Pension service supervision method and system based on image recognition AI model algorithm

Abstract

The invention provides a pension service supervision method and system based on an image recognition AI model algorithm. The method comprises the following steps: aiming at respiratory arrest, pressure sores and limb stiffness risks in a bedridden old-age care scene, firstly, establishing a physiological abnormality library containing parameters such as a respiratory rate threshold value and a body pressure critical value; a physical sign monitoring device is deployed to collect body pressure hot area distribution data of the bedridden person in real time, breathing interval abnormity, heartbeat attenuation and body pressure exceeding areas are analyzed through an AI model, and a retention risk index is generated. And when the monitoring data exceeds an abnormal library threshold value, triggering multi-modal early warning information, fusing risk area positioning, abnormal duration and temperature gradient change, outputting a multi-modal early warning report combining images and texts, and guiding the nursing personnel to perform targeted intervention. According to the technical scheme provided by the invention, real-time monitoring and multi-mode risk early warning of health hidden dangers in bed are realized, the occurrence rate of complications such as respiratory arrest and pressure sores is reduced, and the precision and response efficiency of nursing for the aged are improved.

Description

Technical Field

[0001] The present application relates to the field of AI model algorithm technology, and in particular to a method and system for supervising elderly care services based on image recognition AI model algorithm. Background Art

[0002] In bedridden elderly care scenarios, the elderly are prone to health risks such as respiratory decline, pressure sores caused by local tissue compression, and muscle atrophy caused by limited limb movement due to long-term bed rest. Such scenarios urgently require a technical solution that can monitor multi-dimensional physiological indicators (such as respiratory rate, body pressure distribution, and limb mobility) in real time to achieve early warning of respiratory arrest, pressure sore formation, and limb stiffness. The technology must meet core requirements such as non-invasive monitoring, multimodal data fusion, and high-precision anomaly detection, and be able to operate stably in complex environments (such as obstruction by bedding and interference from body movement).

[0003] Currently, the typical technical solution for bedridden health monitoring is a localized pressure monitoring system based on body pressure distribution sensors. This system deploys a high-density array of pressure sensors on the mattress surface to collect real-time pressure distribution data from bedridden individuals. It then uses a threshold algorithm to identify localized high-pressure areas (such as the sacrum and heels, which are high-risk areas for pressure ulcers) and uses heatmap visualization to prompt caregivers to adjust body position. Some systems can also predict the level of pressure ulcer risk based on pressure trends.

[0004] Existing solutions rely solely on body pressure data and are unable to synchronously monitor key vital signs such as respiration and heartbeat, resulting in missed risks such as respiratory arrest and cardiopulmonary dysfunction. Body pressure data lacks fusion analysis with physiological signals (such as respiration and heartbeat), making it difficult to establish a correlation model between body pressure distribution and respiratory depression and limb rigidity, limiting the accuracy of early warnings. Static pressure thresholds cannot reflect dynamic risk fluctuations caused by changes in body position and individual differences, and can easily reduce system credibility due to false alarms / missed alarms. Summary of the Invention

[0005] This application provides a method and system for supervising elderly care services based on an image recognition AI model algorithm, which is used to solve the problems of low processing efficiency and poor accuracy of AI model algorithms in the existing technology.

[0006] In the first aspect, this application provides a method for supervising elderly care services based on an image recognition AI model algorithm, including:

[0007] Establish a database of abnormal physiological parameters based on the health risks of respiratory arrest, pressure sores, and limb rigidity in bedridden elderly care scenarios;

[0008] The vital sign monitoring device deployed in the bedridden area collects the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person, and matches the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory frequency threshold value of the physiological parameter abnormality library;

[0009] Obtaining the body pressure and thermal zone distribution data of the bedridden area, combining it with the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggering an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and thermal zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person;

[0010] Based on the detention risk index, the area of the body pressure distribution critical value exceeding the standard corresponding to the physiological parameter abnormality library is extracted, and combined with the duration of the abnormal breathing interval and the heartbeat intensity attenuation, the bed rest risk warning information is generated through the image recognition AI model;

[0011] Based on the bed rest risk warning information, a multimodal warning report is generated.

[0012] Optionally, the multiple key body regions include the sacrum, coccyx, hip, and heel regions; the spatial superposition rule of the body pressure and thermal zone distribution data and the respiratory micro-movement amplitude is used to trigger the image recognition AI model to analyze the multiple key body regions of the bedridden person and generate a retention risk index for the bedridden person, including:

[0013] Based on the continuous areas where the pressure value exceeds the set threshold in the body pressure and thermal zone distribution data, the time series integral of the pressure duration of each area is calculated, and the static pressure area with the integral value exceeding the first critical value is screened out;

[0014] Extract the peak-to-valley difference sequence of respiratory micro-movement amplitude within the sliding time window, calculate the variance of the difference sequence for the overlapping area of adjacent windows, and screen out low-activity areas with variance values below the second critical value;

[0015] Grid-coding and matching the spatial coordinates of the static pressure area with the spatial coordinates of the low-activity area, and selecting a composite risk area within the grid unit that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression;

[0016] Based on the mapping relationship between the composite risk area and the sacrum, hip, and heel regions in the human anatomical coordinate system, the image recognition AI model dynamically allocates the optical resolution enhancement ratio of the target area according to the preset weight coefficient;

[0017] Based on the enhanced image data, the skin texture deformation characteristics of the target area are extracted, and the retention risk index of each key area of the body is calculated.

[0018] Optionally, the grid-coding and matching of the spatial coordinates of the static pressure area and the spatial coordinates of the low activity area, and selecting a composite risk area in the grid unit that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression, includes:

[0019] Dividing the bed area into orthogonal grid cells of equal resolution, performing grid attribution mapping on the spatial coordinates of the static pressure area and the spatial coordinates of the low activity area, and screening grid cells that simultaneously meet the dual conditions of pressure integral exceeding limit and respiratory fluctuation suppression through Boolean logic operations;

[0020] Based on the spatial continuity of adjacent biconditional grid cells, discrete cells were merged through morphological dilation operations to generate a composite risk area covering the sacrum, hip, and heel mapping areas. The boundaries of the composite risk area were jointly calibrated by the pressure gradient mutation points and the respiratory amplitude jump points.

[0021] Optionally, the skin texture deformation characteristics include the contour curvature change rate and edge gradient direction distribution parameters of the compressed area;

[0022] The method of extracting the skin texture deformation characteristics of the target area based on the enhanced image data and calculating the retention risk index of each key area of the body includes:

[0023] The contour pixels of the compressed area of the target area in the enhanced image data are extracted, and the continuous closed boundaries of the compressed area are retained through multi-scale spatial filtering to generate a contour sequence of the compressed area;

[0024] Calculating the curvature difference values of adjacent contour point sets along the tangent direction of the contour sequence, performing sliding window statistics on the number of times the sign of the curvature difference values changes, and generating the contour curvature change rate of the compressed area;

[0025] In the inner area of the compressed area contour, along the normal path perpendicular to the contour direction, a frequency distribution histogram of the cosine value of the angle between the gradient amplitude direction and the normal direction is statistically analyzed, and the cumulative proportion of the cosine value of the angle in the histogram falling within a preset negative interval is calculated to extract the edge gradient direction distribution parameter;

[0026] According to the window statistical results of the contour curvature change rate and the cumulative proportion of the edge gradient direction distribution parameters, the retention risk index of each key area of the body is generated through a linear weighted fusion formula.

[0027] Optionally, the step of extracting, based on the retention risk index, areas of the physiological parameter abnormality library corresponding to critical values of body pressure distribution exceeding the standard, and combining the duration of the abnormal breathing interval and heartbeat intensity attenuation to generate bed rest risk warning information through the image recognition AI model includes:

[0028] According to the retention risk index, the body pressure distribution critical values corresponding to the sacrum, hip and heel regions are matched from the physiological parameter abnormality library, and the exceeding-standard areas whose pressure values continuously exceed the body pressure distribution critical values and whose spatial distribution area is larger than the set threshold are screened out;

[0029] Calculating the phase offset of adjacent respiratory cycles within the abnormal respiratory interval period, extracting a sequence of mutation points in the respiratory signal where the phase difference between adjacent peaks exceeds a preset angle, and statistically analyzing the density distribution of the mutation point sequence within the duration of the abnormal respiratory interval period;

[0030] In the body pressure thermal zone distribution data corresponding to the exceeding standard area, a temperature gradient profile line is extracted along the direction of maximum decrease of the body pressure gradient, and the covariance coefficient of the temperature decay rate and the heartbeat intensity decay rate within a unit distance on the profile line is calculated;

[0031] The density distribution and covariance coefficient are input into the image recognition AI model, and a multimodal fusion is performed on the geometric topological features of the exceeded area and the time-frequency domain features of the respiratory and heartbeat signals to generate a feature vector representing the intensity of pressure and physiological coupling;

[0032] According to the mapping relationship between each dimensional parameter in the feature vector and the preset risk level in the physiological parameter abnormality library, the local weight coefficient of the body pressure distribution critical value is dynamically adjusted to output bed rest risk warning information.

[0033] Optionally, the density distribution and covariance coefficient are input into the image recognition AI model, and multimodal fusion is performed on the geometric topological features of the exceeded area and the time-frequency domain features of the respiratory and heartbeat signals to generate a feature vector representing the intensity of pressure and physiological coupling, including:

[0034] Performing multi-scale decomposition on the geometric topological features of the exceeding-standard area, extracting the frequency domain distribution of the contour curvature change rate at different scales, and generating a multi-scale frequency domain descriptor of the geometric topological features;

[0035] Performing segmented analysis on the time-frequency domain features of the respiratory and heartbeat signals, extracting the time-frequency energy distribution of high-density areas in the density distribution, and generating a time-frequency energy descriptor of the respiratory and heartbeat signals;

[0036] Mapping the covariance coefficient to the geometric topological feature space of the exceeding area, calculating the local correlation between the temperature decay rate and the contour curvature change rate, and generating a temperature and geometry coupling descriptor;

[0037] Input the multi-scale frequency domain descriptor of the geometric topological features, the time-frequency energy descriptor of the respiratory and heartbeat signals, and the temperature and geometric coupling descriptor into the image recognition AI model, and generate an initial feature vector through the spatial alignment and weight distribution rules of the multimodal features;

[0038] The initial eigenvector is compressed in eigenspace, and the dimensional parameters with the highest correlation with the critical value of pressure distribution, density distribution, and covariance coefficient in the eigenvector are extracted to generate a eigenvector representing the intensity of pressure-physiological coupling.

[0039] Optionally, generating a multimodal warning report based on the bed rest risk warning information includes:

[0040] Based on the bed rest risk warning information, threshold calibration is performed on the time periods in the physiological parameter abnormality library that do not meet the body micro-movement cycle standard;

[0041] Based on the calibrated body pressure distribution critical value, the regional segmentation parameters of the body pressure heat zone distribution data are reconstructed. The pre-trained convolutional neural network is used to perform skin texture wrinkle depth recognition and erythema feature extraction on key areas of the body that continuously exceed the calibration threshold, generating a multimodal early warning report including abnormal respiratory waveforms, heat zone exceeding standard distribution and calibrated thresholds.

[0042] Secondly, this application provides a pension service supervision system based on an image recognition AI model algorithm, including:

[0043] Establish a module to build a physiological parameter abnormality library based on the health risks of respiratory arrest, pressure sore risks, and limb rigidity in bedridden elderly care scenarios;

[0044] A matching module collects the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person through a vital sign monitoring device deployed in the bedridden area, and matches the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory frequency threshold value of the physiological parameter abnormality library;

[0045] a generation module that obtains the body pressure and heat zone distribution data of the bedridden area, combines the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggers an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and heat zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person;

[0046] The generation module is further configured to extract, based on the retention risk index, areas of excess body pressure distribution corresponding to critical values in the physiological parameter abnormality library, and generate bed rest risk warning information through the image recognition AI model in combination with the duration of the abnormal breathing interval and the heartbeat intensity attenuation;

[0047] The generating module is further configured to generate a multimodal warning report based on the bed rest risk warning information.

[0048] In a third aspect, an embodiment of the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for supervising elderly care services based on an image recognition AI model algorithm as described in the first aspect above.

[0049] In a fourth aspect, an embodiment of the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for supervising elderly care services based on an image recognition AI model algorithm as described in the first aspect.

[0050] In an embodiment of the present application, a physiological parameter abnormality library is established based on the health risks of respiratory arrest, pressure ulcer risk and limb stiffness in bedridden elderly care scenarios; a vital sign monitoring device deployed in the bedridden area is used to collect heartbeat vibration signals and respiratory micro-movement amplitudes of bedridden persons, and based on the respiratory frequency threshold of the physiological parameter abnormality library, respiratory interval abnormality and heartbeat intensity attenuation characteristics in the monitoring signal are matched; body pressure thermal zone distribution data of the bedridden area is obtained, and combined with the respiratory micro-movement amplitude collected by the vital sign monitoring device, an image recognition AI model is triggered to analyze multiple key body areas of the bedridden person through the spatial superposition rule of body pressure thermal zone distribution data and respiratory micro-movement amplitude to generate a retention risk index for the bedridden person; based on the retention risk index, the area exceeding the standard corresponding to the critical value of the body pressure distribution in the physiological parameter abnormality library is extracted, and combined with the duration of the respiratory interval abnormality and heartbeat intensity attenuation, bedridden risk warning information is generated through the image recognition AI model; based on the bedridden risk warning information, a multimodal warning report is generated.

[0051] The technical solution of this application has the following beneficial effects:

[0052] By presetting indicators such as respiratory rate thresholds and body pressure distribution critical values, a standardized risk criterion library is constructed to provide a quantitative benchmark for subsequent monitoring. Non-contact sensors are used to capture heartbeat vibrations and respiratory micro-movement signals in real time, and combined with the abnormal library threshold to quickly identify abnormal respiratory intervals and cardiac function decline, thereby improving the sensitivity of early warning. By integrating body pressure thermal zones and respiratory micro-movement data, high-risk areas are located through spatial superposition rules, and AI models are used to analyze skin deformation characteristics to achieve accurate risk assessment of pressure ulcers and limb stiffness. The spatiotemporal relationship between physiological abnormalities and areas with excessive body pressure is dynamically associated to generate multi-dimensional risk warning information. Graphical data and risk level prompts are integrated to output visual reports to guide nursing staff in targeted interventions.

[0053] Furthermore, through the spatial coupling of pressure and respiratory data and dynamic resolution enhancement technology, skin micro-deformations and signs of tissue ischemia in areas with high incidence of pressure ulcers can be accurately identified, reducing the risk of misjudgment due to a single parameter, improving the accuracy of risk assessment of pressure ulcers and limb stiffness in bedridden people, and providing anatomical positioning basis for differentiated care.

[0054] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0056] Figure 1 A flowchart of a method for supervising elderly care services based on an image recognition AI model algorithm provided by this application is shown;

[0057] Figure 2 The figure shows a schematic diagram of the structure of an elderly care service supervision system based on an image recognition AI model algorithm provided by this application;

[0058] Figure 3 A schematic structural diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0061] This application aims to build a dynamic early warning system for the risks of respiratory arrest, pressure sores and limb stiffness in bedridden elderly care scenarios through multimodal data fusion and intelligent analysis technology. First, integrate clinical data and historical cases to establish a physiological abnormality library containing parameters such as respiratory rate thresholds and body pressure critical values as a risk judgment benchmark; secondly, deploy a non-contact vital sign monitoring device in the bed area to collect heartbeat vibration signals and respiratory micro-movement amplitudes in real time, and extract respiratory interval abnormalities and heartbeat intensity attenuation characteristics through signal filtering and time-frequency analysis algorithms; simultaneously obtain high-precision body pressure hot zone distribution data, combine the spatial distribution of respiratory micro-movement amplitudes, and use coordinate mapping rules to superimpose the body pressure exceeding standard area and the respiratory depression area, triggering the image recognition AI model to locate and analyze the complex risk area and generate a retention risk index; generate risk level classification and early warning information through a multi-threshold linkage mechanism; finally, integrate the abnormal area heat map, risk timeline and text summary to generate a multimodal early warning report combining pictures and texts to support precise intervention by nursing staff.

[0062] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0063] Figure 1 A flowchart of a method for supervising elderly care services based on an image recognition AI model algorithm is provided for an embodiment of the present application. Figure 1 As shown, the method includes:

[0064] 101. Establish a database of abnormal physiological parameters based on the health risks of respiratory arrest, pressure sores, and limb rigidity in bedridden elderly care scenarios;

[0065] Physiological Parameter Abnormality Database: This database includes respiratory rate thresholds (e.g., a normal respiratory rate is 12-20 breaths / minute; a rate below 8 is considered respiratory depression), body pressure distribution thresholds (e.g., sacral pressure >32 mmHg for two hours triggers pressure ulcer risk), and limb mobility thresholds (e.g., a joint range of motion <30° for four hours is considered rigidity). Health hazards include common risks associated with bed rest, such as respiratory arrest, pressure ulcers, and limb rigidity.

[0066] Based on clinical data and historical case analysis, we integrated parameters such as respiratory rate, body pressure distribution, and limb range of motion to construct a standardized abnormality library. First, by statistically analyzing the respiratory rate distribution of 1,000 bedridden patients, we set the apnea threshold to an interval greater than 10 seconds and the respiratory depression threshold to less than 8 breaths / minute. Combined with pressure sensing experimental data, we defined the sacral body pressure critical value as 32 mmHg (which triggers a pressure ulcer warning for 2 hours), and based on joint motion sensor data, we set the limb stiffness threshold to an activity angle less than 30° for 4 hours, thus establishing a physiological parameter abnormality library.

[0067] A nursing home established an abnormality database based on data from 100 long-term bedridden patients. The thresholds for respiratory rate were set at 8-24 breaths / minute, sacral pressure at 32 mmHg, and hip range of motion at 30°. When these data exceeded these limits, an alert was triggered.

[0068] 102. Using a vital sign monitoring device deployed in the bedridden area, collect the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person, and match the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory rate threshold value in the physiological parameter abnormality library;

[0069] Vital sign monitoring device: A non-contact piezoelectric sensor array or millimeter-wave radar is used to collect heartbeat vibration signals (such as heart rate) and respiratory micro-movement amplitude (such as the distance of chest and abdominal rise and fall). Abnormal breathing interval: The time between two breaths exceeds a preset threshold (such as >10 seconds).

[0070] A piezoelectric sensor array (sensitivity 0.1Pa) and a millimeter-wave radar (frequency band 60GHz) are deployed under the mattress to collect heartbeat vibration signals (main frequency 1-2Hz) and respiratory micro-movement amplitude (accuracy 0.5mm) in real time. After the signal is denoised using wavelet transform, the respiratory interval time series (such as the interval between adjacent peaks) and the heartbeat amplitude envelope are extracted. Using a sliding window (window length 5 minutes, step length 1 minute), the standard deviation of the respiratory interval (such as >15% is considered disordered) and the heartbeat attenuation rate (an amplitude drop of >20% within 30 minutes is considered abnormal) are calculated, matching the abnormal respiratory interval and heartbeat intensity attenuation characteristics in the monitoring signal.

[0071] For example, Mr. Zhang (a bedridden elderly man)'s piezoelectric sensor detected a breathing interval of 12 seconds (exceeding the threshold of 10 seconds), and the millimeter-wave radar showed a 50% decrease in breathing amplitude. The system determined it to be a respiratory depression event.

[0072] 103. Obtaining the body pressure and thermal zone distribution data of the bedridden area, combining it with the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggering an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and thermal zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person;

[0073] Body pressure thermal zone distribution data: A high-density pressure sensor mattress generates a pressure distribution heat map, identifying high-pressure areas (such as the sacrum and coccyx highlighted in red). Spatial superposition rules: Align the amplitude of respiratory micro-movements (such as chest and abdominal rise and fall) with the body pressure distribution coordinates to locate areas of overlapping respiratory depression and high pressure.

[0074] High-density fiber optic pressure sensing mattress (resolution 1cm 2 / point) to generate body pressure thermal zone distribution data, locate high-pressure areas (such as sacral pressure of 35 mmHg for 3 hours), and align the respiratory micro-movement amplitude (millimeter wave radar data) with the body pressure distribution through coordinate mapping. If the pressure in a certain area is greater than the critical value and the corresponding respiratory amplitude decreases by more than 50%, it is marked as a composite risk area. The YOLOv5 model is triggered to perform image recognition (input RGB + infrared image) on key areas such as the sacrum and heels, and analyze the skin fold depth (OpenCV edge detection) and temperature gradient (thermal imaging temperature difference > 2°C) to generate a retention risk index for bedridden people.

[0075] For example, Mr. Zhang's body pressure thermogram showed that the sacral pressure was 35 mmHg for 3 hours, and the respiratory amplitude in the corresponding area decreased by 70%. The AI model recognized that the skin texture deformation in this area was obvious, and the system generated a retention risk index of 78 points (high risk).

[0076] 104. Based on the detention risk index, extract the area of the body pressure distribution critical value exceeding the standard in the physiological parameter abnormality library, combine the duration of the abnormal breathing interval and the heartbeat intensity attenuation, and generate bed rest risk warning information through the image recognition AI model;

[0077] In the embodiment of the present application, based on the retention risk index, the coordinates of the area where the body pressure exceeds the standard (such as the sacral coccyx coordinates X:15, Y:20) are extracted, combined with the duration of abnormal breathing (such as a cumulative 6 minutes) and thermal imaging data (local temperature difference of 3.5°C), and input into the random forest model (features include pressure, breathing, temperature, and phase disorder). The model outputs the risk level (such as high risk) and the warning type (pressure sore + respiratory depression), and dynamically adjusts the warning threshold: if the pressure lasts for more than 4 hours, a bed rest risk warning information is generated.

[0078] Mr. Zhang suffered from sacral pressure for 4 hours, respiratory arrest for 6 minutes, and thermal imaging temperature difference of 3.5℃. The system triggered the "pressure ulcer and high respiratory risk" warning, prompting him to turn over immediately and check his airway.

[0079] 105. Generate a multimodal warning report based on the bed rest risk warning information.

[0080] Multimodal report: PDF / HTML report including heat map (body pressure distribution), line graph (respiratory rate trend), and text summary (risk description).

[0081] This report integrates body pressure thermograms (rendered with Matplotlib), respiratory rate trend graphs (dynamic curves generated with Plotly), infrared thermal images (pseudo-color processing with OpenCV), and text summaries (generated with Natural Language Processing) to generate PDF reports using the Python ReportLab library. The report includes: markers for high-pressure areas in the sacral region; a timeline of respiratory abnormalities (e.g., apnea from 2:00 PM to 2:06 PM); and nursing recommendations (e.g., "Turn the patient over every two hours and check blood oxygen saturation"). The report is automatically pushed to the nursing system, supporting real-time alerts via email and mobile devices, creating a multimodal early warning report.

[0082] The system generates an early warning report for Mr. Zhang, showing the high-pressure area in the sacrum and coccyx, the timeline of abnormal breathing, and nursing recommendations. The nurse adjusts the nursing plan based on the report.

[0083] While Mr. Zhang was bedridden in a nursing home, the system detected a sacral pressure of 35 mmHg (above the threshold of 32 mmHg) and a respiratory rate of 7 breaths / minute (below the threshold of 8). Millimeter-wave radar detected a 60% decrease in respiratory amplitude and a 25% attenuation in cardiac amplitude. The combined body pressure and respiratory data identified the sacral region as a complex risk area. The AI model identified skin deformation and a risk index of 78 points. Combined with the 3.5°C temperature difference in thermal imaging, an early warning of "pressure ulcers and high respiratory risk" was generated. The report was sent to the nurse, who was instructed to turn the patient over every two hours and monitor blood oxygen levels.

[0084] This solution achieves accurate monitoring and graded early warning of bedridden health risks through multimodal data fusion and AI dynamic analysis.

[0085] In order to improve the accuracy of risk assessment, static pressure areas with pressure values exceeding the threshold and duration integral exceeding the limit are screened based on body pressure thermal zone data; at the same time, the peak-to-valley difference sequence of respiratory micro-movement amplitude in the sliding time window is extracted, the variance of the overlapping area of adjacent windows is calculated, and low activity areas with variance below the threshold are screened; the spatial coordinates of the two types of areas are matched by grid coding, and the composite risk areas that meet both pressure integral exceeding the limit and respiratory depression in the grid unit are selected; according to the anatomical mapping relationship, the optical resolution of the AI model is dynamically adjusted, and the enhanced image is enhanced by OpenCV to extract the skin texture deformation features, and the pressure and respiratory depression parameters are combined to calculate the retention risk index of each part to achieve quantitative risk assessment. In some embodiments, multiple key body differences include the sacrum, hip and heel areas; the spatial superposition rule of the body pressure thermal zone distribution data and the respiratory micro-movement amplitude is triggered to analyze the multiple key body areas of the bedridden person and generate a retention risk index for the bedridden person, including:

[0086] 201. Based on the continuous areas where the pressure value exceeds the set threshold in the body pressure and thermal zone distribution data, calculate the time series integral of the pressure duration of each area, and screen out the static pressure areas where the integral value exceeds the first critical value;

[0087] Static pressure zone: refers to the continuous area in the body's pressure and thermal zone where the pressure value continuously exceeds the set threshold (such as 32mmHg), and the pressure duration integral (pressure × duration) exceeds the first critical value (such as integral > 5000). Time series integration: Integrates the curve of pressure data changing over time to quantify the cumulative pressure intensity.

[0088] In order to achieve accurate screening of static pressure areas, this step is based on high-density fiber optic pressure mattress (resolution 1cm 2 The system collects real-time pressure and thermal distribution data of the bedridden area (at each point) to generate a pseudo-color pressure thermogram. A connected domain analysis algorithm (such as OpenCV's findContours) is used to identify continuous regions where pressure exceeds a set threshold (e.g., 32 mmHg). The pressure time series integral of each region is calculated (pressure value × duration; for example, if a region has a pressure of 35 mmHg for 3 hours, the integral = 35 × 3 × 60 = 6300). If the integral exceeds a first critical value (e.g., 5000), the region is marked as statically compressed.

[0089] 202. Extract the peak-to-valley difference sequence of the respiratory micro-movement amplitude within the sliding time window, calculate the variance of the difference sequence for the overlapping area of adjacent windows, and screen the low-activity area with a variance value lower than a second critical value;

[0090] Low activity: When the variance of the peak-to-valley difference of respiratory micro-movements within a sliding time window (e.g., a 5-minute window with a 1-minute step) is less than a second critical value (e.g., variance <0.1), respiratory activity is significantly suppressed. Peak-to-valley difference sequence: A sequence of differences between adjacent peaks and troughs in the respiratory waveform, used to quantify respiratory fluctuation intensity.

[0091] This step uses a 60GHz millimeter-wave radar to monitor the chest and abdominal fluctuations of bedridden patients and extract the respiratory micro-movement amplitude signal (accuracy 0.1mm). A sliding time window (window length 5 minutes, step length 1 minute) is used to segment the respiratory waveform. The peak-to-valley difference sequence (the absolute value sequence of the peak-to-valley difference) of the overlapping area of adjacent windows is calculated, and its variance is quantified (for example, the variance of window 1 is 0.15, and the variance of window 2 is 0.08). If the variance is lower than the second critical value (such as 0.1), the area is determined to be a low-activity area.

[0092] 203. Perform grid coding matching on the spatial coordinates of the static pressure area and the spatial coordinates of the low activity area, and select a composite risk area in the grid unit that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression;

[0093] Grid-coded matching: Mapping body pressure distribution and respiratory amplitude data to a unified grid coordinate system (e.g., 1 cm × 1 cm grid) to screen for spatially overlapping composite risk areas. Composite risk areas are defined as grid cells that simultaneously meet the criteria of pressure integral exceeding the limit (>5000) and respiratory depression (variance <0.1).

[0094] The coordinates of the static pressure region from step 201 and the low-activity region from step 202 were mapped to a unified grid coordinate system (1 cm × 1 cm grid). Missing data areas were filled using bilinear interpolation. Each grid cell was traversed and marked as a composite risk region (e.g., a sacrococcygeal X:16, Y:28 grid) if its body pressure integral was greater than 5000 and its respiratory variance was less than 0.1. Adjacent high-risk grids were merged to create a continuous region outline (e.g., a 10 × 10 cm sacrococcygeal region), and the boundaries were optimized using an edge smoothing algorithm. Finally, a composite risk region was generated that met both pressure integral overlimit and respiratory fluctuation suppression requirements.

[0095] 204. Based on the mapping relationship between the composite risk area and the sacrum, hip, and heel areas in the human anatomical coordinate system, the optical resolution enhancement ratio of the image recognition AI model to the target area is dynamically allocated according to the preset weight coefficient;

[0096] The target area refers to the mapping relationship between the composite risk areas in the human anatomical coordinate system, focusing on key body parts prone to pressure ulcers and postural retention, such as the sacrum, hips, and heels. Anatomical coordinate system mapping: The AI model dynamically allocates the resolution enhancement ratio for different areas based on the human anatomical structure (such as the sacrum corresponding to grid coordinates X:15-20, Y:25-30). Weight coefficient: The regional priority is preset based on the clinical risk level (sacrum weight 0.6, hip 0.3, heel 0.1).

[0097] The optical resolution enhancement ratio of the image recognition AI model (YOLOv5) is dynamically allocated based on the location of the composite risk area in the human anatomical coordinate system (e.g., mapping weight 0.6 for the sacrum, 0.3 for the hip, and 0.1 for the heel). The super-resolution algorithm ESRGAN is used to enhance the image of the target area (e.g., the sacrum), increasing the resolution from 1mm / pixel to 0.1mm / pixel. Simultaneously, the model's attention mechanism (e.g., spatial pyramid pooling layer weights) is adjusted, and the image recognition AI model's optical resolution enhancement ratio for the target area is dynamically allocated according to preset weight coefficients.

[0098] Based on the enhanced image data, the skin texture deformation characteristics of the target area are extracted and the retention risk index of each key area of the body is calculated.

[0099] Skin texture deformation characteristics: Parameters extracted through image analysis include skin wrinkle depth (>0.2mm is abnormal) and erythema area (>5% is high risk). Detention risk index: A quantitative risk value (0-100) calculated by combining deformation characteristics, pressure integral, and respiratory depression parameters.

[0100] Based on enhanced image data, Canny edge detection was used to extract skin wrinkle contours (e.g., wrinkle depth 0.3 mm), and the gray-level co-occurrence matrix (GLCM) was used to analyze texture features (e.g., erythema area percentage 5%). The pressure integral (e.g., 6300), duration of respiratory depression (e.g., 40 minutes), and deformation parameters were integrated to calculate the retention risk index for each key body region using a weighted formula.

[0101] The body pressure and respiratory monitoring of Ms. Li (a bedridden patient) in a nursing home showed a sacral pressure of 35 mmHg for 5 hours. The integral was 35 × 5 × 60 = 10,500 (exceeding the threshold of 5,000), marking it as a static pressure area. The variance of respiratory micro-movement amplitude in adjacent windows was 0.08 and 0.06, marking it as a low activity area. Grid matching revealed that the sacral area (X:16, Y:28) met both the pressure integral exceeding limit and respiratory depression, marking it as a complex risk area. Using a sacral weight of 0.6, the resolution was increased to 0.1 mm / pixel. The AI model identified a wrinkle depth of 0.4 mm. The risk index was 0.4 × 50 + 10,500 × 0.01 + 50 × 10 = 20 + 105 + 50 = 175 points (high-risk threshold > 50), triggering an emergency alert.

[0102] In summary, this solution achieves precise localization and quantitative assessment of pressure ulcer and respiratory depression risk through spatial overlay of body pressure and respiratory data and dynamic resolution enhancement technology. A composite risk area screening algorithm reduces false positive rates, super-resolution image analysis improves the sensitivity of skin deformation detection, and a weighted risk index model supports tiered nursing decision-making.

[0103] In order to accurately locate the combined risk area of pressure ulcers and respiratory depression, in some embodiments, the spatial coordinates of the static pressure area are grid-coded and matched with the spatial coordinates of the low-activity area, and the combined risk area that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression within the grid unit is selected, including:

[0104] 301. Divide the bed rest area into orthogonal grid cells of equal resolution, perform grid attribution mapping on the spatial coordinates of the static pressure area and the spatial coordinates of the low activity area, and select grid cells that simultaneously meet the dual conditions of pressure integral exceeding limit and respiratory fluctuation suppression through Boolean logic operation;

[0105] Equal-resolution orthogonal grid cells: The bed rest area is divided into a grid of fixed dimensions (e.g., 1 cm × 1 cm) with aligned rows and columns, and each grid cell has unique coordinates (e.g., X:15, Y:25). Grid attribution mapping: The coordinates of statically compressed areas (pressure integral > 5000) and low-activity areas (respiratory variance < 0.1) are mapped to corresponding grid cells. Bi-conditional grid cells: Grid cells that simultaneously meet the conditions of pressure integral exceeding the limit and respiratory depression (e.g., a grid with a pressure integral of 6000 and a respiratory variance of 0.05).

[0106] To achieve preliminary screening of complex risk areas, this step first divides the bedridden area into orthogonal grid cells with equal resolution (e.g., 1 cm × 1 cm), each with unique coordinates (e.g., X:15, Y:25). Based on the original coordinate data of the static pressure area (pressure integral > 5000) and the low activity area (respiratory variance < 0.1), the physical coordinates are mapped to grid cells using an affine transformation algorithm (e.g., sacrococcygeal pressure point X:15.3, Y:25.7 corresponds to grid cell X:15, Y:25). Subsequently, all grid cells are traversed, and a Boolean logic "AND" operation is performed on each cell: if the corresponding physical area simultaneously meets the pressure integral exceeding limit (e.g., integral 6000) and respiratory suppression conditions (e.g., variance 0.05), it is marked as a grid meeting the dual conditions of pressure integral exceeding limit and respiratory fluctuation suppression.

[0107] Based on the spatial continuity of adjacent bi-conditional grid cells, discrete cells are merged through morphological dilation operations to generate a composite risk area covering the sacrum, hip and heel mapping areas. The boundary of the composite risk area is jointly calibrated by the pressure gradient mutation point and the respiratory amplitude jump point.

[0108] Morphological dilation: Dilates a discrete biconditional grid using a structural element (e.g., a 3×3 square), merging adjacent cells to form a continuous region. Pressure gradient abrupt change: The coordinate point in the pressure distribution where the pressure drops suddenly (e.g., the boundary from 32 mmHg to 20 mmHg). Respiratory amplitude jump: The moment in the respiratory waveform where the amplitude suddenly drops (e.g., the timestamp from 2 mm to 0.5 mm).

[0109] This step achieves precise generation of composite risk regions through morphological optimization and multi-source data fusion. First, a morphological dilation operation (OpenCV's dilate function, iterated twice) is performed on the biconditional grid using a 3×3 structuring element, merging adjacent discrete cells (e.g., X:15, Y:25 and X:16, Y:25 into a continuous region). Next, pressure gradient abrupt changes (e.g., a sudden drop in pressure from 32 mmHg to 20 mmHg at coordinates X:18, Y:28, with a gradient change rate >10 mmHg / cm) and respiratory amplitude jumps (e.g., a sudden drop of 1.2 mm in respiratory amplitude at 2:05 PM, corresponding to grid X:18, Y:28) are detected at the edges of the merged region. The abrupt changes are connected using the Canny edge detection algorithm to generate a composite risk region covering the sacrum, hip, and heel mapping areas.

[0110] Mr. Wang, a resident of a nursing home, was monitored for body pressure and respiration. The bed rest area was divided into a 1 cm × 1 cm grid. Five grid cells within the sacrococcygeal region at coordinates X:15-20, Y:25-30 satisfied a pressure integral greater than 5000 and a respiratory variance less than 0.1, marking them as biconditional. Morphological dilation merged adjacent grid cells to form a continuous region X:15-18, Y:25-28. A sudden pressure gradient drop at the edge of the region (X:18, Y:28) was detected (from 35 mmHg to 22 mmHg). A sudden 1.2 mm drop in respiratory amplitude at 2:05 PM resulted in a composite risk region outline covering the sacrococcygeal region.

[0111] In summary, this scheme achieves accurate positioning and boundary demarcation of complex risk areas through grid screening and morphological optimization.

[0112] In order to improve the objectivity and interpretability of the pressure ulcer risk index, the enhanced pressure area image is subjected to multi-scale spatial filtering to extract the continuous closed contour boundary and generate a contour sequence; the curvature difference value of the adjacent point set is calculated along the contour tangent direction, and the number of curvature sign changes is counted through a sliding window to characterize the complexity of the contour folds; the distribution of the cosine value of the gradient amplitude direction and the angle between the normal and the normal is counted along the normal direction on the inner side of the contour, and the cumulative proportion of the negative interval is calculated; finally, the curvature change rate and the gradient direction parameter are linearly weighted to generate a quantitative risk index, and the report is output by superimposing a text through a heat map. In some embodiments, the skin texture deformation characteristics include the contour curvature change rate and the edge gradient direction distribution parameters of the pressure area; based on the enhanced image data, the skin texture deformation characteristics of the target area are extracted, and the retention risk index of each key area of the body is calculated, including:

[0113] 401. Extracting contour pixels of the compressed region of the target area in the enhanced image data, retaining the continuous closed boundary of the compressed region through multi-scale spatial filtering, and generating a contour sequence of the compressed region;

[0114] Multi-scale spatial filtering: Uses a Gaussian-Laplacian pyramid of different scales to filter the image, preserving the continuous closed boundaries of compressed areas and eliminating noise interference. Contour sequence: A set of coordinates of continuous closed boundary pixels arranged in clockwise or counterclockwise order.

[0115] In order to achieve accurate extraction of the contour of the compressed area, this step performs multi-scale spatial filtering on the enhanced high-resolution image (such as 0.1mm / pixel image of the sacrum and coccyx). First, Gaussian pyramid decomposition (number of layers = 3, Gaussian kernel σ = 1.5) is used to generate images of different scales, and the Canny edge detection algorithm (low threshold 50, high threshold 150) is used to extract candidate contours in each layer of the image. Subsequently, closed contours (boundaries connected from beginning to end) are screened by chain code analysis, and a continuous closed contour sequence is generated after merging the multi-scale results (such as the sacrum and coccyx contour is composed of 200 pixels arranged clockwise). The parameter setting is based on clinical experimental data (such as the coverage rate of closed contours on the pressure ulcer area reaches 95%), and finally the contour sequence of the compressed area is output.

[0116] 402. Calculate the curvature difference values of adjacent contour point sets along the tangent direction of the contour sequence, perform sliding window statistics on the number of times the sign of the curvature difference values changes, and generate the contour curvature change rate of the compressed area;

[0117] Curvature difference value: The difference in curvature change between adjacent contour points (e.g., point A has a curvature of 0.2, point B has a curvature of -0.1, and the difference is 0.3). Sliding window statistics: Count the number of curvature sign changes using a fixed window length (e.g., 10 pixels) and step size (e.g., 5 pixels).

[0118] This step quantifies the complexity of contour wrinkles by the curvature change rate. Along the tangent direction of the contour sequence, the curvature of the adjacent contour point set (such as points i-1, i, i+1) is calculated (based on the curvature radius of the three-point fitting circle, curvature = 1 / radius), and the curvature difference of the adjacent points is calculated (such as point i has a curvature of 0.3, point i+1 has a curvature of -0.1, and the difference is 0.4). A sliding window (window length 10 pixels, step length 5 pixels) is used to count the number of sign changes of the curvature difference value within the window (such as from positive to negative or negative to positive). For example, a sign change of 3 times within a window indicates high wrinkle fluctuation. The parameters are based on skin wrinkle morphology research (the curvature change rate of healthy skin is <2 times / window) to generate the contour curvature change rate of the compressed area.

[0119] 403. In the inner region of the contour of the compressed area, along the normal path perpendicular to the contour direction, a frequency distribution histogram of the cosine value of the angle between the gradient amplitude direction and the normal direction is statistically analyzed, and the cumulative proportion of the cosine value of the angle in the histogram that is within a preset negative interval is calculated to extract the edge gradient direction distribution parameter;

[0120] Normal path: The extension path perpendicular to the contour boundary (e.g., extending 5 pixels inward from the contour point). Gradient direction distribution parameter: The cosine value distribution of the angle between the gradient direction and the normal direction (cosine value < -0.5 indicates that the gradient direction is inward).

[0121] This step analyzes the indentation characteristics of skin deformation based on the gradient direction distribution. In the inner area of the contour, extend 5 pixels along the normal direction (the path perpendicular to the boundary), and calculate the gradient amplitude and direction by the Sobel operator. Count the cosine value of the angle between the gradient direction and the normal (such as the angle between the gradient direction and the normal is 60°, the cosine value is 0.5), and generate a frequency distribution histogram (binning interval 0.1). Calculate the cumulative proportion of the preset negative interval (cosine value <-0.5) (such as a proportion of 65% indicates that the gradient is significantly indented). The parameters are derived from the study of early skin deformation of pressure ulcers (the proportion of indented areas >60% is high risk), and the edge gradient direction distribution parameters are output.

[0122] 404. Generate a retention risk index for each key body region using a linear weighted fusion formula based on the window statistical result of the contour curvature change rate and the cumulative proportion of the edge gradient direction distribution parameter.

[0123] The statistical mean of the contour curvature change rate sequence is directly proportional to the weight coefficient of the retention risk index, and the cumulative proportion of the edge gradient direction distribution parameter is inversely proportional to the weight coefficient of the retention risk index. The linear weighted fusion formula is as follows: the risk index is calculated according to the preset weights (such as the curvature change rate weight of 0.6 and the gradient parameter weight of 0.4).

[0124] This step generates a quantitative risk index through feature fusion. The curvature change rate (such as the number of sign changes 3 times / window) and the gradient direction parameter (such as the negative interval accounts for 65%) are normalized to the range of 0-1 (curvature parameter = 3 / 5 = 0.6, gradient parameter = 65 / 100 = 0.65). The retention risk index is calculated according to the linear weighted formula (curvature weight 0.6, gradient weight 0.4): 0.6×0.6+0.65×0.4=0.62 (high-risk threshold>0.5). The parameter weights are based on clinical data regression analysis (curvature contributes 60% to pressure ulcers), and finally the retention risk index of each key area of the body is generated.

[0125] Mr. Zhang from a nursing home suffered from a sacral compression region: Enhanced images showed the sacral contour, and a closed contour sequence (180 points) was extracted after multi-scale filtering. Sliding window statistics showed that the curvature sign changed four times per window (high fold complexity). Gradient direction statistics showed that the negative interval accounted for 65% (significant inward collapse). The risk index = 0.8 (curvature) × 0.6 + 0.65 (gradient) × 0.4 = 0.74 (high risk), triggering an alert.

[0126] In summary, this scheme realizes the quantitative assessment of skin deformation risk through multi-scale contour extraction, curvature change analysis and gradient direction statistics.

[0127] To improve the sensitivity and specificity of bed rest risk warning, the critical body pressure values of the sacrum, hip, and heel were matched from the physiological parameter abnormality library based on the retention risk index. Areas with continuously exceeded pressure limits and an area greater than the set threshold were screened. Phase offsets were calculated for adjacent respiratory cycles during periods of abnormal respiratory intervals, and the density of mutation points was statistically distributed over the duration of the exceeded areas. Temperature profiles were extracted along the direction of maximum gradient descent in the body pressure thermal zone, and the covariance coefficients of temperature decay rate and heartbeat decay rate were calculated. The density distribution and covariance coefficients were input into the ResNet-50 model, and the geometric features of the exceeded areas were integrated with the time-frequency features of respiration and heartbeat to generate a 128-dimensional feature vector. Based on the mapping relationship between the feature vector and the preset risk level, the weight of the critical body pressure value was dynamically adjusted to output graded warning information.

[0128] In some embodiments, the excessive area corresponding to the critical value of body pressure distribution in the physiological parameter abnormality library is extracted, combined with the duration of the abnormal breathing interval and the heartbeat intensity attenuation, and the bed rest risk warning information is generated by the image recognition AI model, including:

[0129] 501. Based on the retention risk index, match the critical values of body pressure distribution corresponding to the sacrum, hip, and heel regions from the physiological parameter abnormality library, and screen out the areas where the pressure values continuously exceed the critical values of body pressure distribution and the spatial distribution area is larger than the set threshold;

[0130] Body pressure distribution threshold: the maximum pressure threshold allowed in different body regions (such as 32 mmHg for the sacrum and 28 mmHg for the hips); Exceeding area: the pressure continuously exceeds the critical value and the distribution area is greater than the set threshold (such as 10 cm 2 ) area.

[0131] This step screens high-risk pressure areas by linking the retention risk index with the physiological parameter library. First, the critical values of body pressure distribution in the sacral, hip, and heel regions (such as 32 mmHg for the sacral, and 28 mmHg for the hip) are retrieved from the physiological parameter abnormality library, and the areas that need to be monitored are determined based on the retention risk index (such as a sacral index of 0.8). The connectedComponents algorithm of OpenCV is used to identify areas in the pressure sensor data that continuously exceed the critical value (such as a sacral pressure of 35 mmHg for 3 hours), and the area is calculated (number of pixels × area of a single pixel, such as 1 cm 2 / pixel) screening distribution area>10cm 2 Exceeding the standard area (such as 12cm in the sacrum and coccyx) 2Parameter settings are based on clinical research (pressure ulcer risk is positively correlated with area), and the final output pressure value continuously exceeds the critical value of body pressure distribution and the spatial distribution area is larger than the set threshold.

[0132] 502. Calculate the phase offset of adjacent respiratory cycles within the abnormal respiratory interval period, extract a mutation point sequence in which the phase difference between adjacent peaks in the respiratory signal exceeds a preset angle, and calculate the density distribution of the mutation point sequence within the duration of the abnormal respiratory interval period;

[0133] Phase offset: The phase angle corresponding to the time difference between the peaks of adjacent respiratory cycles (e.g., a time difference of 0.5 seconds corresponds to a phase difference of 90°). Mutation point density: The number of phase mutation points (phase difference > 90°) that occur per unit time.

[0134] This step focuses on analyzing the spatiotemporal correlation between abnormal breathing and excessive pressure. During periods of abnormal breathing intervals (e.g., apnea lasting 5 minutes), the instantaneous phase of the respiratory signal is extracted using a Hilbert transform, and the phase difference between adjacent peaks is calculated (e.g., a peak interval of 0.6 seconds corresponds to a phase difference of 108°). If the phase difference is greater than the preset angle (90°), it is marked as a mutation point (e.g., three mutations are detected in a certain period). The density of mutation points within the duration of the excessive pressure area is calculated.

[0135] 503. Extract the temperature gradient profile line along the direction of maximum decrease of the body pressure gradient from the body pressure thermal zone distribution data corresponding to the exceeding standard area, and calculate the covariance coefficient of the temperature decay rate and the heartbeat intensity decay rate within a unit distance on the profile line;

[0136] Temperature gradient profile: The temperature curve along the direction of maximum decrease in body pressure gradient (e.g., from 35 mmHg to 20 mmHg in the negative x-axis direction). Covariance coefficient: The covariance between the temperature decay rate (°C / cm) and the heart rate decay rate (bpm / minute), quantifying the correlation between the two.

[0137] This step reveals the risk of tissue ischemia through covariance analysis of temperature and heart rate. The temperature profile line is extracted along the maximum gradient direction in the body pressure thermal zone (such as the negative direction of the X-axis when the sacral pressure drops from 35 mmHg to 20 mmHg), and the temperature decay rate per unit distance (such as 2°C / cm) and the heart rate decay rate (5 bpm per minute) are calculated. The covariance coefficient is calculated using the covariance formula. If the coefficient is greater than 0.8 (strong positive correlation), it indicates that the temperature drop is synchronized with the weakening of cardiac function. The parameters are based on the correlation study of thermodynamics and hemodynamics, and the covariance coefficient of the temperature decay rate and the heart rate intensity decay rate per unit distance on the output surface line is obtained.

[0138] 504. Input the density distribution and covariance coefficient into the image recognition AI model, perform multimodal fusion of the geometric topological features of the exceeded area and the time-frequency domain features of the respiratory and heartbeat signals, and generate a feature vector representing the intensity of pressure-physiological coupling;

[0139] Geometric topological features: Parameters such as the area of the exceeded area, shape factor (perimeter squared / area), and centroid coordinates. Time-frequency domain features: Wavelet packet energy of respiratory and heartbeat signals (energy proportion in the 0.5-4 Hz frequency band).

[0140] This step integrates multi-source data to generate risk feature vectors. The geometric topological features of the exceeding area (such as the area of 12cm 2 , shape factor 1.3), combined with the time-frequency characteristics of the respiratory and heartbeat signals (wavelet packet decomposition extracts 0.5-4Hz low-frequency energy accounting for 70%), the mutation point density (3 times / hour), covariance coefficient (0.88) and other parameters are input into the fully connected layer, and the characteristic vector representing the intensity of pressure and physiological coupling is output.

[0141] 505. According to the mapping relationship between the dimensional parameters in the feature vector and the preset risk levels in the physiological parameter abnormality library, the local weight coefficient of the body pressure distribution critical value is dynamically adjusted, and bed rest risk warning information is output.

[0142] Local weight coefficient: Dynamically adjust the regional weight of the body pressure critical value according to the eigenvector (for example, the sacral weight increases from 0.6 to 0.8).

[0143] This step implements dynamic parameter optimization and graded early warning. A random forest model is used to classify risk levels based on the mapping relationship between the eigenvector dimensions (such as area, covariance coefficient, and low-frequency energy) and the preset risk levels in the physiological parameter library (high risk: eigenvalue > 0.7). If the sacrum and coccyx are determined to be high-risk, the body pressure threshold weight is dynamically adjusted, and a eigenvector representing the strength of the pressure-physiological coupling is output.

[0144] Sacrococcygeal monitoring of Mr. Wang in a nursing home: retention risk index 0.8, matching sacrococcygeal critical value 32mmHg, screening area 12cm 2 The respiratory phase mutation point density is 4 times / hour (12 times in 3 hours). The temperature decay rate is 2°C / cm, the heart rate decay rate is 6bpm / minute, the covariance coefficient is 0.88, and the characteristic vector contains an area of 12cm. 2 , shape factor 1.3, wavelet energy 75%, and output 128-dimensional vector. The model determined high risk, and the sacrococcygeal critical value weight was adjusted to 0.8 (25.6 mmHg), triggering an emergency warning.

[0145] In summary, this solution achieves coupled risk warning for pressure and physiological signals through dynamic critical value adjustment and multimodal feature fusion. The covariance coefficient quantifies the temperature-heartbeat correlation (increasing accuracy by 25%), the random forest model maps risk levels with an error rate of less than 5%, and the dynamic weighting mechanism increases warning sensitivity by 40%. In actual application, data from a pilot program at a nursing home showed a 35% reduction in false alarm rates for pressure ulcers and a reduction in nursing response time to under 20 minutes, validating the technology's efficiency and reliability.

[0146] In order to improve the characterization capability of the coupling characteristics of pressure and physiological signals, the geometric topological features of the exceeding standard area are subjected to wavelet multi-scale decomposition, the frequency domain energy distribution of the curvature change rate at different scales is extracted, and a multi-scale frequency domain descriptor is generated; the time-frequency domain features of the respiratory and heartbeat signals are segmented and analyzed, the time-frequency energy distribution of the high-density period is extracted, and a time-frequency energy descriptor is generated; the covariance coefficient is mapped to the geometric space, and the local correlation between the temperature decay rate and the contour curvature is calculated through the local spatial autocorrelation model to generate a temperature-geometry coupling descriptor; the three types of descriptors are input into the ResNet-50 model, and a 256-dimensional initial feature vector is generated through spatial alignment and attention weight allocation; the autoencoder is further used to compress the features, and the dimensions that are strongly correlated with the pressure distribution critical value, density distribution, and covariance coefficient are screened to generate a 32-dimensional high-information feature vector for dynamic risk assessment. In some embodiments, the density distribution and covariance coefficient are input into the image recognition AI model, and the geometric topological features of the exceeding standard area and the time-frequency domain features of the respiratory and heartbeat signals are multimodally fused to generate a feature vector representing the strength of the pressure and physiological coupling, including:

[0147] 601. Perform multi-scale decomposition on the geometric topological features of the exceeding-standard area, extract the frequency domain distribution of the contour curvature change rate at different scales, and generate a multi-scale frequency domain descriptor of the geometric topological features;

[0148] Multiscale frequency domain descriptor: This extracts the energy distribution parameters of geometric topological features (such as contour curvature) in different frequency bands through wavelet multiscale decomposition, which is used to describe the multiscale characteristics of shape complexity. Contour curvature change rate: The difference in curvature between adjacent points along a contour sequence reflects the rate of change of the local curvature.

[0149] To achieve multi-scale characterization of geometric topological features, this step uses wavelet transform to perform frequency-domain decomposition of the rate of change of contour curvature in the exceeding-standard area. First, a three-level wavelet decomposition is performed on the contour curvature sequence (e.g., 200 points) (using db4 as the mother wavelet) to generate low-frequency (0.5-2Hz), medium-frequency (2-4Hz), and high-frequency (4-8Hz) sub-band signals. The energy contribution of each sub-band signal is calculated (e.g., low frequency accounts for 70%), forming a multi-scale frequency-domain descriptor of the geometric topological features.

[0150] 602. Perform segmented analysis on the time-frequency domain features of the respiratory and heartbeat signals, extract the time-frequency energy distribution of high-density areas in the density distribution, and generate a time-frequency energy descriptor of the respiratory and heartbeat signals;

[0151] Time-frequency energy descriptor: This extracts the energy distribution of respiratory and heartbeat signals in specific frequency bands through time-frequency analysis (such as wavelet packet decomposition), reflecting the rhythmic characteristics of physiological signals. High-density region: The time period with a respiratory mutation point density greater than 3 times / minute.

[0152] This step quantifies rhythm abnormalities in the respiratory and heartbeat signals through time-frequency analysis. The respiratory and heartbeat signals are segmented into 5-minute windows, and high-density regions (mutation point density > 3 beats / minute) are extracted. Wavelet packet decomposition (tree depth = 3) is then used to extract the energy fraction (e.g., 80%) in the 0.5-4 Hz frequency band to generate a time-frequency energy descriptor for the respiratory and heartbeat signals.

[0153] 603. Map the covariance coefficient to the geometric topological feature space of the exceeding-standard area, calculate the local correlation between the temperature decay rate and the contour curvature change rate, and generate a temperature and geometry coupling descriptor;

[0154] Temperature and geometry coupling descriptor: quantifies the local spatial correlation between temperature decay rate and profile curvature change (e.g., a covariance coefficient > 0.7 indicates a strong correlation). Local correlation: calculates the degree of synergistic change between temperature and geometric features using a spatial autocorrelation model (Moran's I index).

[0155] This step uses spatial statistical modeling to model the coupling relationship between temperature and geometric features. Covariance coefficients (e.g., a temperature-heartbeat covariance of 0.85) are mapped to the local coordinates of the geometric contour. Moran's I index (e.g., I = 0.6) is used to calculate the local spatial autocorrelation of the temperature decay rate and curvature change. The calculation is based on an adjacency weight matrix (threshold distance = 5 cm). If I > 0.5, the region is considered strongly coupled, and a temperature-geometry coupling descriptor is generated.

[0156] 604. Input the multi-scale frequency domain descriptor of the geometric topological features, the time-frequency energy descriptor of the respiratory and heartbeat signals, and the temperature and geometric coupling descriptor into the image recognition AI model, and generate an initial feature vector through spatial alignment and weight distribution rules of multimodal features;

[0157] Multimodal feature alignment: Mapping feature vectors from different modalities to the same spatial coordinate system (e.g., aligning geometric and physiological signals through affine transformation). Weight assignment rule: Dynamically assigning weights based on feature importance (e.g., weight 0.5 for geometric features, 0.3 for physiological features).

[0158] This step integrates multimodal features to generate a high-dimensional vector. The spatial coordinate systems of the geometric, physiological, and temperature descriptors are aligned through affine transformation, input into a ResNet-50 model (with pre-trained weights fine-tuned), and weights are assigned using the SE attention module (geometric 0.5, physiological 0.3, coupling 0.2), outputting the initial feature vector.

[0159] 605. Perform feature space compression on the initial feature vector, extract the dimensional parameters with the highest correlation with the pressure distribution critical value, density distribution, and covariance coefficient in the feature vector, and generate a feature vector representing the pressure and physiological coupling strength.

[0160] Feature space compression: extract high-information features through dimensionality reduction algorithms (such as autoencoders) and remove redundant dimensions. Correlation screening: screen dimensional parameters that are strongly correlated with pressure, density, and covariance (Pearson correlation coefficient > 0.7).

[0161] This step optimizes feature validity through dimensionality reduction and screening. An autoencoder (256-dimensional input, 64-dimensional encoding layer, and 256-dimensional decoding layer) was constructed and trained using mean squared error as the loss function. Initial features were compressed to 64 dimensions. The Pearson correlation coefficients of each dimension with the pressure threshold, density distribution, and covariance were calculated (e.g., the correlation coefficient for dimension 1 was 0.82). Dimensions > 0.7 were selected to generate feature vectors representing the strength of pressure-physiological coupling.

[0162] Sacrococcygeal pressure ulcer risk assessment for patients in a nursing home: The sacrococcygeal contour was decomposed using wavelet analysis, with low-frequency energy accounting for 75%, generating a descriptor of [0.75, 0.2, 0.05]. The respiratory mutation point density was 4 times / minute, resulting in a time-frequency energy descriptor of [0.8, 0.15, 0.05]. The Moran's I index, calculated based on the temperature-curvature correlation, was 0.65, generating a coupling descriptor of [0.65, 0.85]. A ResNet-50 fusion generated a 256-dimensional vector, with attention weights assigned as 0.5 for geometry, 0.3 for physiology, and 0.2 for coupling. The autoencoder was compressed to 32 dimensions, and dimensions 1 (correlation coefficient 0.82), 3 (0.75), and 5 (0.78) were selected to generate the final feature vector.

[0163] In summary, this scheme achieves efficient feature fusion of pressure and physiological signals through multi-scale frequency domain decomposition, time-frequency energy analysis and spatial correlation modeling.

[0164] In order to improve the adaptability and accuracy of the bed rest risk warning system, based on the bed rest risk warning information, the period of time in which the body micro-motion cycle does not meet the standard in the physiological parameter abnormality library is screened, and the sliding window statistics (window length 24 hours, step length 1 hour) are used to calculate the historical pressure distribution mean and standard deviation, and dynamically calibrate the body pressure critical value; reconstruct the body pressure hot zone segmentation parameters, identify the skin wrinkle depth and erythema area (such as 5%) through the pre-trained ResNet-50 model, and simultaneously extract the abnormal respiratory waveform and the hot zone exceeding the standard distribution; integrate the calibrated threshold, wrinkle depth, erythema characteristics and abnormal respiratory waveform segments, generate a multimodal warning report in PDF / HTML format, including a hot zone distribution heat map, a highlight mark of the abnormal respiratory waveform segment and nursing suggestions, and push it to the nursing system through the API interface. In some embodiments, the multimodal warning report generated based on the bed rest risk warning information includes:

[0165] 701. Based on the bed rest risk warning information, threshold calibration is performed on the time periods in the physiological parameter abnormality database that do not meet the body micro-movement cycle standard;

[0166] Micro-movement cycle standard: A preset threshold for the time interval between bedridden patients turning over or moving their limbs (e.g., micro-movement at least once every two hours). Threshold calibration: Dynamically adjusts the critical value of body pressure distribution based on historical data to adapt it to individual differences or environmental changes.

[0167] To achieve dynamic threshold calibration, this step screens the physiological parameter anomaly database for time periods (e.g., no movement for three consecutive hours) that do not meet the body micro-movement cycle (e.g., turning interval >2 hours) based on bed rest risk warning information. A sliding window statistic (window length 24 hours, step length 1 hour) is used to calculate the historical pressure distribution mean (e.g., sacrococcygeal pressure mean 28 mmHg) and standard deviation (e.g., 5 mmHg). The body pressure threshold is dynamically adjusted to the mean + standard deviation (28 + 5 = 33 mmHg) to adapt to individual differences and environmental changes. The calibration parameters are derived from clinical data regression analysis (e.g., individual pressure fluctuation range ±2σ covers 95% of normal values). The updated critical value is written back to the anomaly database and triggers the synchronous adjustment of the warning rules. For example, if a patient's sacrococcygeal pressure tolerance decreases due to long-term bed rest, the original critical value of 32 mmHg is calibrated down to 28 mmHg. The system lowers the risk judgment threshold for this area in real time and performs threshold calibration.

[0168] 702. Reconstruct the regional segmentation parameters of the body pressure heat zone distribution data based on the calibrated body pressure distribution critical value, perform skin texture wrinkle depth recognition and erythema feature extraction on the key areas of the body that continuously exceed the calibration threshold through the pre-trained convolutional neural network, and generate a multimodal early warning report including abnormal breathing waveforms, heat zone exceeding the standard distribution and calibrated thresholds.

[0169] Regional segmentation parameters: The standard for dividing high-risk areas in the body pressure and heat zone distribution data (such as area > 8cm 2 Skin wrinkle depth: quantify the degree of collapse of skin wrinkles through image analysis (depth > 0.2 mm is abnormal).

[0170] This step generates an accurate early warning report through multimodal data fusion. Based on the calibrated body pressure threshold (such as 28mmHg), the hot zone segmentation parameters are reconstructed (the area threshold is from 10cm 2 Adjust to 8cm 2 ), screen for areas that continuously exceed the standard (such as sacral pressure 30mmHg for 2 hours, area 9cm 2 The pre-trained ResNet-50 model (with 0.1mm / pixel enhanced image as input) is used to identify skin wrinkle depth (Canny edge detection calculates wrinkle contours, with depths >0.2mm considered abnormal) and erythema features (HSV color space segmentation with red area >5%). Abnormal respiratory waveforms (such as apnea period 14:00-14:05) and hot spot distribution maps are simultaneously extracted. The calibrated threshold, wrinkle depth (0.3mm), and erythema area (6%) are integrated to generate a multimodal warning report containing abnormal respiratory waveforms, hot spot distribution exceeding the standard, and calibrated thresholds.

[0171] A patient in a nursing home, Ms. Li, had not moved for 4 consecutive hours. The system calculated the historical mean of her sacral pressure as 26 mmHg (standard deviation 4 mmHg), and calibrated the critical value to 26 + 4 = 30 mmHg. 2 ), a sacral pressure of 32mmHg was detected for 2 hours; ResNet-50 identified a wrinkle depth of 0.35mm and erythema accounted for 6%; the respiratory waveform showed apnea from 14:00 to 14:05; a report was generated and pushed, and it was recommended to "turn over every 1.5 hours and check blood oxygen."

[0172] In summary, this scheme significantly improves the individual adaptability and accuracy of the early warning system through dynamic threshold calibration and multimodal data fusion.

[0173] Figure 2 The present application embodiment provides a structural diagram of an elderly care service supervision system based on an image recognition AI model algorithm, such as Figure 2 As shown, the system includes:

[0174] Establish Module 21 to establish a physiological parameter abnormality database based on the health risks of respiratory arrest, pressure sores, and limb rigidity in bedridden elderly care scenarios;

[0175] Matching module 22, which collects the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person through the vital sign monitoring device deployed in the bedridden area, and matches the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory frequency threshold value of the physiological parameter abnormality library;

[0176] A generation module 23 obtains the body pressure and heat zone distribution data of the bedridden area, combines the data with the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggers an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and heat zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person;

[0177] The generating module 23 is further configured to extract, based on the retention risk index, areas of excess body pressure distribution threshold values corresponding to the physiological parameter abnormality library, and generate bed rest risk warning information through the image recognition AI model in combination with the duration of the abnormal breathing interval and the heartbeat intensity attenuation;

[0178] The generating module 23 is further configured to generate a multimodal warning report based on the bed rest risk warning information. Figure 2 The elderly care service supervision system based on the image recognition AI model algorithm can perform Figure 1 The implementation principle and technical effects of the elderly care service supervision method based on the image recognition AI model algorithm described in the illustrated embodiment will not be repeated here. The specific manner in which each module and unit performs operations in the elderly care service supervision system based on the image recognition AI model algorithm in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated here.

[0179] In one possible design, Figure 2 The elderly care service supervision system based on the image recognition AI model algorithm of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0180] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0181] The processing component 32 is used for the above Figure 1 The embodiment of the present invention is a method for supervising elderly care services based on an image recognition AI model algorithm.

[0182] The processing component 32 may include one or more processors to execute computer instructions to perform all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0183] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0184] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0185] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0186] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0187] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0188] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a method for supervising elderly care services based on an image recognition AI model algorithm.

[0189] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0190] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0191] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for supervising elderly care services based on an image recognition AI model algorithm, characterized in that: include: Establish a database of abnormal physiological parameters based on the health risks of respiratory arrest, pressure sores, and limb rigidity in bedridden elderly care scenarios; The vital sign monitoring device deployed in the bedridden area collects the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person, and matches the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory frequency threshold value of the physiological parameter abnormality library; Obtaining the body pressure and thermal zone distribution data of the bedridden area, combining it with the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggering an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and thermal zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person; Based on the detention risk index, the area of the body pressure distribution critical value exceeding the standard corresponding to the physiological parameter abnormality library is extracted, and combined with the duration of the abnormal breathing interval and the heartbeat intensity attenuation, the bed rest risk warning information is generated through the image recognition AI model; Based on the bed rest risk warning information, a multimodal warning report is generated.

2. The method according to claim 1, characterized in that Multiple key body areas include the sacrum, hip, and heel regions. The spatial superposition rule of the body pressure and thermal zone distribution data and the respiratory micro-movement amplitude is used to trigger the image recognition AI model to analyze multiple key body areas of the bedridden person and generate a retention risk index for the bedridden person, including: Based on the continuous areas where the pressure value exceeds the set threshold in the body pressure and thermal zone distribution data, the time series integral of the pressure duration of each area is calculated, and the static pressure area with the integral value exceeding the first critical value is screened out; Extract the peak-to-valley difference sequence of respiratory micro-movement amplitude within the sliding time window, calculate the variance of the difference sequence for the overlapping area of adjacent windows, and screen out low-activity areas with variance values below the second critical value; Grid-coding and matching the spatial coordinates of the static pressure area with the spatial coordinates of the low-activity area, and selecting a composite risk area within the grid unit that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression; Based on the mapping relationship between the composite risk area and the sacrum, hip, and heel regions in the human anatomical coordinate system, the image recognition AI model dynamically allocates the optical resolution enhancement ratio of the target area according to the preset weight coefficient; Based on the enhanced image data, the skin texture deformation characteristics of the target area are extracted, and the retention risk index of each key area of the body is calculated.

3. The method according to claim 2, characterized in that The spatial coordinates of the static pressure area are matched with the spatial coordinates of the low activity area by grid coding, and a composite risk area that satisfies both pressure integral exceeding limit and respiratory fluctuation suppression is selected within the grid unit, including: Dividing the bed area into orthogonal grid cells of equal resolution, performing grid attribution mapping on the spatial coordinates of the static pressure area and the spatial coordinates of the low activity area, and screening grid cells that simultaneously meet the dual conditions of pressure integral exceeding limit and respiratory fluctuation suppression through Boolean logic operations; Based on the spatial continuity of adjacent biconditional grid cells, discrete cells were merged through morphological dilation operations to generate a composite risk area covering the sacrum, hip, and heel mapping areas. The boundaries of the composite risk area were jointly calibrated by the pressure gradient mutation points and the respiratory amplitude jump points.

4. The method according to claim 2, characterized in that The skin texture deformation characteristics include the contour curvature change rate and edge gradient direction distribution parameters of the compressed area; The method of extracting the skin texture deformation characteristics of the target area based on the enhanced image data and calculating the retention risk index of each key area of the body includes: The contour pixels of the compressed area of the target area in the enhanced image data are extracted, and the continuous closed boundaries of the compressed area are retained through multi-scale spatial filtering to generate a contour sequence of the compressed area; Calculating the curvature difference values of adjacent contour point sets along the tangent direction of the contour sequence, performing sliding window statistics on the number of times the sign of the curvature difference values changes, and generating the contour curvature change rate of the compressed area; In the inner area of the compressed area contour, along the normal path perpendicular to the contour direction, a frequency distribution histogram of the cosine value of the angle between the gradient amplitude direction and the normal direction is statistically analyzed, and the cumulative proportion of the cosine value of the angle in the histogram falling within a preset negative interval is calculated to extract the edge gradient direction distribution parameter; According to the window statistical results of the contour curvature change rate and the cumulative proportion of the edge gradient direction distribution parameters, the retention risk index of each key area of the body is generated through a linear weighted fusion formula.

5. The method according to claim 1, wherein The method extracts the exceeding area of the critical value of the body pressure distribution corresponding to the physiological parameter abnormality library according to the retention risk index, combines the duration of the abnormal breathing interval and the heartbeat intensity attenuation, and generates bed rest risk warning information through the image recognition AI model, including: According to the retention risk index, the body pressure distribution critical values corresponding to the sacrum, hip and heel regions are matched from the physiological parameter abnormality library, and the exceeding-standard areas whose pressure values continuously exceed the body pressure distribution critical values and whose spatial distribution area is larger than the set threshold are screened out; Calculating the phase offset of adjacent respiratory cycles within the abnormal respiratory interval period, extracting a sequence of mutation points in the respiratory signal where the phase difference between adjacent peaks exceeds a preset angle, and statistically analyzing the density distribution of the mutation point sequence within the duration of the abnormal respiratory interval period; In the body pressure thermal zone distribution data corresponding to the exceeding standard area, a temperature gradient profile line is extracted along the direction of maximum decrease of the body pressure gradient, and the covariance coefficient of the temperature decay rate and the heartbeat intensity decay rate within a unit distance on the profile line is calculated; The density distribution and covariance coefficient are input into the image recognition AI model, and a multimodal fusion is performed on the geometric topological features of the exceeded area and the time-frequency domain features of the respiratory and heartbeat signals to generate a feature vector representing the intensity of pressure and physiological coupling; According to the mapping relationship between each dimensional parameter in the feature vector and the preset risk level in the physiological parameter abnormality library, the local weight coefficient of the body pressure distribution critical value is dynamically adjusted to output bed rest risk warning information.

6. The method according to claim 5, characterized in that The density distribution and covariance coefficient are input into the image recognition AI model, and multimodal fusion is performed through the geometric topological features of the exceeded area and the time-frequency domain features of the respiratory and heartbeat signals to generate a feature vector representing the intensity of pressure and physiological coupling, including: Performing multi-scale decomposition on the geometric topological features of the exceeding-standard area, extracting the frequency domain distribution of the contour curvature change rate at different scales, and generating a multi-scale frequency domain descriptor of the geometric topological features; Performing segmented analysis on the time-frequency domain features of the respiratory and heartbeat signals, extracting the time-frequency energy distribution of high-density areas in the density distribution, and generating a time-frequency energy descriptor of the respiratory and heartbeat signals; Mapping the covariance coefficient to the geometric topological feature space of the exceeding area, calculating the local correlation between the temperature decay rate and the contour curvature change rate, and generating a temperature and geometry coupling descriptor; Input the multi-scale frequency domain descriptor of the geometric topological features, the time-frequency energy descriptor of the respiratory and heartbeat signals, and the temperature and geometric coupling descriptor into the image recognition AI model, and generate an initial feature vector through the spatial alignment and weight distribution rules of the multimodal features; The initial eigenvector is compressed in eigenspace, and the dimensional parameters with the highest correlation with the critical value of pressure distribution, density distribution, and covariance coefficient in the eigenvector are extracted to generate a eigenvector representing the intensity of pressure-physiological coupling.

7. The method according to claim 1, characterized in that Generating a multimodal warning report based on the bed rest risk warning information includes: Based on the bed rest risk warning information, threshold calibration is performed on the time periods in the physiological parameter abnormality library that do not meet the body micro-movement cycle standard; Based on the calibrated body pressure distribution critical value, the regional segmentation parameters of the body pressure heat zone distribution data are reconstructed. The pre-trained convolutional neural network is used to perform skin texture wrinkle depth recognition and erythema feature extraction on key areas of the body that continuously exceed the calibration threshold, generating a multimodal early warning report including abnormal respiratory waveforms, heat zone exceeding standard distribution and calibrated thresholds.

8. A pension service supervision system based on image recognition AI model algorithm, characterized by: include: Establish a module to build a physiological parameter abnormality library based on the health risks of respiratory arrest, pressure sore risks, and limb rigidity in bedridden elderly care scenarios; A matching module collects the heartbeat vibration signal and respiratory micro-movement amplitude of the bedridden person through a vital sign monitoring device deployed in the bedridden area, and matches the abnormal breathing interval and heartbeat intensity attenuation characteristics in the monitoring signal based on the respiratory frequency threshold value of the physiological parameter abnormality library; a generation module that obtains the body pressure and heat zone distribution data of the bedridden area, combines the respiratory micro-movement amplitude collected by the vital sign monitoring device, and triggers an image recognition AI model to analyze multiple key body areas of the bedridden person based on the spatial superposition rule of the body pressure and heat zone distribution data and the respiratory micro-movement amplitude to generate a retention risk index for the bedridden person; The generation module is further configured to extract, based on the retention risk index, areas of excess body pressure distribution corresponding to critical values in the physiological parameter abnormality library, and generate bed rest risk warning information through the image recognition AI model in combination with the duration of the abnormal breathing interval and the heartbeat intensity attenuation; The generating module is further configured to generate a multimodal warning report based on the bed rest risk warning information.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the elderly care service supervision method based on the image recognition AI model algorithm as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for supervising elderly care services based on the image recognition AI model algorithm as described in any one of claims 1 to 7 is implemented.

Images