A chronic disease home nursing system based on body shape recognition

Abstract

The application relates to the technical field of image recognition, and relates to a chronic disease home nursing system based on body state recognition. The system comprises a behavior baseline construction module, an activity segmentation module, a feature decoupling module, an index tracking module, a topology mining module and a strategy generation module. The historical images of a target object are subjected to mode induction analysis to obtain a behavior mode baseline. The real-time image sequence of the target object is subjected to activity segment division to obtain a body state image sequence of the target object. The body state image sequence is subjected to time-space confrontation decoupling to obtain feature description. The body state image sequence is subjected to cumulative deviation degree tracking to obtain an evolution index. The evolution index is subjected to dependent topology mining to obtain a mode correlation graph of the evolution index. Based on historical feedback records and the mode correlation graph, the effect of the evolution index is mapped and deduced to obtain a nursing strategy of the target object. The application can improve the efficiency of reference information generation of the chronic disease home nursing system based on body state recognition.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a home care system for chronic diseases based on body posture recognition. Background Technology

[0002] In the application scenario of posture recognition in home care for chronic diseases, existing technologies generally use a generalized human posture baseline for anomaly detection, without constructing a specific reference standard that combines the individual physiological characteristics and activity patterns of chronic disease patients. This results in a low matching degree between the baseline and the patient's actual posture characteristics, making it difficult to accurately capture subtle and gradual posture changes during the progression of chronic diseases. At the same time, existing posture feature extraction technologies mostly use overall feature direct modeling without effectively decoupling the spatiotemporal features of posture image sequences. This makes them susceptible to interference from the home environment and inconsistencies in movement, causing feature confusion and making it impossible to accurately separate the core posture features related to the progression of chronic diseases. Consequently, the misjudgment and false negative rates of subsequent abnormal posture recognition are high.

[0003] Existing home-based care systems for chronic diseases have significant shortcomings in the analysis of body posture indicators and the generation of care strategies. They often analyze identified body posture indicators in isolation, failing to explore the intrinsic dependencies between different indicators and revealing the logical link between changes in body posture and the progression of chronic diseases. Furthermore, the generation of care strategies does not fully integrate historical patient feedback data, outputting standardized suggestions based solely on general pathological models. This lack of adaptability to individual patient pathological characteristics and body posture response patterns results in insufficient targeting of the generated care strategies, failing to effectively match the personalized care needs of different patients. Consequently, it is difficult to achieve precise intervention in the progression of chronic diseases and optimize care outcomes. Therefore, improving the efficiency of home-based care for chronic diseases based on body posture recognition has become an urgent problem to be solved. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides a home-based chronic disease care system based on body posture recognition, characterized in that the system includes a behavioral baseline construction module, an activity segmentation module, a feature decoupling module, an indicator tracking module, a topology mining module, and a strategy generation module, wherein:

[0005] The behavior baseline construction module is used to perform pattern induction analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object.

[0006] The activity segmentation module is used to segment the real-time image sequence of the target object into activity segments to obtain the body image sequence of the target object;

[0007] The feature decoupling module is used to perform spatiotemporal adversarial decoupling on the body posture image sequence to obtain the feature description of the body posture image sequence;

[0008] The indicator tracking module is used to track the cumulative deviation of the body posture image sequence based on the behavioral pattern baseline, and obtain the evolution indicator of the body posture image sequence.

[0009] The topology mining module is used to perform dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index.

[0010] The strategy generation module is used to perform effect mapping and deduction on the evolution indicators based on the historical feedback records of the target object and the pattern association diagram to obtain the nursing strategy for the target object.

[0011] In a preferred embodiment, when the behavior baseline construction module performs pattern inductive analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object, it is specifically used for:

[0012] The pose change sequence of the target object is obtained by tracing the pose change of the target object's historical images.

[0013] Based on the baseline pose change sequence, motion continuity analysis is performed on the historical image to obtain a motion coherence measure of the historical image;

[0014] By performing time-series pattern mining on the historical images, a description of the activity patterns of the target object can be obtained;

[0015] The baseline behavior pattern of the target object is obtained by fusing the baseline posture change sequence, the motion coherence measure, and the activity pattern description.

[0016] In a preferred embodiment, when the activity segmentation module performs activity segmentation on the real-time image sequence of the target object to obtain the body image sequence of the target object, it is specifically used for:

[0017] Temporal structure recognition is performed on the real-time image sequence of the target object to obtain the state change nodes of the real-time image sequence;

[0018] Based on the state change nodes, motion state discrimination is performed on the real-time image sequence to obtain the active segments of the real-time image sequence;

[0019] The connection strength of the activity segments is evaluated to obtain a coherence score for the activity segments;

[0020] Based on the coherence score, the activity segments are optimized and merged to obtain the body image sequence of the target object.

[0021] In a preferred embodiment, when the feature decoupling module performs spatiotemporal adversarial decoupling on the body posture image sequence to obtain a feature description of the body posture image sequence, it is specifically used for:

[0022] Multi-granularity feature extraction is performed on the body posture image sequence to obtain the hierarchical features of the body posture image sequence;

[0023] The hierarchical features are subjected to hierarchical learning to obtain the hierarchical baseline of the hierarchical features;

[0024] Based on the hierarchical baseline, the hierarchical features are bidirectionally competitively aligned to obtain the alignment baseline representation of the hierarchical features;

[0025] Based on the alignment baseline representation, attention decomposition is performed on the body posture image sequence to obtain the weight distribution of the body posture image sequence;

[0026] Based on the weight distribution, the hierarchical features are weighted and reorganized to obtain the feature components of the hierarchical features;

[0027] The feature components are structured and encoded to obtain the feature description of the body image sequence.

[0028] In a preferred embodiment, when the feature decoupling module performs bidirectional competitive alignment of the hierarchical features based on the hierarchical baseline to obtain the alignment baseline representation of the hierarchical features, it is specifically used for:

[0029] The interaction intensity structure of the hierarchical baselines is performed to obtain the initial competition coefficient matrix of the hierarchical baselines;

[0030] Based on the initial competition coefficient matrix, selective baseline responses are performed on the hierarchical features to obtain the response distribution of the hierarchical features;

[0031] The temporal consistency of the response distribution is evaluated to obtain a confidence score for the response distribution;

[0032] Based on the confidence score, the initial competition coefficient matrix is ​​adaptively optimized for competition intensity.

[0033] Based on the optimized competition coefficient matrix, the hierarchical features are reweighted and aggregated to obtain the aligned baseline representation of the hierarchical features.

[0034] In a preferred embodiment, when the indicator tracking module performs cumulative deviation tracking on the body posture image sequence based on the behavioral pattern baseline to obtain the evolution indicator of the body posture image sequence, it is specifically used for:

[0035] The baseline of the behavioral pattern is spatiotemporally matched with the body posture image sequence to obtain the baseline matching distribution of the body posture image sequence;

[0036] Based on the baseline matching distribution, motion trajectory deviation analysis is performed on the body posture image sequence to obtain the spatiotemporal residuals of the body posture image sequence;

[0037] Multi-scale region aggregation is performed on the spatiotemporal residuals to obtain a saliency region map of the spatiotemporal residuals;

[0038] Based on the salient region map, the temporal coherence of the body image sequence is identified and filtered to obtain the abnormal motion trajectory of the body image sequence;

[0039] The cumulative deviation of the abnormal motion trajectory is obtained by performing pattern intensity integration on the abnormal motion trajectory.

[0040] Based on the accumulated bias, a structured semantic description is performed on the body posture image sequence to obtain the evolution index of the body posture image sequence.

[0041] In a preferred embodiment, when the index tracking module performs mode intensity integration on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory, it is specifically used for:

[0042] The abnormal motion trajectory is mapped to manifold coordinates to obtain a parameter representation of the abnormal motion trajectory;

[0043] Based on the parameter representation, anomaly evaluation sampling is performed on the abnormal motion trajectory to obtain the abnormal intensity sequence of the abnormal motion trajectory;

[0044] Local motion feature analysis is performed on the parameter representation to obtain the local curvature sequence and velocity change rate sequence of the parameter representation;

[0045] Based on the abnormal intensity sequence, the local curvature sequence, and the velocity change rate sequence, a composite weighted integral is performed on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory.

[0046] In a preferred embodiment, when the topology mining module performs dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index, it is specifically used for:

[0047] The feature description is expanded in feature space to obtain a high-dimensional semantic manifold of the feature description;

[0048] Based on the high-dimensional semantic manifold, the evolution index is projected onto the manifold path to obtain the spatiotemporal path of the evolution index.

[0049] Morphological analysis of the spatiotemporal path yields a heterogeneous distribution of curvature.

[0050] Based on the aforementioned curvature heterogeneous distribution, critical surface identification is performed on the high-dimensional semantic manifold to obtain the folded topology of the high-dimensional semantic manifold.

[0051] Homotopy path compression is performed on the folded topology to obtain a simplified homotopy skeleton of the folded topology.

[0052] Based on the simplified homotopy skeleton, the spatiotemporal path is topologically integrated to generate the pattern association graph of the evolution index.

[0053] In a preferred embodiment, when the topology mining module performs homotopy path compression on the folded topology to obtain a simplified homotopy skeleton of the folded topology, it is specifically used for:

[0054] The redundancy of the folded topology is evaluated to obtain the redundancy of the folded topology.

[0055] Based on the redundancy, a visual information flow simulation is performed on the folded topology to obtain the information flow intensity map of the folded topology.

[0056] Based on the information flow intensity map, density peak aggregation is performed on the folded topology to obtain the node clusters of the folded topology.

[0057] Based on the node clusters, the shortest path connection is performed on the folded topology to obtain the simplified homotopy skeleton of the folded topology.

[0058] In a preferred embodiment, when the strategy generation module performs effect mapping deduction on the evolution indicators based on the historical feedback records of the target object and the pattern association graph to obtain the nursing strategy for the target object, it is specifically used for:

[0059] Based on the historical feedback records of the target object, knowledge transfer is performed on the pattern association graph to obtain an experience-weighted causal graph of the pattern association graph.

[0060] Based on the aforementioned experience-weighted causal graph, a virtual intervention response simulation is performed on the evolution indicators to obtain the intervention prediction distribution of the evolution indicators;

[0061] The intervention prediction distribution is subjected to feasibility filtering, and the filtered distribution is prioritized to obtain the intervention priority sequence of the intervention prediction distribution;

[0062] Based on the intervention priority sequence, the intervention prediction distribution is integrated with personalized context to obtain the nursing strategy for the target object.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. This invention utilizes a feature extraction technique that decouples personalized behavioral pattern baseline construction with spatiotemporal adversarial decoupling to accurately capture subtle postural changes related to chronic diseases, improving the accuracy and completeness of postural feature description. Based on a dynamic tracking mechanism using cumulative deviation, it enables precise quantification and continuous monitoring of postural evolution indicators, significantly improving the accuracy of identifying abnormal postural evolution trends and providing reliable data support for chronic disease progression assessment.

[0065] 2. This invention constructs a pattern association graph of evolutionary indicators using topology mining technology, clearly revealing the intrinsic dependencies between postural indicators. Combined with patient historical feedback records, it enables personalized deduction of nursing strategies. This technology can improve the adaptability of nursing strategies to individual patient pathological characteristics and postural response patterns, enhance the pertinence and effectiveness of nursing interventions, help optimize the effects of home care for chronic diseases, and improve the implementation value of nursing plans. Attached Figure Description

[0066] Figure 1 This is a system architecture diagram of a home care system for chronic diseases based on body posture recognition, provided in an embodiment of the present invention.

[0067] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments belong to some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0069] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “said” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0070] Depending on the context, the word "if" or "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0071] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0072] In practice, the server-side equipment deployed in a posture recognition-based chronic disease home care system may consist of one or more devices. This posture recognition-based chronic disease home care system can be implemented as: a business instance, a virtual machine, and hardware devices. For example, this posture recognition-based chronic disease home care system can be implemented as a business instance deployed on one or more devices in a cloud node. Simply put, this posture recognition-based chronic disease home care system can be understood as software deployed on a cloud node, used to provide posture recognition-based chronic disease home care services to various user terminals. Alternatively, this posture recognition-based chronic disease home care system can also be implemented as a virtual machine deployed on one or more devices in a cloud node. This virtual machine contains application software for managing various user terminals. Alternatively, this posture recognition-based chronic disease home care system can also be implemented as a server composed of numerous identical or different types of hardware devices, with one or more hardware devices configured to provide posture recognition-based chronic disease home care services to various user terminals.

[0073] In terms of implementation, a home care system for chronic diseases based on body posture recognition and the user terminal are mutually compatible. That is, if the home care system for chronic diseases based on body posture recognition is implemented as an application installed on a cloud service platform, then the user terminal is a client that establishes a communication connection with the application; or if the home care system for chronic diseases based on body posture recognition is implemented as a website, then the user terminal is implemented as a webpage; or if the home care system for chronic diseases based on body posture recognition is implemented as a cloud service platform, then the user terminal is implemented as a mini-program in an instant messaging application.

[0074] like Figure 1 The figure shown is a system architecture diagram of a home care system for chronic diseases based on body posture recognition provided in an embodiment of the present invention.

[0075] The chronic disease home care system 10 based on body posture recognition described in this invention can be set up in a cloud server. In terms of implementation, it can be implemented as one or more service devices, or as an application installed in the cloud (e.g., a mobile service operator's server, server cluster, etc.), or it can be developed as a website. Depending on the functions implemented, the chronic disease home care system 10 based on body posture recognition may include a behavior baseline construction module 11, an activity segmentation module 12, a feature decoupling module 13, an indicator tracking module 14, a topology mining module 15, and a strategy generation module 16. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0076] In this embodiment of the invention, in a home-based chronic disease care system based on body posture recognition, each of the above-mentioned modules can be implemented independently and can call other modules. Here, "calling" can be understood as one module connecting to multiple modules of another type and providing corresponding services to those connected modules. The home-based chronic disease care system based on body posture recognition provided by this embodiment of the invention allows for adjustment of the system's applicability by adding modules and directly calling them, without modifying the program code. This enables cluster-based horizontal expansion, achieving the goal of quickly and flexibly expanding the system. In practical applications, the above modules can be set in the same device or different devices, or in virtual devices, such as service instances in a cloud server.

[0077] The following describes, with reference to specific embodiments, each component and specific workflow of a home care system for chronic diseases based on body posture recognition:

[0078] The behavior baseline construction module 11 is used to perform pattern induction analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object.

[0079] In this embodiment of the invention, when the behavior baseline construction module performs pattern inductive analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object, it is specifically used for:

[0080] The pose change sequence of the target object is obtained by tracing the pose change of the target object's historical images.

[0081] Based on the baseline pose change sequence, motion continuity analysis is performed on the historical image to obtain a motion coherence measure of the historical image;

[0082] By performing time-series pattern mining on the historical images, a description of the activity patterns of the target object can be obtained;

[0083] The baseline behavior pattern of the target object is obtained by fusing the baseline posture change sequence, the motion coherence measure, and the activity pattern description.

[0084] When tracing the posture changes of a target object in historical images, the position information of key limb parts of the target object in each historical image is first extracted. These key parts include the head, neck, shoulders, elbows, wrists, hips, knees, and ankles. By comparing the position differences of the same key part in adjacent historical images frame by frame, the position change trajectory of each key part over time is sorted out. The change trajectories of all key parts are integrated in chronological order to form the baseline posture change sequence of the target object.

[0085] When performing motion continuity analysis on historical images based on the reference posture change sequence, the smoothness of posture transition between adjacent historical images is judged according to the change trajectory of each key part in the reference posture change sequence. By checking the connection of the position changes of key parts between adjacent frames, it is confirmed whether the posture change conforms to the continuous characteristics of natural motion. The judgment results of the smoothness of the overall motion process are organized to form a motion continuity measure of historical images.

[0086] When mining the temporal patterns of the historical images, the historical images are first sorted according to the order of time, the types of poses of the target object appearing in different time periods are counted, the duration of each pose type and the switching interval between different pose types are recorded, and the activity frequency distribution of the target object in different time periods of the day and different dates of the week is sorted out. These activity features in the time dimension are summarized and organized to form a description of the activity patterns of the target object.

[0087] When fusing the baseline posture change sequence, the motion coherence measure, and the activity pattern description, the core posture change features in the baseline posture change sequence are first sorted out, key information reflecting the smoothness of motion in the motion coherence measure is extracted, and representative time distribution features in the activity pattern description are selected. These three parts of feature information are then fully integrated to ensure that each part of feature information complements each other and has no duplication or redundancy, and finally a personalized behavior pattern baseline that can comprehensively reflect the behavioral characteristics of the target object is formed.

[0088] The beneficial effects include: by constructing a personalized behavioral pattern baseline and extracting postural features using a spatiotemporal adversarial decoupling approach, the system can accurately capture subtle postural changes related to chronic diseases, enabling precise quantification and continuous monitoring of postural evolution indicators, and providing reliable data support for chronic disease progression assessment; by constructing a pattern association graph of evolution indicators through topology mining, the system can reveal the intrinsic dependencies between postural indicators, and combine this with patient historical feedback records to deduce personalized nursing strategies, thereby improving the adaptability of nursing strategies to individual patient pathological characteristics and postural response patterns, enhancing the pertinence and effectiveness of nursing interventions, optimizing the effect of home care for chronic diseases, and improving the value of nursing program implementation.

[0089] The activity segmentation module 12 is used to segment the real-time image sequence of the target object into activity segments to obtain the body image sequence of the target object;

[0090] In this embodiment of the invention, when the activity segmentation module performs activity segmentation on the real-time image sequence of the target object to obtain the body image sequence of the target object, it is specifically used for:

[0091] Temporal structure recognition is performed on the real-time image sequence of the target object to obtain the state change nodes of the real-time image sequence;

[0092] Based on the state change nodes, motion state discrimination is performed on the real-time image sequence to obtain the active segments of the real-time image sequence;

[0093] The connection strength of the activity segments is evaluated to obtain a coherence score for the activity segments;

[0094] Based on the coherence score, the activity segments are optimized and merged to obtain the body image sequence of the target object.

[0095] When performing temporal structure recognition on a real-time image sequence of a target object, the frame order of the real-time image sequence is first sorted out according to the time sequence. The limb movement features and posture information of the target object in the image are extracted frame by frame. By comparing the differences in the limb movement features of the target object and the changes in posture information between adjacent frames, the key frames of the start of the action, the transition of the action, and the termination of the action are located. These key frames are the state change nodes of the real-time image sequence. Each state change node corresponds to a change in the action state of the target object.

[0096] When judging the motion state of a real-time image sequence based on the state change nodes, the real-time image sequence is divided into multiple continuous image frame intervals with the state change nodes as the boundaries. For each image frame in the interval, the action form, limb movement trajectory and action duration of the target object are analyzed one by one to determine the motion state of the target object in each interval. The continuous image frame intervals corresponding to each motion state are integrated to form an activity segment of the real-time image sequence. Each activity segment corresponds to a stable motion state.

[0097] When evaluating the connection strength of the activity segments, the focus is on the boundary region between two adjacent activity segments. The pose and motion trend of the target object in the last frame of the previous activity segment are extracted, and the pose and motion trend of the target object in the first frame of the next activity segment are extracted. The degree of matching of the poses and the continuity of the motion trends are compared, and the smoothness of the transition between adjacent activity segments is analyzed. The evaluation result of this connection smoothness is transformed into specific descriptive information to form the coherence score of the activity segments.

[0098] When performing boundary optimization and merging of the activity segments based on the coherence score, the coherence scores of all activity segments are first reviewed. For adjacent activity segments whose coherence scores meet the preset connection requirements, the boundary frames of the two segments are adjusted, duplicate or redundant image frames at the boundary are removed, and any missing transition image frames are added to make the transition between the two segments smoother. Then, the optimized adjacent activity segments are merged. For adjacent activity segments whose coherence scores do not meet the preset connection requirements, their independent boundaries are maintained and they are not merged. After the above optimization and merging process, a continuous and complete sequence of body images of the target object is finally obtained.

[0099] The beneficial effects are that by performing temporal structure recognition on real-time image sequences of target objects, the nodes of state change can be accurately located, the motion state can be judged based on the nodes to achieve accurate division of activity segments, the continuity score can be obtained by evaluating the connection strength of activity segments, and then the activity segments can be optimized and merged according to the score, ensuring that the final body image sequence is continuous and complete, and can accurately reflect the motion state change process of the target object, providing high-quality image data support for subsequent extraction and analysis of body features.

[0100] The feature decoupling module 13 is used to perform spatiotemporal adversarial decoupling on the body posture image sequence to obtain the feature description of the body posture image sequence;

[0101] In this embodiment of the invention, when the feature decoupling module performs spatiotemporal adversarial decoupling on the body posture image sequence to obtain the feature description of the body posture image sequence, it is specifically used for:

[0102] Multi-granularity feature extraction is performed on the body posture image sequence to obtain the hierarchical features of the body posture image sequence;

[0103] The hierarchical features are subjected to hierarchical learning to obtain the hierarchical baseline of the hierarchical features;

[0104] Based on the hierarchical baseline, the hierarchical features are bidirectionally competitively aligned to obtain the alignment baseline representation of the hierarchical features;

[0105] Based on the alignment baseline representation, attention decomposition is performed on the body posture image sequence to obtain the weight distribution of the body posture image sequence;

[0106] Based on the weight distribution, the hierarchical features are weighted and reorganized to obtain the feature components of the hierarchical features;

[0107] The feature components are structured and encoded to obtain the feature description of the body image sequence.

[0108] When the feature decoupling module performs bidirectional competitive alignment of the hierarchical features based on the hierarchical baseline to obtain the aligned baseline representation of the hierarchical features, it is specifically used for:

[0109] The interaction intensity structure of the hierarchical baselines is performed to obtain the initial competition coefficient matrix of the hierarchical baselines;

[0110] Based on the initial competition coefficient matrix, selective baseline responses are performed on the hierarchical features to obtain the response distribution of the hierarchical features;

[0111] The temporal consistency of the response distribution is evaluated to obtain a confidence score for the response distribution;

[0112] Based on the confidence score, the initial competition coefficient matrix is ​​adaptively optimized for competition intensity.

[0113] Based on the optimized competition coefficient matrix, the hierarchical features are reweighted and aggregated to obtain the aligned baseline representation of the hierarchical features.

[0114] When performing multi-granularity feature extraction on the body image sequence, different feature extraction granularities are first divided from the whole to the local. The overall granularity focuses on the outline and posture layout of the complete body of the target object, while the local granularity is decomposed into limb part groups, single limb joints, and joint-related regions. For each granularity, the grayscale distribution, edge contours, and motion trend information of the corresponding region in the body image sequence are extracted frame by frame. The features extracted at different granularities are arranged in order from coarse to fine to form the hierarchical features of the body image sequence.

[0115] When performing hierarchical learning on the hierarchical features, the learning process is carried out sequentially according to the granularity of the hierarchical features. For each level of features, the common manifestations and inherent attributes of all features under that level are sorted out, the standard presentation form of features within the same level is summarized, and the stable change pattern of the features at that level in different body image frames is recorded. The sorted and summarized common attributes and standard forms are integrated to obtain the hierarchical baseline of the hierarchical features.

[0116] When structuring the interaction intensity of the hierarchical baselines, the degree of correlation between different hierarchical baselines is analyzed, the correlation relationship of the interaction between each level of baselines is sorted out, and this correlation relationship and the degree of interaction are transformed into clearly characterizable correlation information. This correlation information is arranged and organized according to the preset row and column correspondence rules to form the initial competition coefficient matrix of the hierarchical baselines.

[0117] Based on the initial competition coefficient matrix, when selectively responding to the hierarchical features, the matching degree between each hierarchical feature and different hierarchical baselines is determined according to the initial competition coefficient matrix. Hierarchical baselines that meet the matching degree requirements of the hierarchical features are selected, and the hierarchical features only generate targeted responses to the selected hierarchical baselines. The response of each hierarchical feature to the corresponding hierarchical baseline is recorded to form the response distribution of the hierarchical features.

[0118] When evaluating the temporal consistency of the response distribution, the response distributions corresponding to different frames in the body image sequence are sorted out in chronological order. The changes in the response distributions between adjacent frames are compared to check the continuity and coherence of the response distributions in the temporal dimension. It is determined whether the response distributions of different time segments conform to a stable change pattern. The evaluation results of this temporal continuity and coherence are transformed into specific characterization information to obtain the confidence score of the response distribution. The formula for calculating the confidence score is as follows:

[0119] ;

[0120] In the formula, Score the confidence level. For time segment indexing, This represents the total number of time segments. It is an exponential function. The preset decay rate, For the first The response ratio of the dominant baseline in the hierarchical baselines described in each time segment. The mean of the response ratios of the aforementioned response distribution. The standard deviation of the response ratio of the response distribution is given. The preset zero-prevention constant is used. The preset control penalty form parameters, It is an absolute value function. The switching penalty value for the dominant baseline. This is the maximum value of the switching penalty value.

[0121] When performing adaptive optimization of the competition intensity based on the confidence score, the confidence score is compared with a preset scoring standard. If the confidence score does not meet the preset standard, it indicates that the competition intensity of the initial competition coefficient matrix does not match the current feature alignment requirements. According to the direction of the difference between the confidence score and the preset standard, the strength representation of the corresponding relationship in the initial competition coefficient matrix is ​​adjusted to enhance or weaken the competition degree between the relevant level baselines until the confidence score meets the preset standard, thus obtaining the optimized competition coefficient matrix.

[0122] When reweighting and aggregating the hierarchical features based on the optimized competition coefficient matrix, corresponding weights are assigned to different hierarchical features according to the optimized competition coefficient matrix. The weights are matched with the importance of the hierarchical features in the competitive alignment. All hierarchical features are integrated and summarized according to the assigned weights, and the integrated features are uniformly represented to obtain the alignment baseline representation of the hierarchical features.

[0123] Based on the alignment baseline representation, when performing attention decomposition on the body image sequence, the alignment baseline representation is used as a reference to identify regions and time segments in the body image sequence that can accurately reflect the core features of the target object's body posture. The importance of these regions and time segments in the body posture feature representation is determined, and corresponding attention weights are assigned to different regions and time segments. These weight information are then organized according to the corresponding regions and time segments to obtain the weight distribution of the body image sequence.

[0124] Based on the weight distribution, when the hierarchical features are weighted and recombined, the hierarchical feature parts with weights that meet the requirements are selected according to the weight of different regions and time segments in the weight distribution. The selected hierarchical feature parts are sorted according to their weights. The sorted hierarchical feature parts are integrated and spliced ​​together, and duplicate feature information that occurs during the integration process is removed to form the feature components of the hierarchical features.

[0125] When performing structured encoding on the feature components, a fixed encoding structure and encoding rules are preset. The attribute information, morphological features and correlation relationships of the feature components are converted into encoded information that conforms to the encoding structure according to the encoding rules, ensuring that the encoded information can completely and accurately represent all the key information of the feature components. The converted encoded information is then organized according to the preset structure to obtain the feature description of the body image sequence.

[0126] The sequence of body images of the target object is divided into continuous segments at fixed time intervals, and each segment corresponds to a unique identifier, which is the time segment index.

[0127] The total number of time segments is the number of time segments contained in the sequence of body image segments after statistical division.

[0128] Within each time segment, the dominant baseline in the hierarchical baseline is determined, and the ratio of the number of times the dominant baseline is responded to by hierarchical features within the corresponding time segment to the total number of times the hierarchical features are responded to within that time segment is calculated. This ratio is the response ratio of the dominant baseline.

[0129] Collect the response ratios of the dominant baseline for all time segments, sum these response ratios, and then divide by the total number of time segments. The result is the mean response ratio of the response distribution.

[0130] First, calculate the difference between the dominant baseline response ratio and the mean response ratio for each time segment. Then, square each difference, add all the squared results together, divide by the total number of time segments, and take the square root. The result is the standard deviation of the response ratio of the response distribution.

[0131] Based on the chronic disease type of the target object corresponding to the body image sequence, and combined with the experience of processing body data of similar objects in the past, a fixed value is determined, which is the preset decay rate.

[0132] A very small, fixed positive number is selected to avoid the denominator being zero during the calculation process. This positive number is the preset zero-prevention constant.

[0133] Based on the characteristics of the target object's body posture changes and referring to the conventional settings for processing historical body posture features, a fixed value is determined. This value is the preset control penalty form parameter.

[0134] When the dominant baseline in a time segment is different from the dominant baseline in the previous time segment, a corresponding value is determined based on the degree of difference between the two dominant baselines. This value is the dominant baseline switching penalty value.

[0135] Collect the switching penalty values ​​of the dominant baseline for all time segments, and select the one with the largest value. This value is the maximum value of the switching penalty.

[0136] This content is used to calculate the confidence score, which reflects the consistency and stability of the response distribution over time.

[0137] It combines the deviation of the dominant baseline response ratio within each time segment from the overall response distribution, as well as the penalty for switching the dominant baseline, to obtain a result that measures whether the response distribution conforms to a stable temporal pattern.

[0138] The smaller the deviation of the dominant baseline response ratio from the mean response ratio within a certain time segment, the larger the value of the exponential component corresponding to that time segment will be.

[0139] The smaller the switching penalty value of the dominant baseline, the larger the value of the corresponding penalty item will be.

[0140] The product of these two values ​​increases as the deviation decreases and the switching penalty decreases. After averaging the total number of time segments, the confidence score increases accordingly.

[0141] Conversely, when the degree of deviation increases or the switching penalty value increases, the confidence score will decrease accordingly.

[0142] The beneficial effects are as follows: hierarchical baselines are obtained by performing multi-granular feature extraction and hierarchical learning on body posture image sequences. The hierarchical baselines are then aligned bidirectionally through hierarchical baseline interaction intensity structuring, selective response, temporal consistency evaluation, and competitive coefficient matrix optimization to obtain aligned baseline representations. Attention decomposition is then performed based on the aligned baseline representations to obtain weight distributions. Based on the weight distributions, hierarchical features are weighted and recombined to obtain feature components. Finally, feature descriptions are obtained through structured encoding. The entire process can accurately separate the spatiotemporal features of body posture image sequences, eliminate redundant interference information, and ensure that the feature descriptions can completely and accurately represent the core features of the target object's body posture, providing high-quality feature data support for subsequent evolution index analysis and nursing strategy generation.

[0143] By clarifying the acquisition methods of relevant parameters and calculating confidence scores based on deviations in the response distribution and penalties for switching the dominant baseline, the temporal consistency and stability of the response distribution can be accurately measured. Simultaneously, relying on personalized behavioral pattern baseline construction and feature extraction technology decoupled from spatiotemporal adversarial processes, subtle postural changes related to chronic diseases can be precisely captured, improving the accuracy and completeness of postural feature descriptions. Through a dynamic tracking mechanism of cumulative deviation, precise quantification and continuous monitoring of postural evolution indicators can be achieved, providing reliable data support for chronic disease progression assessment. Furthermore, by constructing a pattern association graph of evolution indicators through topology mining technology, the intrinsic dependencies between postural indicators can be clearly revealed. Combined with patient historical feedback records, personalized deduction of nursing strategies can be achieved, improving the adaptability of nursing strategies to individual patient pathological characteristics and postural response patterns, and enhancing the pertinence and effectiveness of nursing interventions.

[0144] The indicator tracking module 14 is used to track the cumulative deviation of the body image sequence based on the behavior pattern baseline to obtain the evolution index of the body image sequence.

[0145] In this embodiment of the invention, when the indicator tracking module performs cumulative deviation tracking on the body posture image sequence based on the behavioral pattern baseline to obtain the evolution indicator of the body posture image sequence, it is specifically used for:

[0146] The baseline of the behavioral pattern is spatiotemporally matched with the body posture image sequence to obtain the baseline matching distribution of the body posture image sequence;

[0147] Based on the baseline matching distribution, motion trajectory deviation analysis is performed on the body posture image sequence to obtain the spatiotemporal residuals of the body posture image sequence;

[0148] Multi-scale region aggregation is performed on the spatiotemporal residuals to obtain a saliency region map of the spatiotemporal residuals;

[0149] Based on the salient region map, the temporal coherence of the body image sequence is identified and filtered to obtain the abnormal motion trajectory of the body image sequence;

[0150] The cumulative deviation of the abnormal motion trajectory is obtained by performing pattern intensity integration on the abnormal motion trajectory.

[0151] Based on the accumulated bias, a structured semantic description is performed on the body posture image sequence to obtain the evolution index of the body posture image sequence.

[0152] When the index tracking module performs pattern intensity integration on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory, it is specifically used for:

[0153] The abnormal motion trajectory is mapped to manifold coordinates to obtain a parameter representation of the abnormal motion trajectory;

[0154] Based on the parameter representation, anomaly evaluation sampling is performed on the abnormal motion trajectory to obtain the abnormal intensity sequence of the abnormal motion trajectory;

[0155] Local motion feature analysis is performed on the parameter representation to obtain the local curvature sequence and velocity change rate sequence of the parameter representation;

[0156] Based on the abnormal intensity sequence, the local curvature sequence, and the velocity change rate sequence, a composite weighted integral is performed on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory.

[0157] When performing spatiotemporal matching between the behavior pattern baseline and the body image sequence, first, each image frame of the behavior pattern baseline and the body image sequence is matched in chronological order. Then, the pose features and motion features of the target object in the body image sequence are compared with the baseline pose change sequence and motion coherence measure in the behavior pattern baseline frame by frame. The matching degree between the features and the baseline of each frame is recorded. The matching degree of all frames is organized according to time and spatial location to obtain the baseline matching distribution of the body image sequence.

[0158] When performing motion trajectory deviation analysis on the body image sequence based on the baseline matching distribution, the difference between the motion trajectory of the target object in the body image sequence and the trajectory corresponding to the behavior pattern baseline is located according to the matching degree of each frame in the baseline matching distribution. These differences are classified and recorded according to time order and spatial location to form the spatiotemporal residual of the body image sequence.

[0159] When performing multi-scale regional aggregation on the spatiotemporal residuals, different regional scales are divided, such as overall body shape, limb group, and single joint. At each scale, the distribution of spatiotemporal residuals in the corresponding region is summarized, the region where residuals are concentrated is marked, and the regions where residuals are concentrated at different scales are integrated and presented to obtain the saliency region map of the spatiotemporal residuals.

[0160] When performing temporal coherence identification and screening of the body image sequence based on the saliency region map, the distribution of residual concentration regions in the saliency region map on time segments is sorted out, and the motion trajectories corresponding to the residual concentration regions that appear continuously in consecutive time segments are screened out. These trajectories are then arranged in chronological order to obtain the abnormal motion trajectories of the body image sequence.

[0161] When performing manifold coordinate mapping on the abnormal motion trajectory, the relative positional relationship of each position point in the abnormal motion trajectory within the body space of the target object is determined, each position point is converted into corresponding coordinate information, and these coordinate information are arranged in the time sequence of the abnormal motion trajectory to obtain the parameter representation of the abnormal motion trajectory.

[0162] When performing anomaly assessment sampling on the abnormal motion trajectory based on the parameter representation, sampling points on the trajectory are selected from the parameter representation at fixed intervals. For each sampling point, the degree of abnormality of the point is determined by combining the deviation of the target object's posture and behavior pattern baseline. These values ​​are arranged in the sampling order to obtain the abnormal intensity sequence of the abnormal motion trajectory.

[0163] When performing local motion feature analysis on the parameter representation, for each point in the parameter representation, the positional changes of its adjacent points are analyzed to determine the curvature of the abnormal motion trajectory at that point, and the curvature of all points is arranged in order to form a local curvature sequence of the parameter representation; at the same time, the change in the trajectory position at that point over time is analyzed, and the changes of all points are arranged in order to form a velocity change rate sequence of the parameter representation.

[0164] When performing a composite weighted integral on the abnormal motion trajectory based on the abnormal intensity sequence, the local curvature sequence, and the velocity change rate sequence, corresponding weights are assigned to the abnormal intensity sequence, the local curvature sequence, and the velocity change rate sequence, respectively. The values ​​of the three sequences corresponding to each position point on the abnormal motion trajectory are combined according to their weights, and these combined values ​​are summarized along the entire length of the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory. The formula for calculating the cumulative deviation is as follows:

[0165] ;

[0166] In the formula, The cumulative deviation, The total arc length of the abnormal motion trajectory. For the abnormal intensity sequence at the arc length position The value at that location, It is an exponential function. These are preset positive real number parameters. For the local curvature sequence at the arc length position The value at that location, These are preset positive real number parameters. For the velocity change rate sequence at the arc length position The value at that location, It is an absolute value function.

[0167] When performing a structured semantic description of the body image sequence based on the cumulative deviation, the degree of abnormal movement, the duration of the abnormal movement, and the corresponding target object's body posture change characteristics are described in combination with the specific circumstances of the cumulative deviation. These descriptions are then organized into a fixed structure to obtain the evolution index of the body image sequence.

[0168] The length of the abnormal motion trajectory from the starting point to the ending point is measured, and this length is the total arc length of the abnormal motion trajectory.

[0169] Retrieve the obtained abnormal intensity sequence, and select the value in the sequence corresponding to the arc length position. This value is the value of the abnormal intensity sequence at the arc length position.

[0170] Based on the target subject's chronic disease type and combined with experience in processing body shape data of similar subjects in the past, a fixed value is determined, which is the preset positive real number parameter.

[0171] Retrieve the obtained local curvature sequence, and select the value in the sequence corresponding to the arc length position. This value is the value of the local curvature sequence at the arc length position.

[0172] Based on the target individual's chronic disease type and referring to the standard settings for processing historical physical characteristics, a fixed value is determined, which is another preset positive real number parameter.

[0173] Retrieve the obtained velocity change rate sequence, and select the value in the sequence corresponding to the arc length position. This value is the value of the velocity change rate sequence at the arc length position.

[0174] This content is used to calculate the cumulative deviation, which reflects the overall degree of abnormality of the abnormal motion trajectory. It combines the values ​​of the abnormal intensity sequence at the arc length position, the values ​​of the local curvature sequence at the arc length position, and the values ​​of the velocity change rate sequence at the arc length position to comprehensively measure the degree of deviation of the abnormal motion trajectory over the entire length. The result is used to form the evolution index of the body image sequence.

[0175] As the value of the abnormal intensity sequence at the arc length position increases, the corresponding correlation value will increase. After summing these values ​​along the total arc length of the abnormal motion trajectory, the cumulative deviation will increase accordingly.

[0176] When the value of the local curvature sequence at the arc length position increases, the corresponding exponential part value will decrease, the associated product part value will decrease, and after summing these values ​​along the total arc length of the abnormal motion trajectory, the cumulative deviation will decrease accordingly.

[0177] When the value of the velocity change rate sequence at the arc length position increases, the corresponding exponential part value will decrease, the associated product part value will decrease, and after summing these values ​​along the total arc length of the abnormal motion trajectory, the cumulative deviation will decrease accordingly.

[0178] The beneficial effects are as follows: By spatiotemporally matching the behavioral pattern baseline with the body posture image sequence to obtain the baseline matching distribution, motion trajectory deviation analysis is carried out based on the distribution to obtain spatiotemporal residuals. After multi-scale regional aggregation, a salient region map is formed. Abnormal motion trajectories are then screened based on this map. Subsequently, manifold coordinate mapping, abnormal assessment sampling, local motion feature analysis, and composite weight integration are performed on the abnormal motion trajectories to obtain the cumulative deviation, thereby forming a structured evolution index. This enables the accurate quantification and continuous monitoring of body posture evolution indicators. At the same time, relying on the feature extraction technology of personalized behavioral pattern baseline construction and spatiotemporal adversarial decoupling, subtle body posture changes related to chronic diseases can be accurately captured, improving the accuracy and completeness of body posture feature description. By using topology mining technology to construct the pattern association graph of evolution indicators, the intrinsic dependence between body posture indicators is clearly revealed. Combined with the patient's historical feedback records, personalized deduction of nursing strategies can be realized, improving the adaptability of nursing strategies to the patient's individual pathological characteristics and body posture response patterns, and enhancing the pertinence and effectiveness of nursing intervention.

[0179] The acquisition path of each content related to cumulative deviation was clarified. By combining the abnormal intensity, local curvature and velocity change rate at each position of the abnormal motion trajectory, the overall abnormality of the trajectory was comprehensively measured, providing a reliable basis for the evolution index of the body image sequence. At the same time, it can accurately reflect the influence of different factors on the cumulative deviation, making the calculation of cumulative deviation more in line with the actual characteristics of the abnormal motion trajectory, and ensuring that the evolution index obtained based on the cumulative deviation can accurately characterize the body changes of the target object.

[0180] The topology mining module 15 is used to perform dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index.

[0181] In this embodiment of the invention, when the topology mining module performs dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index, it is specifically used for:

[0182] The feature description is expanded in feature space to obtain a high-dimensional semantic manifold of the feature description;

[0183] Based on the high-dimensional semantic manifold, the evolution index is projected onto the manifold path to obtain the spatiotemporal path of the evolution index.

[0184] Morphological analysis of the spatiotemporal path yields a heterogeneous distribution of curvature.

[0185] Based on the aforementioned curvature heterogeneous distribution, critical surface identification is performed on the high-dimensional semantic manifold to obtain the folded topology of the high-dimensional semantic manifold.

[0186] Homotopy path compression is performed on the folded topology to obtain a simplified homotopy skeleton of the folded topology.

[0187] Based on the simplified homotopy skeleton, the spatiotemporal path is topologically integrated to generate the pattern association graph of the evolution index.

[0188] When the topology mining module performs homotopy path compression on the folded topology to obtain a simplified homotopy skeleton of the folded topology, it is specifically used for:

[0189] The redundancy of the folded topology is evaluated to obtain the redundancy of the folded topology.

[0190] Based on the redundancy, a visual information flow simulation is performed on the folded topology to obtain the information flow intensity map of the folded topology.

[0191] Based on the information flow intensity map, density peak aggregation is performed on the folded topology to obtain the node clusters of the folded topology.

[0192] Based on the node clusters, the shortest path connection is performed on the folded topology to obtain the simplified homotopy skeleton of the folded topology.

[0193] When performing feature space expansion on the feature description, first sort out all the feature terms included in the feature description, treat each feature term as an independent dimension, map the information content corresponding to each feature in the feature description to the specific position under each dimension, and then integrate the position information of all dimensions to form a continuous structure in a high-dimensional space, thus obtaining the high-dimensional semantic manifold of the feature description.

[0194] When performing manifold path projection on the evolution index based on the high-dimensional semantic manifold, the position of each feature information corresponding to the evolution index is located in the high-dimensional semantic manifold, and these positions are connected sequentially according to the time order corresponding to the evolution index to form a continuous path, thereby obtaining the spatiotemporal path of the evolution index.

[0195] When performing morphological analysis on the spatiotemporal path, the change in the connection direction between each point and its adjacent points on the spatiotemporal path is examined one by one to determine the curvature of the path at each point. The curvature of all points is then distributed and organized according to their positions on the spatiotemporal path to obtain the curvature heterogeneous distribution of the spatiotemporal path.

[0196] When performing critical surface identification on the high-dimensional semantic manifold based on the curvature heterogeneous distribution, the surface parts corresponding to the regions where the curvature changes in the curvature heterogeneous distribution are located in the high-dimensional semantic manifold, and these surface parts are integrated into the overall structure to obtain the folded topology of the high-dimensional semantic manifold.

[0197] When performing structural redundancy assessment on the folded topology, each part of the folded topology is examined one by one, the recurring structural content is identified, the proportion of these recurring contents in the entire structure is calculated, and the assessment result corresponding to this proportion is determined as the redundancy of the folded topology.

[0198] When performing visual information flow simulation on the folded topology based on the redundancy, the non-repetitive key areas for information transmission in the folded topology are identified according to the redundancy. The flow process of information between these key areas is simulated, and the strength of information flow in each area is marked. These marks are then integrated and presented to obtain the information flow intensity map of the folded topology.

[0199] When performing density peak aggregation on the folded topology based on the information flow intensity map, the region with higher information flow intensity is identified as the core point in the information flow intensity map. The regions with related information flow intensity around the core point are aggregated to the vicinity of the corresponding core point. Each core point and its corresponding aggregation region together form the node cluster of the folded topology.

[0200] When performing shortest path connections on the folded topology based on the node clusters, the position of the core point of each node cluster is determined, the shortest connection path between different core point positions is calculated, and these shortest paths are connected sequentially to form a concise continuous structure, thus obtaining the simplified homotopy skeleton of the folded topology.

[0201] When performing topological integration of the spatiotemporal paths based on the simplified homotopy skeleton, the spatiotemporal paths are mapped to the corresponding structural positions of the simplified homotopy skeleton, the relationships between different spatiotemporal paths are marked, and these relationships are integrated with the structure of the simplified homotopy skeleton to generate a pattern association diagram of the evolution index.

[0202] The beneficial effects are as follows: by expanding the feature space of the feature description to obtain a high-dimensional semantic manifold, and then obtaining the spatiotemporal path based on the evolution index projected on the manifold, the curvature heterogeneous distribution is obtained through morphological analysis and the fold topology is identified. Subsequently, the structure is subjected to redundancy assessment, information flow simulation, density peak convergence and shortest path connection to obtain a simplified homotopy skeleton. Finally, based on the skeleton, the spatiotemporal path is integrated to generate a pattern association graph of the evolution index. This can accurately sort out the relationship between feature description and evolution index, eliminate structural redundancy and present it in a simplified topological form, and clearly reveal the intrinsic relationship between different evolution indexes. This provides an intuitive and reliable association basis for subsequent deduction of personalized nursing strategies based on historical feedback records, and helps to improve the adaptability and pertinence of nursing strategies.

[0203] The strategy generation module 16 is used to perform effect mapping and deduction on the evolution indicators based on the historical feedback records of the target object and the pattern association diagram to obtain the nursing strategy for the target object.

[0204] In this embodiment of the invention, when the strategy generation module performs effect mapping deduction on the evolution indicators based on the historical feedback records of the target object and the pattern association graph to obtain the nursing strategy for the target object, it is specifically used for:

[0205] Based on the historical feedback records of the target object, knowledge transfer is performed on the pattern association graph to obtain an experience-weighted causal graph of the pattern association graph.

[0206] Based on the aforementioned experience-weighted causal graph, a virtual intervention response simulation is performed on the evolution indicators to obtain the intervention prediction distribution of the evolution indicators;

[0207] The intervention prediction distribution is subjected to feasibility filtering, and the filtered distribution is prioritized to obtain the intervention priority sequence of the intervention prediction distribution;

[0208] Based on the intervention priority sequence, the intervention prediction distribution is integrated with personalized context to obtain the nursing strategy for the target object.

[0209] When performing knowledge transfer on the pattern association diagram based on the historical feedback records of the target object, the following steps are taken: First, the implementation effects of past nursing measures, the target object's tolerance to different intervention methods, and changes in body posture indicators contained in the historical feedback records are sorted out. The experiential information related to the evolution indicators is extracted, and this experiential information is mapped to each relationship in the pattern association diagram. Each relationship is assigned a weight that reflects the effectiveness of the experience, and the causal strength of the relationship between different evolution indicators is clarified to form the experience-weighted causal diagram of the pattern association diagram.

[0210] When performing virtual intervention response simulations on the evolution indicators based on the empirical weighted causal graph, for each evolution indicator, different types of nursing interventions are simulated to act on that indicator. Based on the weights and causal directions of the relationships in the empirical weighted causal graph, the changing trend of the evolution indicator under each intervention is derived. At the same time, the chain effect of the change of the indicator on other related evolution indicators is analyzed. The prediction results of the changes in the evolution indicators corresponding to all interventions are classified and organized according to the intervention type and the changing trend to obtain the intervention prediction distribution of the evolution indicators.

[0211] When performing feasibility filtering on the intervention prediction distribution, the feasibility of each intervention measure in the intervention prediction distribution is checked one by one, taking into account the target object's physical condition, the implementation conditions of home care, the operational complexity of the care measures, and other practical factors. Intervention prediction results that do not meet the actual conditions, may cause physical burden to the target object, or cannot be implemented in a home setting are eliminated. Then, the filtered prediction results are arranged in order according to the expected improvement of evolution indicators by the intervention measures, the resource input required for implementation, and the target object's past tolerance, to obtain the intervention priority sequence of the intervention prediction distribution.

[0212] When performing personalized contextual integration of the intervention prediction distribution based on the intervention priority sequence, personalized contextual information such as the target object's daily routine, dietary preferences, and family care situation is collected. This information is matched with the various intervention measures in the intervention priority sequence, and the implementation time and specific operational details of the intervention measures are adjusted to suit the target object's daily schedule. Implementation precautions for personalized scenarios are added, and the adjusted intervention measures are integrated into a complete plan according to priority order to obtain the nursing strategy for the target object.

[0213] The beneficial effects are as follows: by transferring the historical feedback records of the target subjects to a pattern association graph to form an experience-weighted causal graph, and relying on this graph to perform virtual intervention response simulation on the evolution indicators to obtain the intervention prediction distribution, the intervention priority sequence is obtained through feasibility filtering and priority arrangement, and then integrated with personalized contextual information to form a nursing strategy. This can fully integrate past nursing experience and the correlation logic of evolution indicators to ensure the rationality and pertinence of intervention prediction, screen out effective intervention measures that are suitable for home scenarios and the physical condition of the target subjects, and at the same time, make the nursing strategy fit the actual life of the target subjects through personalized adaptation and adjustment, improve the feasibility and acceptance of nursing measures, and help to achieve precise intervention and effect optimization of home care for chronic diseases.

[0214] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0215] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0216] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A home care system for chronic diseases based on body posture recognition, characterized in that, The system includes a behavior baseline construction module, an activity segmentation module, a feature decoupling module, an indicator tracking module, a topology mining module, and a policy generation module, wherein: The behavior baseline construction module is used to perform pattern induction analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object. The activity segmentation module is used to segment the real-time image sequence of the target object into activity segments to obtain the body image sequence of the target object; The feature decoupling module is used to perform spatiotemporal adversarial decoupling on the body posture image sequence to obtain the feature description of the body posture image sequence; The indicator tracking module is used to track the cumulative deviation of the body posture image sequence based on the behavioral pattern baseline, and obtain the evolution indicator of the body posture image sequence. The topology mining module is used to perform dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index. The strategy generation module is used to perform effect mapping and deduction on the evolution indicators based on the historical feedback records of the target object and the pattern association diagram to obtain the nursing strategy for the target object.

2. The chronic disease home care system based on body posture recognition as described in claim 1, characterized in that, When the behavior baseline construction module performs pattern inductive analysis on the historical images of the target object to obtain the behavior pattern baseline of the target object, it is specifically used for: The pose change sequence of the target object is obtained by tracing the pose change of the target object's historical images. Based on the baseline pose change sequence, motion continuity analysis is performed on the historical image to obtain a motion coherence measure of the historical image; By performing time-series pattern mining on the historical images, a description of the activity patterns of the target object can be obtained; The baseline behavior pattern of the target object is obtained by fusing the baseline posture change sequence, the motion coherence measure, and the activity pattern description.

3. The chronic disease home care system based on body posture recognition as described in claim 1, characterized in that, When the activity segmentation module performs activity segmentation on the real-time image sequence of the target object to obtain the body image sequence of the target object, it is specifically used for: Temporal structure recognition is performed on the real-time image sequence of the target object to obtain the state change nodes of the real-time image sequence; Based on the state change nodes, motion state discrimination is performed on the real-time image sequence to obtain the active segments of the real-time image sequence; The connection strength of the activity segments is evaluated to obtain a coherence score for the activity segments; Based on the coherence score, the activity segments are optimized and merged to obtain the body image sequence of the target object.

4. The chronic disease home care system based on body posture recognition as described in claim 1, characterized in that, When the feature decoupling module performs spatiotemporal adversarial decoupling on the body posture image sequence to obtain the feature description of the body posture image sequence, it is specifically used for: Multi-granularity feature extraction is performed on the body posture image sequence to obtain the hierarchical features of the body posture image sequence; The hierarchical features are subjected to hierarchical learning to obtain the hierarchical baseline of the hierarchical features; Based on the hierarchical baseline, the hierarchical features are bidirectionally competitively aligned to obtain the alignment baseline representation of the hierarchical features; Based on the alignment baseline representation, attention decomposition is performed on the body posture image sequence to obtain the weight distribution of the body posture image sequence; Based on the weight distribution, the hierarchical features are weighted and reorganized to obtain the feature components of the hierarchical features; The feature components are structured and encoded to obtain the feature description of the body image sequence.

5. A home care system for chronic diseases based on body posture recognition as described in claim 4, characterized in that, When the feature decoupling module performs bidirectional competitive alignment of the hierarchical features based on the hierarchical baseline to obtain the aligned baseline representation of the hierarchical features, it is specifically used for: The interaction intensity structure of the hierarchical baselines is performed to obtain the initial competition coefficient matrix of the hierarchical baselines; Based on the initial competition coefficient matrix, selective baseline responses are performed on the hierarchical features to obtain the response distribution of the hierarchical features; The temporal consistency of the response distribution is evaluated to obtain a confidence score for the response distribution; Based on the confidence score, the initial competition coefficient matrix is ​​adaptively optimized for competition intensity. Based on the optimized competition coefficient matrix, the hierarchical features are reweighted and aggregated to obtain the aligned baseline representation of the hierarchical features.

6. A home care system for chronic diseases based on body posture recognition as described in claim 1, characterized in that, When the indicator tracking module performs cumulative deviation tracking on the body posture image sequence based on the behavioral pattern baseline to obtain the evolution indicator of the body posture image sequence, it is specifically used for: The baseline of the behavioral pattern is spatiotemporally matched with the body posture image sequence to obtain the baseline matching distribution of the body posture image sequence; Based on the baseline matching distribution, motion trajectory deviation analysis is performed on the body posture image sequence to obtain the spatiotemporal residuals of the body posture image sequence; Multi-scale region aggregation is performed on the spatiotemporal residuals to obtain a saliency region map of the spatiotemporal residuals; Based on the salient region map, the temporal coherence of the body image sequence is identified and filtered to obtain the abnormal motion trajectory of the body image sequence; The cumulative deviation of the abnormal motion trajectory is obtained by performing pattern intensity integration on the abnormal motion trajectory. Based on the accumulated bias, a structured semantic description is performed on the body posture image sequence to obtain the evolution index of the body posture image sequence.

7. A home care system for chronic diseases based on body posture recognition as described in claim 6, characterized in that, When the index tracking module performs pattern intensity integration on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory, it is specifically used for: The abnormal motion trajectory is mapped to manifold coordinates to obtain a parameter representation of the abnormal motion trajectory; Based on the parameter representation, anomaly evaluation sampling is performed on the abnormal motion trajectory to obtain the abnormal intensity sequence of the abnormal motion trajectory; Local motion feature analysis is performed on the parameter representation to obtain the local curvature sequence and velocity change rate sequence of the parameter representation; Based on the abnormal intensity sequence, the local curvature sequence, and the velocity change rate sequence, a composite weighted integral is performed on the abnormal motion trajectory to obtain the cumulative deviation of the abnormal motion trajectory.

8. A home care system for chronic diseases based on body posture recognition as described in claim 1, characterized in that, When the topology mining module performs dependency topology mining on the evolution index based on the feature description to obtain the pattern association graph of the evolution index, it is specifically used for: The feature description is expanded in feature space to obtain a high-dimensional semantic manifold of the feature description; Based on the high-dimensional semantic manifold, the evolution index is projected onto the manifold path to obtain the spatiotemporal path of the evolution index. Morphological analysis of the spatiotemporal path yields a heterogeneous distribution of curvature. Based on the aforementioned curvature heterogeneous distribution, critical surface identification is performed on the high-dimensional semantic manifold to obtain the folded topology of the high-dimensional semantic manifold. Homotopy path compression is performed on the folded topology to obtain a simplified homotopy skeleton of the folded topology. Based on the simplified homotopy skeleton, the spatiotemporal path is topologically integrated to generate the pattern association graph of the evolution index.

9. A home care system for chronic diseases based on body posture recognition as described in claim 8, characterized in that, When the topology mining module performs homotopy path compression on the folded topology to obtain a simplified homotopy skeleton of the folded topology, it is specifically used for: The redundancy of the folded topology is evaluated to obtain the redundancy of the folded topology. Based on the redundancy, a visual information flow simulation is performed on the folded topology to obtain the information flow intensity map of the folded topology. Based on the information flow intensity map, density peak aggregation is performed on the folded topology to obtain the node clusters of the folded topology. Based on the node clusters, the shortest path connection is performed on the folded topology to obtain the simplified homotopy skeleton of the folded topology.

10. A home care system for chronic diseases based on body posture recognition as described in claim 1, characterized in that, When the strategy generation module performs effect mapping and deduction on the evolution indicators based on the historical feedback records and the pattern association diagram of the target object to obtain the nursing strategy for the target object, it is specifically used for: Based on the historical feedback records of the target object, knowledge transfer is performed on the pattern association graph to obtain an experience-weighted causal graph of the pattern association graph. Based on the aforementioned experience-weighted causal graph, a virtual intervention response simulation is performed on the evolution indicators to obtain the intervention prediction distribution of the evolution indicators; The intervention prediction distribution is subjected to feasibility filtering, and the filtered distribution is prioritized to obtain the intervention priority sequence of the intervention prediction distribution; Based on the intervention priority sequence, the intervention prediction distribution is integrated with personalized context to obtain the nursing strategy for the target object.

Images