Robust output method and system for nursing critical pressure area facing incomplete pressure structure scene

Abstract

The application discloses a nursing key pressure area robust output method and system for a pressure structure incomplete scene, and the method comprises the following steps: establishing a topological coordinate reference based on pressure weighted second moment, generating a key pressure area candidate through longitudinal projection and transverse envelope analysis; constructing a dynamic neighborhood based on the candidate area, correcting the reliability of the intra-frame structure characteristics of the candidate area, and outputting a reliability score, an abnormal label and a parameter correction amount; adopting a multi-layer time sequence convolution gate network to cross-frame gate multiple frames of candidate states, combining the reliability correction result to form a multi-level quality stratified output strategy, and correcting the key pressure area output. The application solves the problem of pressure area generation under the condition of structure loss by driving the pure pressure topology without relying on external contour information, guarantees the inter-frame continuity through dynamic neighborhood write-back, realizes the technical improvement from single-frame reliability to cross-frame stability through double-layer quality control, and can be widely applied to intelligent nursing mattresses, pressure sore risk monitoring and ward bedside monitoring systems.

Description

Technical Field

[0001] This invention relates to the field of pressure sensing and analysis technology, and in particular to a robust output method and system for critical pressure zones in nursing care scenarios with incomplete pressure structures. Background Technology

[0002] In scenarios such as long-term bedridden care, pressure ulcer risk warning, intelligent mattress monitoring, and bedside monitoring in wards, pressure sensor arrays have become an important means of obtaining information on the distribution of contact pressure between the human body and the supporting surface. Pressure arrays can continuously and non-invasively reflect the contact state between the human body and the bed surface, and are therefore widely used for sleep posture determination, bed exit monitoring, body movement analysis, and pressure risk assessment. As clinical nursing gradually shifts from coarse-grained observation to refined intervention, nurses are paying increasing attention to the continuous pressure on key risk areas such as the sacrum, heels, and shoulders, the degree of pressure concentration, and the changing trends of these risk areas over time.

[0003] In existing technologies, methods for human posture analysis and pressure zone identification based on pressure arrays can be mainly divided into three categories. The first category converts the pressure matrix into a color heatmap, and then uses classification, detection, or segmentation networks from the field of computer vision (such as the YOLO series, Hyper-YOLO, etc.) to directly output the sleeping posture category or body part boundaries. Based on the identification results, the average and peak pressures of each body part are then statistically analyzed. For example, Chinese patent application CN120708287A discloses a method for sleeping posture and body part identification based on pressure sensors. This method first normalizes, interpolates, and smooths the pressure matrix, then renders it as a heatmap. Subsequently, it uses a YOLOv8 classification network to output the sleeping posture category, and a Hyper-YOLO segmentation network to output the body part boundaries. Finally, it outputs the pressure index for each body part. The advantage of this type of method is that it can output semantically clear body part-level pressure information, but its core relies on relatively complete body part boundaries and clear visual segmentation results, requiring a high degree of completeness in the input pressure structure.

[0004] The second type of method geometrically divides pressure regions based on inflection points, contour boundaries, or fixed bed surface partitions in the pressure distribution. It then statistically analyzes the pressure characteristics of these divided regions and uses machine learning models (such as random forests) to output sleeping posture status or assess sleep quality. For example, Chinese patent application CN118402783A discloses a sleeping posture detection system based on human body pressure distribution. This method extracts human body pressure values ​​from mattress pressure data, detects inflection points in the pressure distribution using sliding windows and first derivatives, divides the human body contour into regions based on these inflection points, extracts representative pressure sensor location points, and combines a personalized sleeping posture database and a random forest model to output sleeping posture status. This type of method has good interpretability, but the region division logic is based on a relatively complete human body contour and representative pressure points. When the pressure structure is missing, problems such as changes in the meaning of regions and distortion of representative points can easily occur.

[0005] The third type of method is based on overall characteristics such as center of gravity, principal axis, and boundary curve to analyze posture or movement. This type of method does not explicitly output body parts, but judges body movement or posture changes through changes in overall pressure distribution. Its advantage is that it has high calculation efficiency, but it is difficult to output key pressure areas with clear nursing significance (such as the sacrococcygeal region), and it is difficult to directly serve pressure ulcer risk assessment.

[0006] While the aforementioned existing technologies can achieve good results under ideal conditions (i.e., intact pressure structure, clear human body contour, and stable posture), they generally have the following two prominent problems in real clinical scenarios.

[0007] First, it relies heavily on a complete pressure structure. When patients are in states such as transitioning between lying down, partially out of bed, lying on one side of the bed, partially obscured by blankets, experiencing local sensor failure, or insufficient contact at the edge of the pressure array, the input pressure matrix often exhibits characteristics such as missing structure, incomplete contours, or fragmented contact areas. Under these conditions, the first type of method fails to identify the location or causes boundary jumps because it cannot obtain complete human semantic segmentation results; the second type of method fails to segment the region due to unstable inflection point detection or distortion of representative points; and the third type of method struggles to define and track critical pressure areas with nursing significance. Therefore, how to stably generate critical pressure areas under conditions of incomplete pressure structures has become an urgent technical challenge to be solved.

[0008] Second, the output layer lacks a stable control mechanism for abnormal frames, transitional frames, and incomplete structures. Most existing methods employ a single-frame independent analysis approach, meaning that each frame of pressure data is preprocessed, feature extracted, and output independently, lacking inter-frame state continuity and stability control. When brief pressure structure disturbances occur in consecutive frames (such as instantaneous pressure redistribution during turning over, or localized elevation when the patient adjusts their position), single-frame analysis methods are prone to candidate region drift, boundary jumps, or abrupt changes in output values, resulting in discontinuous and unstable monitoring results over time. For nurses, the true value of the system lies not only in identifying the body part or posture in the current frame, but also in its ability to stably output key pressure areas and their pressure indicators for nursing assessment even under various undesirable conditions, avoiding the impact of frequent output jumps on clinical decisions.

[0009] To address the aforementioned issues, some studies have attempted to post-process the output results using temporal filtering, result smoothing, or state machines. However, these methods typically treat stability control as an independent post-processing step, disconnected from the generation process of the pressure zone, making it difficult to adaptively adjust the stabilization strategy based on the intrinsic state of the pressure topology. Therefore, there is an urgent need for a two-layer stabilization control mechanism that can directly generate critical pressure zones from the pressure topology without relying on the complete human body contour, and integrates frame-level reliability correction and cross-frame temporal gating, to achieve robust output of critical pressure zones under incomplete pressure structure conditions. Summary of the Invention

[0010] To achieve the above-mentioned objectives and other advantages of the present invention, a first objective of the present invention is to provide a robust output method for critical pressure zones in scenarios with incomplete pressure structures, comprising the following steps: Data preprocessing steps: Obtain the original pressure matrix of the bed pressure array, preprocess it, and generate an effective pressure matrix; Pressure topology candidate region generation steps: Based on the effective pressure matrix, establish an overall topological coordinate reference consistent with the current pressure structure. Through longitudinal pressure projection analysis along the main topological direction and width envelope analysis along the transverse topological direction, generate one or more key pressure zone candidates, and simultaneously extract candidate state vectors and pressure index sets describing the candidate regions of the current frame. Dynamic neighborhood parameter write-back step: Define each candidate key pressure area as a dynamic neighborhood containing center, length, width, orientation angle and acceptance threshold, and use the abnormal state label output by the credibility correction module and parameter correction amount to perform frame-level write-back correction of the parameters of the dynamic neighborhood to generate the corrected candidate area parameters. Credibility correction step: Input the candidate state vector into the credibility correction model built on multi-branch XGBoost to obtain the credibility score, abnormal state label and parameter correction amount of the current frame; Temporally stable gating steps: Combine the candidate state vectors, confidence scores, abnormal state labels and parameter corrections of multiple consecutive frames into a temporal state sequence, input it into a stable gating model built on a multi-layer temporal convolutional gating network, and obtain the updated gating value, state category and hold control signal used to control the output result; Key pressure zone output steps: Based on the confidence score, abnormal state label, updated gate value, state category and hold control signal, the final key pressure zone and its pressure index generated based on the corrected candidate zone parameters are corrected and output.

[0011] Furthermore, the pressure topology candidate region generation step includes: Calculate the pressure weighted center of gravity of the current pressure structure and construct the pressure weighted second-order moment matrix. By performing eigendecomposition on the matrix, obtain the principal topological axis direction and the transverse topological axis direction. In this way, establish a principal axis relative coordinate system consistent with the current pressure structure. At the same time, calculate the principal axis stability parameters used to characterize the stability of the principal topological direction. Under the main axis relative coordinate system, a longitudinal pressure projection is constructed along the main topology direction, and longitudinal features including the number of main peaks, peak positions, inter-peak distances, valley depths, and peak widths are extracted from the projection. In the relative coordinate system of the principal axis, a lateral width envelope is constructed at each position along the principal topology direction, and lateral features including width extrema, width contraction segments, and width abrupt change segments are extracted; Based on the longitudinal and lateral features, one or more key pressure zone candidates are generated, and for each candidate zone, its center position, initial length along the main axis, initial width along the lateral direction, orientation angle, and initial acceptance threshold are defined. The set of pressure indicators includes at least local average pressure, local peak pressure, pressure-bearing area, and pressure concentration.

[0012] Furthermore, the candidate state vector includes: principal axis stability parameters, number of main peaks, inter-peak distance, valley depth, width extreme value parameters, effective contact area, edge contact ratio, direction change, center drift, local gradient concentration, and pressure concentration.

[0013] Furthermore, in the dynamic neighborhood parameter write-back step, the dynamic neighborhood parameter correction is expressed as: Among them, the corrected center ,length ,width and threshold The following formula is used to fuse the parameter correction amount output by the credibility correction step from the previous frame or initial parameters: in, This is the center offset correction amount. This is the length correction amount. This is the width correction amount. For threshold correction, direction angle It remains unchanged within a single frame.

[0014] Furthermore, the credibility correction model includes: A feature input layer is used to receive the candidate state vector; The feature standardization layer is used to standardize features of different dimensions. A credibility regression sub-model, built on XGBoost, is used to output the credibility score of the current frame, which is mapped to the [0,1] interval; A parameter-corrected regression sub-model, built on XGBoost, is used to output the parameter correction amount; An abnormal state classification sub-model, built on XGBoost, is used to output the abnormal state labels, which include at least normal state, mild abnormal state, and severe abnormal state.

[0015] Furthermore, the credibility correction step also includes a partitioning decision based on the credibility score: When the confidence score is greater than or equal to the first threshold, it is determined to be a high confidence interval, and the correction result of the current frame is directly used for updating; When the confidence score is between the second threshold and the first threshold, it is determined to be in the medium confidence interval, and the current frame result needs to be judged in combination with the result of the time-stability gating step; When the confidence score is less than the second threshold, it is determined to be in a low confidence interval, and the current frame result is not used for updating.

[0016] Furthermore, when applying the parameter correction amount in the dynamic neighborhood parameter write-back step, the correction magnitude is also controlled based on the abnormal state label: When the abnormal status label is in the normal status, the parameter correction amount participates in the write-back at a 100% ratio; When the abnormal status label is a mild abnormal status, the parameter correction amount is reduced by a scaling factor of less than 100% before being used for write-back. When the abnormal status label is a severe abnormal status, the parameter correction amount is not involved in the write-back.

[0017] Furthermore, the time-stability gating module includes a time-series input layer, a first one-dimensional convolutional layer, a first batch of normalization layers, a first ReLU activation layer, an expanded one-dimensional convolutional layer, a second batch of normalization layers, a second ReLU activation layer, a global average pooling layer, a fully connected layer, and a multi-head output layer connected in sequence. The temporal input layer is used to combine the candidate state vectors, confidence scores, abnormal state labels and parameter corrections of multiple consecutive frames into a temporal tensor. The global average pooling layer is used to compress the temporal features of the entire time window into a fixed-length representation; The multi-head output layer is used to output the update gate value, state category, and hold control signal in parallel.

[0018] Furthermore, in the timing stability gating step, the following output logic is executed based on the updated gating value, state category, and hold control signal: When it is determined that an update is allowed and the state is stable, accept the correction result of the current frame and output the fine-grained key pressure area; When it is determined that the previous result should be maintained, the critical pressure area is not updated, and the output result of the previous stable frame is maintained. When a rollback is determined, discard the fine-grained results of the current frame and output the predefined coarse-grained critical pressure area; When multiple consecutive frames are in a severely abnormal state, the output of the critical pressure area is temporarily suppressed.

[0019] Furthermore, the pressure indicators output in the key pressure zone output step include one or more of the following: local average pressure, local peak pressure, pressure area, and pressure concentration degree; the output information also includes the credibility score used to characterize the reliability of the output results, and the state category used to characterize the current scenario.

[0020] A second objective of this invention is to provide a robust output system for critical pressure zones in scenarios with incomplete pressure structures, for performing the aforementioned method, the system comprising: The data preprocessing module is used to preprocess the original pressure matrix to generate an effective pressure matrix; The pressure topology candidate region generation module is used to generate key pressure zone candidates and candidate state vectors based on the effective pressure matrix. The dynamic neighborhood parameter write-back module is used to perform frame-level write-back correction of the dynamic neighborhood parameters of the candidate region. The credibility correction module, which embeds a credibility correction model built on a multi-branch XGBoost, is used to output credibility scores, abnormal state labels, and parameter correction amounts. The temporally stable gating module embeds a stable gating model built on a multi-layer temporally convolutional gating network, which is used to output the update gating value, state category and hold control signal; The critical pressure zone output module is used to output the final critical pressure zone and its pressure index based on the control information output by each module.

[0021] A third objective of the present invention is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0022] A fourth objective of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0023] Compared with the prior art, the beneficial effects of the present invention are: The method proposed in this invention does not rely on semantic recognition of human body parts and extraction of complete human body contours. It directly calculates the center position, longitudinal peak-valley structure, lateral envelope width, principal axis direction angle, and acceptance threshold through pressure topology, and directly generates key pressure areas with clinical nursing significance. This solves the problem of traditional methods relying on the complete human body pressure structure. It can still stably output effective pressure areas in abnormal scenarios such as incomplete pressure distribution, missing human body contours, and transitions during turning over, which significantly improves the algorithm's scenario adaptability and robustness.

[0024] This invention represents the candidate region of the critical pressure zone as an adjustable dynamic neighborhood, and modifies the parameters of the center, length, width and acceptance threshold. Through real-time write-back detection of the neighborhood parameters at the frame level, it realizes adaptive correction and continuous tracking of the candidate region, transforming the traditional single-frame independent detection into cross-frame state continuation. It effectively solves the problems of target drift, state change and inter-frame interruption that occur in the existing scheme under the conditions of local pressure loss and boundary disturbance, and ensures the continuity and stability of pressure zone tracking.

[0025] This invention constructs a two-layer stable control system of single-frame correction and temporal constraints: the first layer filters out the influence of single-frame structural defects or interference noise through frame reliability assessment, abnormal state identification and parameter adaptive correction; the second layer introduces cross-frame temporal stability gating, which determines the update strategy of the pressure zone parameters in combination with the evolution relationship of continuous frames, realizing the technical improvement from single-frame detection to continuous stable output, so that the algorithm output has real-time response capability while meeting the strict requirements of nursing monitoring for continuity, robustness and interpretability.

[0026] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart outlining the overall scheme for a robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures; Figure 2 A flowchart illustrating a robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures; Figure 3 Here is a flowchart of the data preprocessing process; Figure 4 Flowchart for generating pressure topology candidate regions; Figure 5 A diagram of the reliability correction module structure; Figure 6 This is a diagram of a multi-layer temporal convolutional gated network architecture; Figure 7 A block diagram of a robust output system for critical pressure zones in nursing care scenarios with incomplete pressure structures; Figure 8 This is a schematic diagram of a computer device. Figure 9 This is a schematic diagram of a computer-readable storage medium. Detailed Implementation

[0028] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0029] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0030] The drawing numbers in this application are only used to distinguish the steps in the scheme and are not used to limit the execution order of the steps. The specific execution order is as described in the specification.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0032] This invention provides a technical solution that can stably output key pressure zones and their pressure indicators even when the pressure structure obtained from bed surface pressure acquisition is incomplete, the human body contour is incomplete, or the pressure is in a transitional state. Addressing the needs of nursing care and pressure ulcer risk analysis, this invention achieves stable generation, updating, and output of key pressure zones through pressure topology candidate region generation, dynamic neighborhood parameter write-back for frame-level reliability correction, and a two-layer reliability control using a multi-layer temporal convolutional gating network. The specific solution is as follows: Example 1 A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, such as... Figure 1 , Figure 2 as well as Figure 7 As shown, it includes the following steps: S100, Data preprocessing steps: Obtain the original pressure matrix of the bed surface pressure array, preprocess it, and generate an effective pressure matrix; For the time of input The original pressure matrix The output is the effective pressure matrix. .in, Indicates the row coordinate index of the pressure array. Indicates the column coordinate index of the pressure array. This indicates the sampling time or frame number. The module's internal components include a baseline removal unit, a threshold filtering unit, a connected component filtering unit, and a mild smoothing unit. For example... Figure 3 As shown, the baseline removal unit is used to remove empty bed background values ​​and slowly changing environmental disturbances; the threshold filtering unit is used to remove noise points below the minimum effective pressure threshold; the connected component filtering unit is used to delete isolated noise points that are too small in area, too short in duration, or clearly separated from the contact area of ​​the main body; and the mild smoothing unit is used to reduce the impact of random spike noise on peak and valley extraction.

[0033] The key parameters in this step include the effective pressure threshold. (In this embodiment, 5-10 g / cm² is used) Minimum connected region area threshold (This embodiment uses 4-9 effective sensing units), edge noise suppression threshold (In this embodiment, 3-5 g / cm² is used) and the width of the smooth window. (This embodiment uses a 3×3 to 5×5 window). The above parameter settings ensure that the preprocessing results maintain the continuity of the contour while avoiding smoothing out locally compressed structures.

[0034] This step primarily addresses the interference caused by baseline drift, environmental noise, discrete interference points, and local weak contact points in the original pressure matrix, providing stable input for subsequent topology extraction and candidate region generation. Its design principle involves effective contact screening of the original signal, preserving as much as possible the spatial distribution characteristics of the actual pressure structure, and avoiding topological distortion due to excessive smoothing or over-pruning.

[0035] S200, Pressure Topology Candidate Region Generation Step: Based on the effective pressure matrix, establish an overall topological coordinate reference consistent with the current pressure structure. Through longitudinal pressure projection analysis along the main topological direction and width envelope analysis along the transverse topological direction, generate one or more key pressure zone candidates, and simultaneously extract candidate state vectors and pressure index sets describing the candidate regions of the current frame. This step primarily enables the generation of candidate critical pressure zones with nursing significance based on the current pressure structure when the human body contour is incomplete, the posture is in a transitional state, or local sensing structures are missing. Its design principle is that it does not directly rely on the semantic boundaries of the human body, but rather characterizes the pressure structure through the overall topological direction, longitudinal pressure distribution, and lateral contact width, thereby establishing the candidate region on the basis of the pressure topology. The pressure topology candidate region generation process is as follows: Figure 4 As shown.

[0036] In some embodiments, the pressure topology candidate region generation step includes: First, establish the overall topological coordinate reference of the current pressure structure using pressure-weighted second-order moments, and then calculate the pressure-weighted center of gravity. This is to avoid the sensitivity of fixed bed coordinates to attitude changes and contour loss. And further construct the second-order moment matrix: By performing eigenvalue decomposition on the second-order moment matrix, the eigenvector corresponding to the largest eigenvalue is taken as the direction of the principal topological axis. The eigenvector corresponding to the smallest eigenvalue is taken as the direction of the horizontal topological axis. This yields the principal topological direction vector and the transverse topological direction vector, thereby establishing a relative coordinate system consistent with the current pressure structure. The significance of this relative coordinate system lies in freeing subsequent analysis from dependence on fixed bed surface coordinates and the complete human body contour. To characterize the stability of the principal topological direction, principal axis stability parameters are constructed. ,in and These are the maximum and minimum eigenvalues ​​of the second-order moment matrix, respectively. To avoid tiny constants with a denominator of zero.

[0037] After obtaining the relative coordinate system, map the effective pressure points to the principal axis coordinate system: Constructing longitudinal pressure projection along the main topological direction The system extracts longitudinal features from the projection, including the number of main peaks, peak positions, inter-peak distances, valley depths, peak widths, peak areas, and tail attenuation lengths, to identify the main pressure-bearing sections and low-pressure transition sections on the pressure topology.

[0038] Construct a width envelope in the horizontal direction. For each vertical position, extract the lateral envelope features, including the width extreme value position, width contraction segment, width abrupt segment, and width continuity; modify the vertical candidate segments so that the generation of candidate regions does not depend on the complete contour segmentation, but on the topological width structure.

[0039] The longitudinal features mainly reflect the pressure distribution along the overall extension direction, while the lateral features mainly reflect the degree of contact expansion at different longitudinal positions. The module generates one or more candidate key pressure zones based on the peak-valley transition relationship, width extreme value relationship, and the degree of local pressure gradient concentration.

[0040] Each candidate region is centered. Initial length along the principal axis Initial width along the lateral direction Direction angle Initial acceptance threshold Local average pressure Local peak pressure and the area under pressure Parameters are described. Among them, It is determined jointly by the main peak center, the peak-valley transition point, and the local gradient high value region. Determined by the vertical peak width and the distance between adjacent valleys. Related to the local width envelope value, Inherit the main topological direction or the local main direction. These parameters are used to define the initial acceptance conditions for the candidate region in the current frame. The specific calculations for these four parameters are explained below: Central position Calculation: First, define the first... The location of the main peak is The positions of its left and right adjacent valley values ​​are respectively and , No. The proportion of each main peak at a given threshold The width of the lower peak is The lateral width envelope value near the main peak is .

[0041] At this point, the local pressure gradient energy density along the transverse weighting center is: ;in, Indicates the relationship with the first The horizontal local search intervals corresponding to each main peak.

[0042] The longitudinal center position is ;in, , , ,in The peak value is the main peak value. and These represent the valley depths relative to the left and right valleys of the main peak.

[0043] Therefore, the first The center location of each candidate area It can be represented as: ;in, The unit vector in the main topological direction. It is a unit vector in the horizontal topological direction.

[0044] Initial length along the principal axis Calculate: the first The initial length of each candidate region along the principal axis Determined by the combination of peak-valley spacing and peak width: ;in, and For length combination coefficients, , The main axis length primarily reflects the overall coverage span between peaks and valleys, while the peak width is only used as a correction term for the local concentration.

[0045] Initial width in the lateral direction Calculate: the first The initial width of each candidate region along the lateral direction Determined by the combined width envelope near the main peak and the local average width: in, and For width combination coefficient, and The width is biased towards the immediate lateral expansion at the location of the main peak, while retaining a portion of the average width of the candidate segments as a smoothing constraint.

[0046] Direction angle Calculate: the first Candidate region orientation angle Inheriting the main topological direction and making fine adjustments based on local peak-valley skew, it is represented as: in, and These represent the main topological direction vectors, respectively. Components in the row and column directions, This is a local directional correction coefficient to prevent excessive pull on the main axis due to lateral width asymmetry.

[0047] Initial acceptance threshold Calculate: the first Initial acceptance threshold for each candidate region Represented as: ;in, , , , Indicates the first Local average pressure and local peak pressure of each candidate region.

[0048] Based on the above steps, the geometric and initial control parameters of the pressure topology candidate region are generated, and the candidate state vector is output simultaneously. It is used for subsequent reliability correction and timing stability gating.

[0049] S300, Dynamic Neighborhood Parameter Write-back Step: Define each candidate key pressure area as a dynamic neighborhood containing center, length, width, orientation angle and acceptance threshold, and use the abnormal state label output by the credibility correction module and parameter correction amount to perform frame-level write-back correction of the parameters of the dynamic neighborhood to generate the corrected candidate area parameters. This step primarily addresses issues that may arise after the initial formation of candidate regions, such as excessively large or small ranges, offsets, or unstable acceptance thresholds under conditions of structural defects, local perturbations, and transitional frames. Its design principle is to treat the candidate region as an adjustable dynamic neighborhood. By writing back and correcting the core parameters of the neighborhood, the candidate region can stably change in accordance with variations in the pressure structure, avoiding the generation of inter-frame fragmentation regions caused by completely regenerating the region in each frame.

[0050] In terms of design, each candidate region is defined as a dynamic neighborhood. .in, Indicates the neighborhood center, Indicates the length along the main topological direction. Indicates the width along the horizontal topological direction. Indicates the neighborhood orientation angle. This represents the acceptance threshold for the neighborhood. To achieve adaptive adjustment, the module introduces a neighborhood length correction factor. Neighborhood width correction amount Center offset correction amount and accept threshold correction amount The corrected neighborhood parameters are expressed as follows: In the specific implementation, the dynamic neighborhood parameter write-back module receives initial parameters from the pressure topology candidate region generation module. , , , , And local pressure indicators, as well as the set of parameter corrections from the reliability correction module. During single-frame write-back, the orientation angle It remains unchanged and is only updated when the pressure topology candidate region generation is re-executed in the next frame.

[0051] This step rewrites the corrected parameters back into the candidate region definition process and recalculates the region range and related pressure indicators accordingly. This avoids situations where the candidate region drifts, jumps, or is incorrectly updated due to relying solely on the initial numerical chain output under conditions of incomplete contours and transition frames, thus achieving stable output of the final result.

[0052] S400, Credibility Correction Step: Input the candidate state vector into the credibility correction model built on multi-branch XGBoost to obtain the credibility score, abnormal state label and parameter correction amount of the current frame; This step primarily addresses potential misjudgments in single-frame analysis results during real-world clinical scenarios, such as partial loss of pressure structures, incomplete contact between the trunk or lower limbs, wrinkles or localized interference from bedding, or sitting or partially off the bed. Its design principle involves transforming the current frame's candidate state into a structured feature vector, using a frame-level credibility assessment method to determine the reliability of the current result, and simultaneously outputting correction suggestions for the dynamic neighborhood.

[0053] The input to this step is the candidate state of the current frame, and the output is the credibility score of the current frame. Abnormal status tags and neighborhood parameter correction Candidate state vector ,in Main spindle stability parameters Number of main peaks Interpeak distance, For valley depth, and For the extreme value parameter of width, For effective contact area, For edge contact ratio, For the change in direction, For center drift amount, For local gradient concentration, This refers to pressure concentration. The above parameters are extracted simultaneously by the pressure topology candidate region generation module while outputting the geometric parameters of the candidate regions, and together they describe whether the candidate regions generated in the current frame possess structural consistency, spatial continuity, and nursing significance.

[0054] like Figure 5 As shown, the main structure of the credibility correction model includes a feature input layer, a feature standardization layer, and an XGBoost sub-model. The feature input layer receives the structured features of the current frame; the feature standardization layer performs uniform scaling on parameters of different dimensions; and the three XGBoost sub-models respectively handle credibility regression, parameter correction regression, and abnormal state classification tasks.

[0055] Specifically, the feature input layer is used to receive candidate state vectors. ; The feature standardization layer is used to standardize features of different dimensions, and the output is: ;in, and These are the mean vector and standard deviation vector of the features obtained during the training phase, respectively. The standardized feature vectors are then fed into three XGBoost sub-model branches (confidence regression sub-model, parameter-corrected regression sub-model, and abnormal state classification sub-model), each of which is composed of CART decision trees concatenated in a gradient boosting manner.

[0056] The credibility regression submodel outputs the credibility score of the current frame. The parameter-corrected regression submodel outputs a set of dynamic neighborhood correction values. The abnormal state classification sub-model outputs abnormal state labels. For the first A boosting tree, the output of the tree model can be denoted as: To control model complexity and generalization ability, a maximum tree depth is set for each sub-model. Minimum number of leaf node samples Row sampling ratio Sampling ratio parameter.

[0057] The credibility regression results are expressed as follows: ;in, The number of trees in the credibility regression sub-model. The learning rate is used. CART regression tree, maximum tree depth Minimum number of leaf node samples Learning rate Row sampling ratio Column sampling ratio Its output is normalized after being mapped by the Sigmoid algorithm. An interval is used to represent the credibility of the candidate state of the current frame.

[0058] The parameter-corrected regression results are expressed as follows: ;in, , Adjust the number of trees in the regression sub-model to adjust the parameters. The learning rate is used. CART regression tree, maximum tree depth Minimum number of leaf node samples Learning rate Row sampling ratio Column sampling ratio The function of this branch is to directly convert the candidate state deviation of the current frame into a rewritable neighborhood correction.

[0059] The abnormal state classification results are represented as follows: ;in, The number of trees in the abnormal state classification regression sub-model. The learning rate is used. CART classification tree, maximum tree depth Minimum number of leaf node samples Learning rate Row sampling ratio Column sampling ratio . Three categories are preferred, namely Indicates a normal state. Indicates a mild abnormal condition. This indicates a severely abnormal condition.

[0060] The outputs of the three sub-models are used in subsequent calculations as follows: Credibility Score Used for frame-level reliability partitioning. Defined as a high-confidence interval, Defined as a medium confidence interval, Defined as a low confidence interval. A high confidence interval indicates that the current frame correction result can be directly used for updating; a medium confidence interval indicates that the current frame result needs to be further judged in conjunction with the timing gating result, and the corrected parameters are weighted and fused with the previous stable result; a low confidence interval indicates that the current frame result is not directly updated, and the previous stable result is maintained or the output is rolled back to coarse-grained.

[0061] Parameter correction set Used for writing back parameters in the dynamic neighborhood. Its calculation source is the regression output of the parameter-corrected regression sub-model; specifically, Used to correct the length of the candidate region along the principal axis. Used to correct the width of the candidate region in the lateral direction. Used to correct the center offset of the candidate region Used to correct the accept threshold. The corrected write-back parameters are as follows: To prevent excessively large corrections from causing abrupt changes in the neighborhood, boundary constraints are added to each correction value to limit its range. , , ,in , , .

[0062] Abnormal status label Used for proportional control of parameter correction amount on dynamic neighborhood write-back. When The set of parameter correction values ​​represents the normal state. It directly participates in dynamic neighborhood write-back at a 100% rate; when When representing a mildly abnormal state, the set of parameter corrections is scaled by a scaling factor. Shrink it before participating in the write-back, that is, use Replace the original ;when When indicating a severe abnormal state, the current frame parameter correction set is not included in the write-back, which is equivalent to using... And directly retain the result of the previous stable frame.

[0063] S500, Temporal Stable Gating Step: Combine the candidate state vectors, confidence scores, abnormal state labels and parameter corrections of multiple consecutive frames into a temporal state sequence, input it into a stable gating model built on a multi-layer temporal convolutional gating network, and obtain the updated gating value, state category and hold control signal used to control the output of the result; This step primarily addresses the issue of inconsistent single-frame results but unstable overall output caused by short-term abrupt changes and persistent drift between consecutive frames in clinical scenarios such as turning over, partial bed evacuation, or movement at the bedside. Its design principle utilizes the evolutionary relationships of candidate states within continuous time windows, introducing a multi-layer temporal convolutional gating network to perform cross-frame stabilization gating on whether the current result allows for updating, ensuring that the output meets the continuity requirements for clinical nursing use.

[0064] Design multi-layer temporal convolutional gating networks to enhance temporal state modeling capabilities. For example... Figure 6 As shown, the model includes a temporal input layer, a one-dimensional convolutional layer, a batch normalization layer, a ReLU activation layer, a dilated one-dimensional convolutional layer, a global average pooling layer, a fully connected layer, and a multi-head output layer.

[0065] In this structure, one-dimensional convolutional layers primarily capture local short-term fluctuations, dilated convolutional blocks mainly model state transition relationships over longer time periods, global average pooling layers compress sequence-level information into a fixed-length representation, fully connected layers enhance the gating decision boundary, and multi-head output layers output corresponding update gating values, current state categories, and result-holding control signals. Through this layered design, the temporally stable gating module can effectively utilize the joint evolution of the current frame's confidence score, parameter corrections, and historical candidate states to stably control result updates.

[0066] The module will use the candidate state vectors of the most recent 3 frames. Current frame credibility score Abnormal status label when and the set of parameter corrections Combined into a time-series state sequence: in, Provides numerical evidence as to whether an update was performed in the current frame. Provide a priori classification of the current frame as normal, mildly anomalous, or severely anomalous. Provides information on the neighborhood correction magnitude to be performed in the current frame. Provides the original structured topology state. Since the current time window length is [value missing]... Therefore, the timing-stable gating module sees a quaternion input of 3 consecutive frames at any given time, i.e. , and With three sets of inputs, the model can simultaneously utilize the structural evolution, confidence changes, anomaly evolution, and correction changes of the last three frames to determine whether an update is allowed.

[0067] The temporal input layer is used to arrange the states of the most recent three frames in chronological order to form a temporal tensor. The temporal input dimension for a single frame is set to... ,in The dimension of the candidate state vector is... This indicates the credibility rating dimension. This indicates the dimension of the abnormal state label encoding. This indicates the dimension of the parameter correction, therefore The model input can be represented as: ;in, .

[0068] The first one-dimensional convolutional layer is used to extract local change patterns within a short temporal neighborhood, represented as: in, Indicates the number of output channels. Indicates the kernel length. Indicates the expansion coefficient. BN represents the batch normalization layer, and ReLU represents the linear rectified activation layer.

[0069] The second extended one-dimensional convolutional layer is used to expand the receptive field and capture evolutionary relationships across multiple time steps. Represented as: The third extended one-dimensional convolutional layer can further increase the expansion coefficient to enhance the temporal recognition capability of the turning-over transition and semi-out-of-bed stages. , is represented as: After completing the temporal convolutional encoding, the global average pooling layer compresses the temporal features of the entire time window into a fixed-length representation, using global average pooling concatenated with the features at the end of the time window: in, This indicates that global average pooling is performed over the entire time window to obtain a compressed representation of the overall trend of the most recent 3 frames; This represents the characteristics of the last time step.

[0070] Subsequently, the fully connected layer further forms a shared timing representation for gating and discrimination: Based on this, the multi-head gating output layer generates and updates the gating value. Current status category and maintain control signals for results .

[0071] Among them, updating the gating value ,for The numerical gating results within the interval are used to determine whether the current fine-grained critical pressure zone is allowed to be updated. Then it is determined that fine-grained updates are allowed, if Then it is determined that the boundary is being maintained. Then it is determined that fine-grained updates are not allowed; current state category Used to mark the current state (stable state, turning-over transition state, semi-out-of-bed state, or boundary movement state); the result maintains the control signal. It is used to control the output state of the result (maintain the previous stable result, accept the current correction result, or fall back to the coarse-grained critical pressure zone output).

[0072] When the system determines and To maintain a stable state, at the same time When the system determines that the current correction result is acceptable, it is allowed to use the corrected dynamic neighborhood to generate new fine-grained critical pressure regions; when the system determines... or It is a transition state, and at the same time When the system determines that it should maintain the previous stable result, it does not output a new fine-grained result, but instead maintains the previous stable result; when the system determines... , For severe abnormal related states (semi-out-of-bed state or boundary movement state), or When a system is determined to revert to coarse-grained output, it reverts to output in the coarse-grained critical compression zone. When multiple consecutive frames are in a state of severe abnormality and meet the conditions for leaving the bed or severe structural loss, the output in the critical compression zone is temporarily suppressed. Through this step, the system achieves a performance improvement from single-frame reliability to stable reliability across frames, resolving clinical misjudgments such as false outputs that are reliable in a single frame but discontinuous in time, erroneous updates caused by short-term peak-valley structural instability during the turning transition, short-term abrupt changes in candidate region parameters but with a stable overall trend, and misjudgments caused by short-term residual candidate regions before and after leaving the bed.

[0073] S600, Key Pressure Zone Output Step: Based on the confidence score, abnormal state label, updated gate value, state category, and hold control signal, the final key pressure zone and its pressure index generated based on the corrected candidate zone parameters are corrected and output.

[0074] This step primarily transforms the aforementioned analysis and control results into output information that nursing staff can directly use. Its design principle is to simultaneously output pressure indicators directly related to nursing judgments while ensuring the stability of the spatial area, enabling the results to directly serve pressure ulcer risk assessment and nursing intervention.

[0075] The inputs for this step include: the corrected critical pressure zone parameters from the dynamic neighborhood parameter write-back module. This is used to determine the location, orientation, and spatial extent of the final output region; local pressure indices derived from the candidate region generation module or recalculated based on the corrected neighborhood. This is used to generate average pressure, peak pressure, pressure-bearing area, and pressure concentration; the credibility score comes from the credibility correction module. and abnormal status labels This is used to indicate the reliability and anomaly attributes of the current output result; the updated gate value comes from the time-series stability gating module. Status labels and maintain control signal This is used to determine whether the current output is a fine-grained update, a stable hold, or a coarse-grained rollback.

[0076] The output of this step includes the final key pressure zone information (location, direction, length, width, and boundary range), corresponding area pressure information (average pressure, peak pressure, pressure area, and pressure concentration), output status labels (stable, transitional, semi-out-of-bed, or boundary movement status), and the current result reliability.

[0077] The output strategy employs four modes: fine-grained update mode, stability maintenance mode, coarse-grained rollback mode, and output suppression mode. Fine-grained update mode is suitable for precise nursing analysis in stable states; stability maintenance mode is suitable for transitional states or boundary confidence intervals, used to maintain the previous stable key pressure area unchanged; coarse-grained rollback mode is suitable for scenarios with low confidence in the current frame, obvious anomalies, or partial structural loss, used to maintain basic continuous output. The output strategy is scored based on confidence level. Result maintain control signal Current status category Update the gate value The evaluation is carried out in stages.

[0078] The output strategy is as follows: Fine-grained update: At that time, directly output fine-grained updates; hour, =Accept the current correction result. =Stable state, Output fine-grained updates; Maintain stability: hour, = Maintain the previous stable result, and keep the output stable; hour, =Accept the current correction result. =Stable state, or The output remains stable. hour, =Accept the current correction result. =The transitional state is stable, and the output remains consistent. hour, = Maintain the previous stable result, and keep the output stable; hour, =Accept the current correction result. The output remains stable. Coarse-grained reduction: hour, =Revert to coarse-grained output; output coarse-grained revert; hour, =Accept the current correction result. Output coarse-grained rollback; hour, =Revert to coarse-grained output; output coarse-grained revert; hour, =Accept the current correction result. Output coarse-grained rollback; It should be noted that coarse-grained critical pressure zone output refers to outputting a simplified pressure zone of a preset fixed size (such as 10cm×10cm) and its statistical pressure indicators (including the average pressure, peak pressure and pressure area within the zone) based on the pressure center or the position of the global maximum pressure point when the system determines that the current frame does not meet the fine-grained output conditions, so as to ensure the basic availability of the system under abnormal scenarios.

[0079] Output suppression: consecutive multiple frames , =Boundary shift or partial bed detachment, output suppression; Through the overall scheme and the collaborative design of each innovative module, this invention establishes a complete technical chain from the original pressure matrix to the output of stable critical pressure areas for nursing care. While retaining the interpretability and high efficiency of structured numerical analysis, this technical chain improves the stability of results in complex scenarios through reliability correction and temporal gating. This enables the invention to continuously output critical pressure areas and their pressure indicators that are of nursing significance, even under conditions of incomplete pressure structures and incomplete human body contours.

[0080] This embodiment provides a robust method for outputting key pressure zones in nursing scenarios with incomplete pressure structures, applied to a bed surface pressure array monitoring system for intelligent nursing mattresses. This method aims to address the problem that existing pressure analysis methods struggle to continuously, stably, and accurately output key pressure zones and their pressure indicators for use in nursing care and pressure ulcer analysis in high-frequency clinical scenarios such as incomplete pressure structures, incomplete human body contours, or when patients are in transitional positions, partially out of bed, partially reclining, partially obstructed, or experiencing local sensor failure.

[0081] Example 2 Based on the same concept, this embodiment also provides a robust output system for critical pressure areas in nursing scenarios with incomplete pressure structures. It applies a robust output method for critical pressure areas in nursing scenarios with incomplete pressure structures provided in Embodiment 1. For a detailed description of the robust output method for critical pressure areas in nursing scenarios with incomplete pressure structures provided in Embodiment 1, please refer to the corresponding description in Embodiment 1, which will not be repeated here.

[0082] It is understood that the robust output system for critical pressure zones in scenarios with incomplete pressure structures provided in this embodiment includes corresponding hardware structures and / or software modules to perform each function in order to achieve the above-mentioned functions. Combining the units and algorithm steps of the examples disclosed in this embodiment, this embodiment can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the technical solution of this embodiment.

[0083] A robust output system for critical pressure zones in nursing care scenarios with incomplete pressure structures, such as... Figures 1-2 ,as well as Figure 7 As shown, the system includes: The data preprocessing module is used to preprocess the original pressure matrix to generate an effective pressure matrix; The pressure topology candidate region generation module is used to generate key pressure zone candidates and candidate state vectors based on the effective pressure matrix. The dynamic neighborhood parameter write-back module is used to perform frame-level write-back correction of the dynamic neighborhood parameters of the candidate region. The credibility correction module, which embeds a credibility correction model built on a multi-branch XGBoost, is used to output credibility scores, abnormal state labels, and parameter correction amounts. The temporally stable gating module embeds a stable gating model built on a multi-layer temporally convolutional gating network, which is used to output the update gating value, state category and hold control signal; The critical pressure zone output module is used to output the final critical pressure zone and its pressure index based on the control information output by each module.

[0084] The pressure topology candidate region generation module outputs two sets of results. The first set consists of the candidate region's geometric and initial control parameters, including... Local pressure indicators in the candidate region And candidate area identification information; among which , , and Used to define the geometric range of the candidate region. Used to define the initial acceptance threshold for the candidate region. , This represents the local average pressure and the local peak pressure. , The first group represents the pressure-bearing area and the degree of pressure concentration. The second group consists of candidate state vectors. It is a structured state description extracted synchronously during the candidate region generation process, including principal axis stability parameters. Number of main peaks Interpeak distance Valley depth Width extreme parameters , Effective contact area Edge contact ratio , change in direction Center drift Local gradient concentration and pressure concentration .

[0085] The credibility correction module outputs a credibility score. Abnormal status tags and the set of parameter corrections ,in These correspond to the correction amounts for the center point position, vertical length, horizontal width, and acceptance threshold of the generated candidate region, respectively. Credibility Score Abnormal status tags and the set of parameter corrections The data is passed to the timing stability gating module to construct the timing state sequence; parameter correction set. The parameters are synchronously sent to the dynamic neighborhood parameter write-back module to generate the corrected candidate region parameters.

[0086] The timing-stable gating module outputs update permission and rollback control information, including update gating values, to the dynamic neighborhood parameter write-back module. Status labels and maintain control signal This is used to control whether the correction result of the current frame takes effect directly, is merged with the previous stable result, or reverts to a coarse-grained output. The final output module uses the corrected candidate region parameters, local pressure index, confidence result, and gating result to form the final output.

[0087] The data sources for the final output module include three parts: first, the corrected candidate region parameters from the dynamic neighborhood parameter write-back module. First, it is used to determine the spatial extent and orientation of the final critical pressure zone; second, it comes from the pressure topology candidate region generation module or from local pressure indices recalculated based on the corrected neighborhood. The first part is used to output average pressure, peak pressure, pressure area, and pressure concentration; the second part is control information from the timing stability gating module and the current frame reliability correction module. It is used to determine whether the current result is a fine-grained update, a stable hold, or a coarse-grained rollback, and to provide the confidence level and status description of the current output result.

[0088] This invention establishes a complete method and process to achieve stable output of critical pressure zones in nursing care under incomplete pressure structures. A bed surface pressure array continuously collects a two-dimensional pressure distribution matrix between the human body and the support surface. The system first preprocesses the raw pressure data to obtain an effective pressure structure that reflects the actual contact state. Then, based on the effective pressure structure, it extracts overall topological information to generate candidate critical pressure zones. Next, it adaptively corrects the neighborhood range and acceptance threshold according to the candidate states. Finally, combining the current frame's reliability judgment and the stability judgment of consecutive frames, it determines the output state of the current candidate zone (direct output, corrected output, maintaining the previous stable result, or reverting to a coarser-grained output). The final output is a stable result for the critical pressure zones in nursing care and their pressure indicators.

[0089] Example 3 A computer device 700, such as Figure 8 As shown, the system includes a memory 710, a processor 720, and a computer program 730 stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of a robust output method for critical pressure zones in scenarios with incomplete pressure structures. For a detailed description of the method, please refer to the corresponding description in the above method embodiments; it will not be repeated here.

[0090] Example 4 A computer-readable storage medium, such as Figure 9 As shown, a computer program is stored thereon. When executed by a processor, the computer program implements the steps of a robust output method for critical pressure areas in scenarios with incomplete pressure structures. For a detailed description of the method, please refer to the corresponding description in the above method embodiments, which will not be repeated here.

[0091] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0092] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

[0093] The apparatus, computer device, and non-volatile computer storage medium and method provided in the embodiments of this specification are corresponding. Therefore, the apparatus, computer device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, computer device, and non-volatile computer storage medium will not be repeated here.

[0094] Those skilled in the art will also know that, besides implementing the controller in the form of purely computer-readable program code, the same functions can be achieved by logically programming the method steps, making the controller take the form of logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers (PLCs), and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices included within it for implementing various functions can also be considered structures within that hardware component. Alternatively, the devices for implementing various functions can be considered as both software units implementing the method and structures within a hardware component.

[0095] The systems, apparatuses, or units described in the above embodiments can be implemented by computer chips or physical entities, or by products with certain functions. For ease of description, the above apparatuses are described separately as various units based on their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in one or more software and / or hardware.

[0096] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0097] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0101] This specification may be described in the general context of computer-executable instructions, such as program units, that are executed by a computer. Generally, program units include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification may also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program units may reside in local and remote computer storage media, including storage devices.

[0102] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0103] The above description is merely an embodiment of this specification and is not intended to limit the scope of one or more embodiments of this specification. Various modifications and variations can be made to one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of one or more embodiments of this specification.

Claims

1. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, characterized in that, Includes the following steps: Data preprocessing steps: Obtain the original pressure matrix of the bed pressure array, preprocess it, and generate an effective pressure matrix; Pressure topology candidate region generation steps: Based on the effective pressure matrix, establish an overall topological coordinate reference consistent with the current pressure structure. Through longitudinal pressure projection analysis along the main topological direction and width envelope analysis along the transverse topological direction, generate one or more key pressure zone candidates, and simultaneously extract candidate state vectors and pressure index sets describing the candidate regions of the current frame. Dynamic neighborhood parameter write-back step: Define each candidate key pressure area as a dynamic neighborhood containing center, length, width, orientation angle and acceptance threshold, and use the abnormal state label output by the credibility correction module and parameter correction amount to perform frame-level write-back correction of the parameters of the dynamic neighborhood to generate the corrected candidate area parameters. Credibility correction step: Input the candidate state vector into the credibility correction model built on multi-branch XGBoost to obtain the credibility score, abnormal state label and parameter correction amount of the current frame; Temporally stable gating steps: Combine the candidate state vectors, confidence scores, abnormal state labels and parameter corrections of multiple consecutive frames into a temporal state sequence, input it into a stable gating model built on a multi-layer temporal convolutional gating network, and obtain the updated gating value, state category and hold control signal used to control the output result; Key pressure zone output steps: Based on the confidence score, abnormal state label, updated gate value, state category and hold control signal, the final key pressure zone and its pressure index generated based on the corrected candidate zone parameters are corrected and output.

2. The robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1, is characterized in that... The pressure topology candidate region generation step includes: Calculate the pressure weighted center of gravity of the current pressure structure and construct the pressure weighted second-order moment matrix. By performing eigendecomposition on the matrix, obtain the principal topological axis direction and the transverse topological axis direction. In this way, establish a principal axis relative coordinate system consistent with the current pressure structure. At the same time, calculate the principal axis stability parameters used to characterize the stability of the principal topological direction. Under the main axis relative coordinate system, a longitudinal pressure projection is constructed along the main topology direction, and longitudinal features including the number of main peaks, peak positions, inter-peak distances, valley depths, and peak widths are extracted from the projection. In the relative coordinate system of the principal axis, a lateral width envelope is constructed at each position along the principal topology direction, and lateral features including width extrema, width contraction segments, and width abrupt change segments are extracted; Based on the longitudinal and lateral features, one or more key pressure zone candidates are generated, and for each candidate zone, its center position, initial length along the main axis, initial width along the lateral direction, orientation angle, and initial acceptance threshold are defined. The set of pressure indicators includes at least local average pressure, local peak pressure, pressure-bearing area, and pressure concentration.

3. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 2, is characterized in that... The candidate state vector includes: principal axis stability parameters, number of main peaks, inter-peak distance, valley depth, width extreme parameters, effective contact area, edge contact ratio, direction change, center drift, local gradient concentration, and pressure concentration.

4. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1, is characterized in that... In the dynamic neighborhood parameter write-back step, the dynamic neighborhood parameter correction is represented as follows: Among them, the corrected center ,length ,width and threshold The following formula is used to fuse the parameter correction amount output by the credibility correction step from the previous frame or initial parameters: in, This is the center offset correction amount. This is the length correction amount. This is the width correction amount. For threshold correction, direction angle It remains unchanged within a single frame.

5. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1, characterized in that... The credibility correction model includes: A feature input layer is used to receive the candidate state vector; The feature standardization layer is used to standardize features of different dimensions. A credibility regression sub-model, built on XGBoost, is used to output the credibility score of the current frame, which is mapped to the [0,1] interval; A parameter-corrected regression sub-model, built on XGBoost, is used to output the parameter correction amount; An abnormal state classification sub-model, built on XGBoost, is used to output the abnormal state labels, which include at least normal state, mild abnormal state, and severe abnormal state.

6. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 5, is characterized in that... The credibility correction step also includes a partitioning decision based on the credibility score: When the confidence score is greater than or equal to the first threshold, it is determined to be a high confidence interval, and the correction result of the current frame is directly used for updating; When the confidence score is between the second threshold and the first threshold, it is determined to be in the medium confidence interval, and the current frame result needs to be judged in combination with the result of the time-stability gating step; When the confidence score is less than the second threshold, it is determined to be in a low confidence interval, and the current frame result is not used for updating.

7. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 5, is characterized in that... When applying the parameter correction amount in the dynamic neighborhood parameter write-back step, the correction magnitude is also controlled based on the abnormal state label: When the abnormal status label is in the normal status, the parameter correction amount participates in the write-back at a 100% ratio; When the abnormal status label is a mild abnormal status, the parameter correction amount is reduced by a scaling factor of less than 100% before being used for write-back. When the abnormal status label is a severe abnormal status, the parameter correction amount is not involved in the write-back.

8. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1, characterized in that... The time-stability gating module includes a time-series input layer, a first one-dimensional convolutional layer, a first batch of normalization layers, a first ReLU activation layer, an expanded one-dimensional convolutional layer, a second batch of normalization layers, a second ReLU activation layer, a global average pooling layer, a fully connected layer, and a multi-head output layer connected in sequence. The temporal input layer is used to combine the candidate state vectors, confidence scores, abnormal state labels and parameter corrections of multiple consecutive frames into a temporal tensor. The global average pooling layer is used to compress the temporal features of the entire time window into a fixed-length representation; The multi-head output layer is used to output the update gate value, state category, and hold control signal in parallel.

9. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1 or 8, characterized in that... In the timing-stabilized gating step, the following output logic is executed based on the updated gating value, state category, and hold control signal: When it is determined that an update is allowed and the state is stable, accept the correction result of the current frame and output the fine-grained key pressure area; When it is determined that the previous result should be maintained, the critical pressure area is not updated, and the output result of the previous stable frame is maintained. When a rollback is determined, discard the fine-grained results of the current frame and output the predefined coarse-grained critical pressure area; When multiple consecutive frames are in a severely abnormal state, the output of the critical pressure area is temporarily suppressed.

10. A robust output method for critical pressure zones in nursing care scenarios with incomplete pressure structures, as described in claim 1, characterized in that... The pressure indicators output in the key pressure zone output step include one or more of the following: local average pressure, local peak pressure, pressure area, and pressure concentration; the output information also includes the credibility score used to characterize the reliability of the output results, and the state category used to characterize the current scenario.

11. A robust output system for critical pressure zones in nursing care scenarios with incomplete pressure structures, characterized in that, The system for performing the method as described in any one of claims 1 to 10 includes: The data preprocessing module is used to preprocess the original pressure matrix to generate an effective pressure matrix; The pressure topology candidate region generation module is used to generate key pressure zone candidates and candidate state vectors based on the effective pressure matrix. The dynamic neighborhood parameter write-back module is used to perform frame-level write-back correction of the dynamic neighborhood parameters of the candidate region. The credibility correction module, which embeds a credibility correction model built on a multi-branch XGBoost, is used to output credibility scores, abnormal state labels, and parameter correction amounts. The temporally stable gating module embeds a stable gating model built on a multi-layer temporally convolutional gating network, which is used to output the update gating value, state category and hold control signal; The critical pressure zone output module is used to output the final critical pressure zone and its pressure index based on the control information output by each module.

12. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 10.

Images