological time sequence signal of a target user; the intelligent processing module is used for generating human body supporting characteristics according to the body pressure data, generating a health assessment result according to the physiological time sequence signal, associating the human body supporting characteristics with the health assessment result and generating an adjusting instruction; the adjusting module is used for respectively controlling a plurality of air bags of the mattress to be inflated and deflated according to the adjusting instruction. According to the embodiment of the invention, health monitoring of the user is realized while the sleep quality is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of smart home, and particularly to a dynamically adaptive air suspension mattress and a control method thereof. Background Art

[0002] During the airbag adjustment of traditional mattresses, only human body support information (such as the pressure distribution of various body parts, anatomical landmark points, etc.) may be considered, or only health assessment information (such as physiological indicators like heart rate, respiratory rate, etc.) may be considered. The mattress cannot comprehensively consider the mutual influence and correlation between these two aspects during adjustment.

[0003] Specifically, the Chinese patent with the publication number "CN113558427A" discloses a method for adjusting the airbag of a mattress, which obtains multiple pressure values of multiple airbags when a human body lies on the mattress; obtains a non-ergonomic curve based on the multiple pressure values; adjusts the multiple pressure differences corresponding to the multiple airbags to adjust the non-ergonomic curve into an ergonomic curve; and obtains a recommended softness and hardness based on the multiple pressure differences and user basic information. It fails to consider health assessment information. Therefore, health risks are easily overlooked when adjusting according to human body support information.

[0004] Traditional mattresses have limitations in terms of supportiveness and health monitoring. They cannot adjust the support characteristics in real time according to the user's physical condition and sleep stage, and it is difficult to meet the high requirements of users for sleep quality and health monitoring. For example, when a user has body movements during sleep, resulting in a change in the pressure at the lumbar spine, a traditional mattress may only make simple support adjustments based on the pressure change, rather than making more reasonable adjustment decisions in combination with the user's health condition (such as whether there is a risk of apnea). When a traditional mattress attempts to optimize human body support, potential risks in health assessment may be overlooked. For example, overemphasizing strong support for the waist may cause the user to have difficulty breathing, especially in users at risk of respiratory diseases, and this conflict will further exacerbate the decline in the adjustment effect. On the contrary, if the mattress focuses on the intervention of health assessment information, such as adjusting the inclination angle of the mattress to improve breathing, it may lead to insufficient support for some parts of the body, causing physical discomfort and pain.

[0005] Therefore, how to achieve health monitoring of users while improving sleep quality has become an urgent technical problem to be solved. Summary of the Invention

[0006] The technical problem solved by the present invention is that traditional mattresses have limitations in terms of supportiveness and health monitoring, cannot adjust the support characteristics in real time according to the user's physical condition and sleep stage, and it is difficult to meet the high requirements of users for sleep quality and health monitoring.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, a dynamic adaptive air suspension mattress, the mattress comprising: a data acquisition module for acquiring the body pressure data of the mattress and the physiological time series signal of the target user; an intelligent processing module for generating a human body support feature according to the body pressure data, generating a health assessment result according to the physiological time series signal, associating the human body support feature and the health assessment result and generating an adjustment instruction; an adjustment module for controlling the inflation and deflation of a plurality of airbags of the mattress respectively according to the adjustment instruction.

[0008] Preferably, the human body support feature includes a pressure gradient tensor and the coordinates of anatomical landmark points; the health status assessment result is used to characterize the joint distribution of the sleep stage and the disease risk.

[0009] Preferably, the intelligent processing module includes a convolutional neural network, and the convolutional neural network sequentially includes an input layer, a convolutional layer, a pooling layer and an output layer; the input layer is used for normalizing the body pressure data into tensor data; the convolutional layer is used for performing convolutional processing on the tensor data; the pooling layer is used for performing downsampling processing on the tensor data obtained by convolutional processing according to spatial pyramid pooling; the output layer is used for processing the tensor data after downsampling processing to obtain the human body support feature.

[0010] Preferably, the convolutional layer includes a deformable convolutional kernel; the convolutional neural network further includes a fusion layer provided between the input layer and the convolutional layer; the input layer is further used for receiving the input of the biometric features and medical image data of the target user, mapping the biometric features into low-dimensional biometric feature vectors, and extracting the medical image feature tensors from the medical image data; the fusion layer is used for splicing the biometric feature vectors, the medical image feature tensors and the body pressure data to obtain a spliced image, and generating an offset field according to the spliced image; the deformable convolutional kernel is used for performing convolutional processing on the tensor data according to the offset field.

[0011] Preferably, the intelligent processing module further includes a long short-term memory recurrent neural network, and the long short-term memory recurrent neural network is used for adjusting the weight coefficient of the forgetting gate of the long short-term memory recurrent neural network according to the classification result of the sleep stage.

[0012] Preferably, the sleep stage includes a non-rapid eye movement period and a rapid eye movement period, and the weight coefficient of the forgetting gate corresponding to the non-rapid eye movement period is lower than the weight coefficient of the forgetting gate corresponding to the rapid eye movement period.

[0013] Preferably, the intelligent processing module further includes a fusion unit, and the fusion unit is used for: obtaining an attention weight matrix by associating the human body support feature and the health assessment result through a cross-attention formula; performing weighted fusion on the human body support feature and the health assessment result according to the attention weight matrix to generate a multi-dimensional adjustment weight vector; generating an adjustment instruction according to the multi-dimensional adjustment weight vector.

[0014] Preferably, the adjustment module is configured to determine a target adjustment airbag according to an adjustment instruction, determine airbags within a preset distance of the target adjustment airbag as cooperative airbags, and adjust the inflation and deflation of the target adjustment airbag and the cooperative airbags according to the adjustment instruction.

[0015] Preferably, the adjustment module is further configured to deflate an overpressure airbag in response to detecting that the pressure of the overpressure airbag exceeds a preset safety threshold within a preset time period.

[0016] In a second aspect, the present invention provides a control method for a dynamically adaptive air suspension mattress, the control method including: acquiring body pressure data of the mattress and physiological timing signals of a target user; generating a human body support feature according to the body pressure data, generating a health assessment result according to the physiological timing signals, associating the human body support feature and the health assessment result and generating an adjustment instruction; respectively controlling the inflation and deflation of a plurality of airbags of the mattress according to the adjustment instruction.

[0017] Advantages of the present invention: By generating a human body support feature according to the body pressure data, generating a health assessment result according to the physiological timing signals, associating the human body support feature and the health assessment result and generating an adjustment instruction, and respectively controlling the inflation and deflation of a plurality of airbags of the mattress according to the adjustment instruction, it is possible to comprehensively consider the needs of sleep quality and health monitoring when adjusting the inflation and deflation of the airbags, thereby realizing health monitoring of the user while improving sleep quality. Description of the Drawings

[0018] Figure 1 It is a system structure diagram of the dynamically adaptive air suspension mattress provided by the embodiment of the present application. Detailed Embodiments

[0019] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed embodiments of the present invention with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments.

[0020] The dynamically adaptive air suspension mattress provided by the embodiment of the present application is applicable to sleep health management. In a sleep scenario, the mattress can adapt to the physical and physiological changes of the user in real time to improve sleep quality and monitor health conditions; the mattress is also applicable to medical rehabilitation assistance. In a medical environment, the mattress can adjust the personalized support and pressure distribution according to the physical condition and rehabilitation needs to promote rehabilitation and prevent complications.

[0021] Please refer to Figure 1 , Figure 1 which is a system structure diagram of the dynamically adaptive air suspension mattress provided by the embodiment of the present application. The mattress includes:

[0022] A data acquisition module for obtaining the body pressure data of the mattress and the physiological time series signals of the target user.

[0023] Among them, the mattress includes an N×M independently controllable airbag matrix. Each airbag is equipped with a piezoelectric film pressure sensor and a micro solenoid valve. The inflation and deflation of the airbag are controlled by devices such as an air pump and a solenoid valve to achieve differential adjustment in multiple regions. Biomechanical pressure arrays, inertial measurement units, and environmental sensors are provided on the airbag to obtain the body pressure data of the mattress and the physiological time series signals of the target user for forming the body pressure data.

[0024] Among them, the pressure data is the pressure exerted by the target user on the mattress when the mattress supports the target user. The pressure data is in the form of a distribution map of pressures at various locations on the mattress. The physiological time series signals include physiological parameter time series signals (such as heart rate variability, breathing rhythm, etc.), and further may include: the turning frequency of the target user, the posture conversion interval, and environmental parameters (such as temperature and humidity fluctuations, noise duration, etc.).

[0025] An intelligent processing module for generating human support features based on the body pressure data, generating a health assessment result based on the physiological time series signals, associating the human support features and the health assessment result, and generating an adjustment instruction.

[0026] Among them, the intelligent processing module sequentially includes a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory Recurrent Neural Network).

[0027] Specifically, the CNN sequentially includes an input layer, a convolutional layer, a pooling layer, and an output layer. The input layer is used to normalize the body pressure data (pressure matrix) into a tensor of 0-1. The convolutional layer may be provided with multiple convolutional layers. For example, 3 convolutional layers, and each convolutional layer uses a 3×3 convolutional kernel with a stride of 2 to adapt to the deformation of the human body contour. The pooling layer selects spatial pyramid pooling (SPP, 4-level downsampling) to retain key regions. A body support demand heat map containing 9-dimensional anatomical partition weight data (sacrum / lumbar vertebra / thoracic vertebra, etc.) is output in the output layer, that is, the human support features are obtained. The human support features include a pressure gradient tensor and anatomical landmark coordinates.

[0028] The pressure gradient tensor is a tensor representing the pressure change trend formed by calculating the pressure gradient between regions of the mattress based on the body pressure data. This tensor contains the change rate and direction information of the pressure in different directions, which helps to determine the parts that need to be supported and the pressure relieved.

[0029] The anatomical landmark coordinates are the projection coordinates of the key anatomical landmarks (such as the acromion, lumbar vertebra, knee joint, etc.) of the body on the pressure distribution map (mattress) identified by analyzing the distribution pattern and characteristics of the body pressure data and combining human anatomy knowledge, providing a position reference for precise adjustment.

[0030] Furthermore, traditional mattresses cannot dynamically adjust the support method according to individual differences of users (such as body type, disease status). Therefore, preferably, the convolutional layer of the CNN model is a deformable convolution layer (Deformable Convolution), and the deformation amount of its convolution kernel can be adjusted according to the biometric features of the target user obtained from user input or external devices (such as body fat scales, height and weight meters, etc.), or can also be adjusted according to the medical imaging data of the user. Biometric features, for example, basic information such as age, gender, height, weight, body type, etc., can reflect the individual differences of users and provide a basis for network parameter adjustment; the medical imaging data of users, such as X-rays, CTs, MRIs, etc., can provide detailed information about the internal structure of the user's body, especially the conditions of parts closely related to mattress support, such as bones, muscles, joints, etc.

[0031] Specifically, for obese users (high BMI), the biometric features trigger the offset field to expand the sampling range of the convolution kernel at the shoulders / hips, enhancing the pressure gradient detection; for the case where the medical imaging shows lordosis of the lumbar spine, it will cause the offset field to adjust the convolution kernel to focus on the lumbar spine area, enhancing the local support weight; for patients with lumbar disc herniation, the medical imaging drives the offset field to generate a "concave" sampling pattern in the lumbar spine area, enabling the convolution kernel to identify the abnormal pressure distribution that needs decompression.

[0032] In a traditional convolutional neural network (CNN), the geometric shape of the convolution kernel (such as a 3×3 square) and the sampling position are fixed and independent of the input content, which limits the adaptability of the model to complex scenarios such as human body contour deformation and personalized anatomical features. Therefore, the introduction of a deformable convolution kernel (Deformable Convolution) realizes the bidirectional coupling of the geometric structure of the convolution kernel and the input data through dynamic deformation amounts (offsets).

[0033] The LSTM model includes an input gate, a forget gate, a cell state, and an output gate. The input gate is used to receive the physiological time-series signals of the target user and screen the physiological time-series signals to determine the part of the current input signal to be added to the cell state. Specifically, for physiological signals, the input gate can screen out the features related to the health status, such as body movements or heart rate variability patterns within a specific frequency range; the forget gate is used to dynamically adjust the historical memory weights. The forget gate is used to determine how much information in the cell state at the previous moment needs to be forgotten. In sleep health assessment, the forget gate can help the LSTM model ignore irrelevant long-term information, such as past irrelevant body movement data, which may have less impact on the current health status assessment; the cell state is used to store the long-term sleep stage features. It is updated through the combined action of the forget gate and the input gate, retaining both important historical information and adding relevant features at the current moment. At each time step, the cell state first multiplies element-wise with the cell state at the previous moment according to the output of the forget gate to achieve the forgetting of old information, and then updates the cell state by adding element-wise the output of the input gate and the candidate state (new information calculated from the current input and the previous hidden state); the output gate is used to generate the health status assessment result and also to determine how much information in the cell state should be passed to the hidden state at the next moment or be used as the model output at the current moment. The final output (hidden state) of the LSTM model contains the screened and integrated physiological signal features, which are closely related to the health status. Through one or more fully connected layers, the hidden state is mapped to the evaluation dimensions of the health status, such as the joint distribution of sleep stages (e.g., wakefulness, light sleep, deep sleep, rapid eye movement period) and disease risks (e.g., sleep apnea, insomnia, etc.). In a classification task, the softmax activation function can be used to convert the output into a probability distribution, representing the likelihood of different health states or disease risks. For example, the output result may show that the probability of belonging to the deep sleep stage at the current moment is 80%, and the risk of having sleep apnea is 10%.

[0034] At different stages, the physiological state of the human body and the sensitivity to the environment, etc. are all different. According to the activity characteristics of the brain and body during sleep, two main stages are divided, the rapid eye movement period and the non-rapid eye movement period. During the non-rapid eye movement period, the physiological state of the human body is relatively stable, and new physiological data can better reflect the current health status. The weight coefficient of historical memory can be set lower than that in the rapid eye movement period, so that the LSTM pays relatively less attention to the previously stored historical health data and focuses more on the newly input physiological data at the current moment; while in the rapid eye movement period, due to the complex and changeable physiological activities, more historical data needs to be combined for comprehensive judgment.

[0035] Associate human body support features with health assessment results and generate adjustment instructions. To splice the spatial feature vector (output of CNN) and the temporal feature vector (output of LSTM) in the fusion layer, the resulting vector contains information in both space and time. The fully connected layer maps the fused comprehensive feature vector to the dimension of the control instruction. The control instruction can be a parameter for adjusting the sleep environment, such as adjusting the hardness of the mattress, the height of the pillow, etc. The fully connected layer generates specific adjustment instructions by learning the mapping relationship between the comprehensive feature and the control instruction. For example, if the model finds that the current sleep stage is deep sleep and the body pressure distribution is uneven, it may generate an instruction to adjust the mattress to provide better support.

[0036] In another alternative approach, the intelligent processing module further includes a fusion unit, and the fusion unit is used for: obtaining an attention weight matrix by associating human body support features and health assessment results through the cross-attention formula; performing weighted fusion on the human body support features and health assessment results according to the attention weight matrix to generate a multi-dimensional adjustment weight vector; generating adjustment instructions according to the multi-dimensional adjustment weight vector.

[0037] Among them, the human body support features (spatial dimension) and the health assessment results (temporal dimension) are fused through the heterogeneous tensor cross-attention formula, and the mathematical expression of the cross-attention formula is:

[0038]

[0039] Preferably, the adjustment module is used to determine the target adjustment airbag according to the adjustment instruction, determine the airbags within the preset distance of the target adjustment airbag as cooperative airbags, and adjust the inflation and deflation of the target adjustment airbag and the cooperative airbags according to the adjustment instruction to avoid the problem of discontinuous support caused by independent airbag control (such as the neighborhood collapse when the waist is inflated). For example, set adjacent 3×3 airbags as a cooperative group; the difference in inflation and deflation volume within the group is limited to ±10% (such as the main airbag is inflated by 1.0 kPa, and the neighborhood is inflated by 0.9 - 1.1 kPa).

[0040] Preferably, the adjustment module is further used to respond to detecting that the pressure of the overpressure airbag exceeds the preset safety threshold within a preset time period, and deflate the overpressure airbag to avoid the risk of bursting caused by airbag overpressure. For example, if the pressure > 8 kPa lasts for 2 seconds, then deflate it.

[0041] In a second aspect, an embodiment of the present application further provides a control method for a dynamically adaptive air suspension mattress. The control method includes: obtaining the body pressure data of the mattress and the physiological time series signal of the target user; generating human body support features according to the body pressure data, generating health assessment results according to the physiological time series signal, associating the human body support features and the health assessment results and generating adjustment instructions; respectively controlling the inflation and deflation of multiple airbags of the mattress according to the adjustment instructions.

[0042] In the embodiments of the present application, by generating a human body support feature according to body pressure data, generating a health assessment result according to a physiological time series signal, associating the human body support feature and the health assessment result and generating an adjustment instruction, and respectively controlling a plurality of air bags of a mattress to inflate and deflate according to the adjustment instruction, it is possible to comprehensively consider the needs of sleep quality and health monitoring when adjusting the inflation and deflation of the air bags, so as to realize health monitoring of users while improving sleep quality.

[0043] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc. These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, and the instruction device implements the functions specified in one process Figure 1 one process or multiple processes and / or Figure 1 boxes or multiple boxes.

[0044] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A dynamically adaptive air suspension mattress, characterized in that: The mattress comprises: A data acquisition module, used to obtain body pressure data of the mattress and physiological time series signals of the target user; an intelligent processing module, for generating a human body support feature according to the body pressure data, generating a health assessment result according to the physiological time series signal, associating the human body support feature with the health assessment result and generating an adjustment instruction; The regulating module is used to control the inflation and deflation of the multiple air bags of the mattress according to the regulating instructions.

2. The mattress according to claim 1, characterized in that: The human body support characteristics include a pressure gradient tensor and anatomical landmark coordinates; the health status assessment result is used to characterize the joint distribution of sleep stages and disease risks.

3. The mattress according to claim 2, characterized in that: The intelligent processing module includes a convolutional neural network, which includes an input layer, a convolution layer, a pooling layer and an output layer in sequence; The input layer is used to normalize the body pressure data into tensor data; The convolution layer is used to perform convolution processing on the tensor data; The pooling layer is used to downsample the tensor data processed by convolution according to spatial pyramid pooling; The output layer is used to process the tensor data after the downsampling process to obtain the human body support features.

4. The mattress according to claim 3, characterized in that: The convolution layer includes a deformable convolution kernel; the convolution neural network also includes a fusion layer arranged between the input layer and the convolution layer; The input layer is further used to receive the input of the biometric features and medical image data of the target user, and map the biometric features into low-dimensional biometric feature vectors, and perform feature extraction on the medical image data to obtain a medical image feature tensor; The fusion layer is used to splice the biological feature vector, the medical image feature tensor and the body pressure data to obtain a spliced ​​image, and generate an offset field according to the spliced ​​image; The deformable convolution kernel is used to perform convolution processing on the tensor data according to the offset field.

5. The mattress according to claim 2, characterized in that: The intelligent processing module also includes a long short-term memory recursive neural network, which is used to adjust the weight coefficient of the forget gate of the long short-term memory recursive neural network according to the classification result of the sleep stage.

6. The mattress according to claim 5, characterized in that: The sleep stages include non-rapid eye movement (NREM) and rapid eye movement (REM) periods, and the weight coefficient of the forget gate corresponding to the non-REM period is lower than the weight coefficient of the forget gate corresponding to the REM period.

7. The mattress according to claim 2, characterized in that: The intelligent processing module further comprises a fusion unit, which is used for: Associating the human body support feature and the health assessment result through a cross attention formula to obtain an attention weight matrix; Performing weighted fusion on the human body support feature and the health assessment result according to the attention weight matrix to generate a multi-dimensional adjustment weight vector; The adjustment instruction is generated according to the multi-dimensional adjustment weight vector.

8. The mattress according to claim 1, characterized in that: The adjustment module is used to determine a target adjustment airbag according to the adjustment instruction, determine an airbag within a preset distance of the target adjustment airbag as a cooperative airbag, and adjust the target adjustment airbag and the cooperative airbag to inflate and deflate according to the adjustment instruction.

9. The mattress according to claim 1, characterized in that: The regulating module is further configured to deflate the overpressure airbag in response to detecting that the pressure of the overpressure airbag exceeds a preset safety threshold within a preset time period.

10. A control method for a dynamically adaptive air suspension mattress, characterized in that: The control method comprises: Acquiring body pressure data of the mattress and physiological time series signals of the target user; Generate a human body support feature according to the body pressure data, generate a health assessment result according to the physiological time series signal, associate the human body support feature with the health assessment result and generate an adjustment instruction; The multiple air bags of the mattress are controlled to inflate and deflate respectively according to the adjustment instructions.

Images