Fall risk classification method based on daily walking foot pressure data, medium and equipment

Abstract

The invention discloses a fall risk classification method based on daily walking foot pressure data, a medium and equipment, and belongs to the technical field of data classification prediction. The method comprises the following steps: acquiring foot pressure data of a target in a daily walking process; performing preprocessing and feature extraction on the foot pressure data to obtain foot pressure feature data; the foot pressure feature data serve as samples to be classified, a tumble risk classification result is obtained through a trained fuzzy classifier, and for the foot pressure feature data, the tumble risk classification result is obtained through the fuzzy classifier by calculating the membership degree between the samples. The method is low in complexity, high in execution efficiency, small in required computing power, capable of responding and processing in real time at an edge terminal, and wider in application.

Description

Technical Field

[0001] The present invention relates to the technical field of data classification and prediction in medical health monitoring, and more specifically, to a fall risk classification method, medium and device based on daily walking foot pressure data. Background Art

[0002] Falls are a common symptom of aging. After a fall, some elderly individuals become restricted or unable to move, and require additional medical care expenses. Accurately assessing the fall risk level of older adults can help them understand their own risks and clarify intervention strategies for caregivers. This allows them to proactively implement fall prevention measures, thus preventing injuries before they occur.

[0003] Currently, fall risk assessment methods primarily include those based on clinical assessment tools, video-based approaches, and wearable systems. Clinical assessment tools primarily assess fall risk using standardized scales administered by healthcare professionals (such as the Morse Fall Assessment Scale and the Berg Balance Scale) and systematic functional tests (such as the Timed Up and Go Test (TUG) and the Five-Stance Sit-to-Stand Test). These tools combine multiple factors, such as fall history, mobility, and medication effects, to create a quantitative score. These tools are suitable for use in settings such as hospitals and nursing homes, but rely on manual experience and lack dynamic monitoring capabilities, making them difficult to cover real-time risks in the home. In video-based fall risk assessment, video analysis systems leverage computer vision technology to capture motion sequences and identify abnormal postures (such as sudden falls). Audio data (such as impact sounds) is used to enhance detection robustness. However, their application is limited by privacy concerns (e.g., low acceptance in bedrooms and bathrooms), the high cost of deploying multiple cameras, and false positives caused by lighting and occlusion. In the fall risk assessment solution based on wearable systems, wearable devices use sensors such as foot pressure, electromyography, and inertia to collect gait and balance data, and use machine learning algorithms to achieve real-time prediction of fall risks and fall event detection. It is highly portable and suitable for home and outdoor scenarios, but faces problems such as poor robustness and low accuracy.

[0004] In the prior art, patent application CN110367991A discloses a method for assessing the fall risk of the elderly, which includes four steps: data preparation, model construction, parameter range estimation, and fall risk assessment. Although this solution can assess the fall risk level of the elderly, it requires the collection of additional physical data, which increases the cost of fall risk assessment. Patent application CN119235299A discloses a real-time fall risk assessment system for the elderly driven by data from smart wearable devices. The system achieves the assessment of the fall risk level by integrating multiple modal sensor signals. However, the system is complex and requires many types of sensors, making it inconvenient for use outside the hospital. In addition, the system uses a linear support vector machine model, which "hard-divides" the sample space with a hyperplane / decision boundary, and is prone to misjudgment of motion data samples in inter-class overlap and fuzzy transition zones.

[0005] After analysis, the existing technology mainly has the following defects: 1) Assessment methods based on clinical tools rely on manual operation and subjective judgment, and are limited in that the assessment results are affected by the experience level of medical staff. Static scales and tests cannot monitor the risk of falls in dynamic activities (such as walking at home) in real time. They also lack the ability to monitor sudden balance disorders and are difficult to cover complex daily environments.

[0006] 2) Video-based fall risk assessment solutions require the assistance of video capture equipment and are only effective within specific areas covered by the equipment. Furthermore, image processing methods only have a high recognition rate for typical human imbalances, leaving blind spots. Furthermore, this approach requires a large amount of equipment, resulting in bulky hardware, cumbersome installation and use, and high costs, making it inconvenient for use in common life scenarios.

[0007] 3) Existing technologies typically implement fall risk assessment by continuously increasing the number and variety of wearable sensors. While these approaches can capture more comprehensive human motion signals through more sensors, they focus solely on data modality, resulting in poor classification model robustness and failure to address sample data with inter-class overlap, non-convex structures, and fuzzy transition regions. Furthermore, in practical applications, this significantly increases the complexity and cost of fall risk assessment, making it difficult to use outside of hospitals or at home.

[0008] 4) Existing technologies typically focus on optimizing the "quality" of sensor signal features when classifying fall risk. For example, they propose novel combinations of features based on sensor signals from different modalities in the hope of improving assessment results. These solutions fail to address fundamental issues from the perspective of classification strategy or logic, and fail to address sample data with overlapping classes, non-convex structures, and fuzzy transition regions.

[0009] 5) Threshold-based fall risk assessment schemes have relatively fixed thresholds, so they have poor robustness or generalization for unfamiliar samples and are difficult to use in actual scenarios. Summary of the Invention

[0010] The purpose of the present invention is to overcome the above-mentioned defects of the prior art and provide a fall risk classification method, medium and device based on daily walking foot pressure data.

[0011] According to a first aspect of the present invention, a method for classifying fall risk based on daily walking foot pressure data is provided. The method comprises the following steps: Collect foot pressure data during the target's daily walking; Preprocessing and feature extraction are performed on the foot pressure data to obtain foot pressure feature data; The foot pressure feature data is used as a sample to be classified, and a trained fuzzy classifier is used to obtain a fall risk classification result. The fuzzy classifier obtains the fall risk classification result by calculating the membership degree between samples for the foot pressure feature data.

[0012] According to a second aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the fall risk classification method based on daily walking foot pressure data of the first aspect are implemented.

[0013] According to a third aspect of the present invention, there is provided a computer device comprising a memory and a processor, wherein a computer program capable of being run on the processor is stored on the memory, wherein when the processor executes the computer program, the steps of the fall risk classification method based on daily walking foot pressure data of the first aspect are implemented.

[0014] Compared to existing technologies, the present invention offers the advantage of providing a fall risk classification scheme based on daily walking foot pressure data. This scheme requires no additional information about the subject (such as height, age, and health status) and does not require any specific actions. Fall risk assessment can be achieved solely with plantar pressure data collected during daily walking. Using this method, subjects can wear a data collection device without being constrained by time or space, and their fall risk level can be automatically and accurately classified with just a few steps. This solves the problem of low fall risk assessment accuracy caused by the complexity of the fall risk assessment process and the difficulty in correctly classifying sample data in the fuzzy transition zone between different fall risk categories and with non-convex distribution structures.

[0015] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0017] Figure 1 is a flowchart of a fall risk classification method based on daily walking foot pressure data according to one embodiment of the present invention; Figure 2 is a schematic diagram of segmenting continuous foot pressure data according to gait cycle and gait phase definitions according to one embodiment of the present invention; Figure 3 The membership function according to one embodiment of the present invention is ( =0.5) Schematic diagram of the changing process; Figure 4 The membership function according to one embodiment of the present invention is ( =2.0) Schematic diagram of the changing process; Figure 5 The membership function according to one embodiment of the present invention is ( =5.0) Schematic diagram of the changing process; Figure 6 is a schematic diagram of the degree of membership of categories according to one embodiment of the present invention; Figure 7 3 is a process diagram of a fall risk classification method based on daily walking foot pressure data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention.

[0019] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0020] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0021] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0022] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0023] Existing fall risk classification methods mainly rely on existing classification models, such as methods based on deep learning and machine learning. The accuracy of deep learning methods depends not only on the performance of the model's own architecture but also on a large amount of sample data. In addition, deep learning algorithms are unexplainable. In fall risk assessment, it is difficult to understand the algorithm's decision-making process from a medical perspective, and it is impossible to ensure the safety of diagnosis or treatment recommendations. In addition, the parameters of deep learning models are huge and the complexity is very high, making it difficult to deploy them in general embedded chips and form wearable smart assessment systems. In addition, deep learning models are sensitive to the influence of data distribution. If the test data differs greatly from the training data (such as individual differences in gait between different subjects), there may be a significant performance degradation. At present, machine learning models do not perform well when used for fall risk classification tasks. This is mainly because: 1) Lack of adaptability to fuzzy class boundaries. People with different fall risk levels exhibit a naturally fuzzy boundary phenomenon. For example, some subjects' motion characteristics are highly similar to those of fall risk level 1, while also being highly similar to those of risk level 2 (e.g., mild imbalance or moderate instability). This leads to a high degree of overlap between classes in the feature space, making it difficult to accurately classify these gray-zone samples, leading to misjudgments and affecting classification accuracy.

[0024] 2) It is not suitable for non-convex sample distribution structures. Motion feature distributions often exhibit nonlinear, non-convex structures. For example, some high-risk samples may be distributed in a "band" between low-risk areas. Traditional classifiers based on linear or convex assumptions, such as SVM (support vector machine) and LDA (linear discriminant analysis), have difficulty accurately fitting such complex boundaries.

[0025] 3) In real-world data, some regions may be highly concentrated with samples of a certain category, while samples of other categories are sparse or even absent, resulting in a phenomenon known as "regional exclusivity." Existing machine learning models often overfit these concentrated regions and perform unstable at the boundaries or in sparse regions, thus affecting classification performance.

[0026] To overcome the defects in the prior art, the present invention provides a fall risk classification method based on daily walking foot pressure data. Figure 1 As shown, the method includes the following steps: Step S110 , collecting foot pressure data during daily walking.

[0027] The device for collecting foot pressure data (ie, plantar pressure data) may be a smart insole system or other foot pressure collection system during walking.

[0028] For example, an intelligent foot pressure insole system is used to collect pressure signals of the human feet during daily walking. The intelligent foot pressure insole system consists of a pair of pressure insoles and a smartphone (installed with an app with pressure data display and storage functions). The pressure insoles and smartphone transmit data via Bluetooth. The pressure insoles are consistent with normal insoles in appearance, size and usage, and can be directly embedded in shoes. In one embodiment, a single pressure insole has 32 pressure sensing unit channels, and a pair of pressure insoles has a total of 64 pressure sensing unit channels. The 64-channel foot pressure data is collected at a frequency of 25Hz by the intelligent foot pressure insole system, and then displayed and saved on the smartphone app. In addition, the weight information of each subject can also be collected synchronously.

[0029] Step S120 , preprocessing and feature extraction are performed on the collected foot pressure data to obtain a foot pressure feature data set, and then a training set, a validation set, and a test set are constructed.

[0030] For example, the pressure data for each channel is filtered using a 4th-order low-pass Butterworth filter with a cutoff frequency of 15 Hz. After filtering, the pressure data for each channel is normalized by dividing it by the corresponding subject's weight. The continuous pressure data for the 32 left channels is then accumulated, and the sum is recorded as Sum_L_pre. Similarly, the continuous pressure data for the 32 right channels is accumulated, and the sum is recorded as Sum_R_pre. Figure 2This diagram illustrates the segmentation of continuous foot pressure data into complete gait cycles and gait phases. The horizontal axis (x-axis) represents the index of the number of sampling points, and the vertical axis represents the total pressure (e.g., in Pa). Curves showing the changes in Sum_L_pre and Sum_R_pre over time during daily walking are shown, with the red curve representing Sum_L_pre and the blue curve representing Sum_R_pre. The left-side pressure data between x+2 and t+1 represents the pressure data during the left single-stance phase. The left-side pressure data between x+1 and x+2 represents the pressure data during the left-side ascent of the double-stance phase. The left-side pressure data between t and x represents the pressure data during the left-side descent of the double-stance phase. The right-side pressure data between x and x+1 represents the pressure data during the right single-stance phase. The right-side pressure data between t and x represents the pressure data during the right-side ascent of the double-stance phase. The right-side pressure data between x+1 and x+2 represents the pressure data during the right-side descent of the double-stance phase. The continuous foot pressure data is divided into several segments according to a single complete gait cycle, with the current right foot contact time t and the next right foot contact time t+1 as the boundary. Each segment contains only the continuous plantar pressure data of the right side and the corresponding continuous plantar pressure data of the left side during a complete gait cycle. In addition, to more accurately capture gait changes, the continuous foot pressure data of the left and right sides in a single gait cycle can be further divided into several foot pressure data segments related to the left side (left single support phase data segment, left rising segment of double support phase, and left falling segment of double support phase) and several foot pressure data segments related to the right side (right single support phase data segment, right rising segment of double support phase, and right falling segment of double support phase) according to medically defined gait phase parameters (left single support phase, right single support phase, rising phase of double support phase, and falling phase of double support phase).

[0031] In this way, the continuous bilateral foot pressure data can be segmented into several data segments with a duration of a single gait cycle. Then, for each data segment with a duration of a single gait cycle, the pressure data during the left single support phase, the pressure data during the left rising period of the double support phase, the pressure data during the left descending period of the double support phase, the pressure data during the right single support phase, the pressure data during the right rising period of the double support phase, and the pressure data during the right descending period of the double support phase are further extracted. The purpose of segmenting the continuous foot pressure data and further extracting the gait phase pressure data segments is to enable the subsequent foot pressure feature extraction work to be carried out smoothly.

[0032] In one embodiment, four feature extraction windows are defined: feature extraction based on bilateral pressure data from a complete single gait cycle; feature extraction based on pressure data from both the left and right single-stance phases; feature extraction based on pressure data from both the left and right ascending phases of the double-stance phase; and feature extraction based on pressure data from both the right and left descending phases of the double-stance phase. Specifically, the same series of feature extraction operations are performed based on each of the four feature extraction windows (assuming the extracted features have a dimension of d). Assuming that n foot pressure data segments with complete gait cycles are segmented from the continuous daily gait data of several subjects, four feature extraction windows of different scales will form four datasets of size n*d, respectively. The labels of the samples in the datasets are defined as follows: if n_j foot pressure data segments with complete gait cycles are segmented from the daily gait data of subject j, and the current subject's fall risk level is p_j, then the labels of all n_j samples are p_j. Through this design, a foot pressure feature dataset can be obtained, and then a training set, a validation set, and a test set can be constructed for classifier training. For example, the training set reflects the correspondence between foot pressure feature data samples and fall risk levels.

[0033] Step S130 : training a fuzzy classifier using the training set and the validation set. The fuzzy classifier obtains a fall risk classification result by calculating the membership degree between samples for the foot pressure feature data.

[0034] Based on the obtained foot pressure feature dataset, the membership between samples and the membership between a sample and the entire category of samples can be further calculated. For example, a Gaussian-Mahalanobis distance fuzzy membership calculation method or other membership function calculation methods can be used.

[0035] 1) Unconstrained Gaussian-Mahalanobis distance fuzzy membership calculation method In order to better understand this method, the definition of prototype points is given. Prototype points refer to a set of sample points in each class that are used to represent the distribution boundary characteristics of the class. For example, prototype points can be obtained by K-means clustering. For the current class , prototype point The specific calculation process is expressed as: (1) in, is a category Samples in Does not belong to the category but belong to other categories; is the Euclidean norm, also called L2 distance; Is the current prototype point In the current Index in class sample; K is currently The total number of prototype points of the class, Representation category q After being split into K groups, the sample set of the i-th group is, Represents all samples that do not belong to this q class.

[0036] After obtaining the prototype point of each class, start calculating any sample point in the data set With class The degree of membership between The concept of membership is a core concept used to describe the relationship between elements and sets, and is used to quantify the degree to which an element belongs to a set. For example, membership The calculation process is expressed as: (2) (3) (4) in, is the sample point With class The degree of membership between Is to be calculated and category The sample points of membership degree. is a category The covariance matrix of all prototype points in , is a parameter that controls the steepness of the attenuation, T represents the transpose, is a category The prototype point in . Is the prototype point in the category The index in . K is currently The total number of prototype points of the class. Formulas (2) and (3) are The calculation function of , any one can be selected in practical application. Formula (4) is The membership function between the point and the prototype point.

[0037] It should be noted that the general form of the Gaussian function is , d Represents a sample x The distance measure (Euclidean distance) between the prototype point (such as the reference point such as the cluster center). The general formula of Mahalanobis distance is . Conventional membership functions are often global formulas, not the accumulation of several membership functions. Conventional membership functions include hypercube, hypersphere or regular polyhedron functions. If the distribution of the data cluster is of the type of hypercube, hypersphere or regular polyhedron at the beginning, the result of the conventional membership calculation is very good. However, if the distribution of the data cluster is not of the assumed hypercube, hypersphere or regular polyhedron type, for example, the data cluster distribution is horseshoe-shaped or banana-shaped, then the error in the membership calculation of sample x to the entire class will be very large. In view of the distribution characteristics of foot pressure data, the membership adopted by the present invention is The calculation formula is based on the sample x The membership of several prototype points is accumulated, so there is no need to preset the distribution of data clusters in advance. In other words, no matter how the data clusters are distributed, the Gaussian-Mahalanobis distance fuzzy membership function can fit the distribution of the clusters well.

[0038] It's important to note that constrained membership functions are like casting plastic with a fixed mold. Take a "spherical mold" or "cubic mold," pour plastic into it, and the resulting membership region can only be a sphere or a cube. This works well if the data clusters are truly spherical, but it's difficult to fit them if the clusters are horseshoe-shaped or banana-shaped. Unconstrained membership functions are like a jigsaw puzzle made of many small clay lumps. Place a bunch of small lumps (each representing a local membership function, such as a small Gaussian peak) at the cluster boundaries and freely piece them together. The resulting shape can be arbitrarily curved, forked, or have holes, all determined by the data.

[0039] In short, membership The calculation process no longer assumes that the cluster is spherical, square or any regular body, but is composed of multiple "local Gaussian peaks". The position and shape (covariance Σ) of each peak are derived from the data itself. Therefore, the membership degree is called The calculation process function is the unconstrained Gaussian-Mahalanobis distance fuzzy membership function.

[0040] It should be noted that is a category The covariance matrix of all prototype points in reflects the correlation between foot pressure features and the main direction of the cluster. It is a parameter that controls the attenuation steepness. The influence of parameters on membership, for the unconstrained Gauss-Mahalanobis membership function The process of change is presented in the form of a graph. Figure 3 、 Figure 4 and Figure 5 They presented =0.5, =2.0 and =5.0, MF1 and MF2 represent different The Gaussian-Mahalanobis membership function of the value, Figure 3 、 Figure 4 and Figure 5 The horizontal axis represents the sample point x， The vertical axis represents the sample points x Membership .

[0041] In order to further understand the concept of Gauss-Mahalanobis distance fuzzy membership, further explanation is given below. Assuming that the current foot pressure feature is a two-dimensional feature, see Figure 6 shown. Figure 6 (a) shows all samples of category q, where the x-axis and y-axis are the two-dimensional coordinates of the samples (marked as x 1 and x 2) The range of x-axis and y-axis is between [0,1], then the point in this 1×1 coordinate plane is denoted as point t. Assume that the category q All samples are prototype points (all samples can be directly regarded as prototype points or the boundary points can be regarded as prototype points by clustering. Compared with directly treating all samples as prototype points, the number of prototype points obtained by clustering is relatively small, which can significantly reduce the amount of membership calculation. Therefore, it is recommended to use clustering method to obtain prototype points. In this example, all sample points are regarded as prototype points). Then, the membership of t to category q is calculated by formula (2) and marked with colors in this 1×1 coordinate plane, as shown in the following example: Figure 6 As shown in (b), the darker the color, the higher the membership. It can be seen that where the original sample concentration is very high, the color is dark red. In order to further illustrate the image, t to category q The membership degree of is displayed in three-dimensional space, such as Figure 6 The three-dimensional coordinate form shown in (c) is shown in Figure 1, where the x-axis and y-axis are the plane coordinates of t, and the x 1 and x 2 means that the z-axis is the degree of membership Similarly, we can see that the farther away from the prototype point, the more t is related to the category q The lower the degree of membership, the higher the closer the relationship.

[0042] 2) Fall risk classification method based on Gaussian-Mahalanobis distance fuzzy membership The Gaussian-Mahalanobis distance fuzzy membership calculation method can be used to achieve fall risk classification prediction.

[0043] Assume that the sample to be classified is , and there are Class samples are kind, kind, kind,…, kind.

[0044] In each feature extraction window In the example, perform the following operations once: Step 11: Obtain the boundary prototype points in each class of samples through clustering method ,in, Indicates category, The range is between 1 and c. The current prototype point is The index within the category sample.

[0045] Step 12: Calculate the sample The degree of membership in each category ,in, Indicates category, The range is between 1 and c.

[0046] Step 13: It can be considered that in the feature extraction window The membership degree of each category is recorded as ,in, Indicates category, The range is between 1 and c. , k is the index of the feature extraction window, is the number of feature extraction windows (for example, m is 4).

[0047] After completing the above three steps in each feature extraction window, perform the following operations: Step 21: Multi-scale fusion. Extract samples from each feature scale window Pair The membership of the sample is linearly weighted, and finally, the sample In the class The final membership degree is: (5) in, Indicates category, The range is between 1 and c.

[0048] Step 22: Find the Sample The sample category with the largest membership , expressed as: (6) in, Indicates category, The range is between 1 and c.

[0049] Step 23: Get the final prediction result, that is, the sample The category is .

[0050] Step S140 : using the trained fuzzy classifier to obtain a fall risk classification result for the actually collected foot pressure data.

[0051] After the training and verification of the fuzzy classifier are completed, it can be used for actual fall risk classification prediction. For example, the application process of the classifier is: using the foot pressure data acquisition system to collect the target's foot pressure data during daily walking; preprocessing the foot pressure data; extracting features from the foot pressure data to obtain foot pressure feature data; normalizing the foot pressure feature data; based on the normalized data, using the trained fuzzy classifier to predict the fall risk classification results.

[0052] To further understand the training and practical application process of fuzzy classifiers, Figure 7 The schematic diagram of the Gaussian-Mahalanobis distance fuzzy membership fall risk classification method based on daily walking foot pressure data is shown. The overall process includes: collecting foot pressure data; preprocessing foot pressure data; extracting features from foot pressure data; normalizing the data to obtain a training set; determining the fuzzy estimate of each class for sample x in each feature window; and using the validation set to determine an optimal value for each feature extraction window. Parameters; linearly weight the membership of each class of sample x in different feature extraction windows; obtain a trained fuzzy classifier. Finally, the trained fuzzy classifier can be used to classify fall risk.

[0053] It should be noted that, without departing from the spirit and scope of the present invention, those skilled in the art may make appropriate changes or modifications to the above embodiments. For example, during the filtering process, the filtering algorithm may be replaced with another filtering algorithm. When calculating the prototype points of each sample class, the clustering algorithm may adopt K-means clustering, hierarchical clustering, density clustering, spectral clustering, etc. For example, in addition to using the points at the boundary of each sample class identified by clustering as prototype points, several sample points in the sample center obtained by clustering or other means may be used as prototype points, or all sample points in the current sample class may be used as prototype points. The fuzzy membership function may be an unconstrained fuzzy membership function based on the Gaussian-Mahalanobis distance, or it may be replaced with another membership function, but the classification effect may be reduced after the replacement. The current sample class membership is calculated by accumulating the membership of the prototype points in the current class or taking the maximum prototype point membership. Alternatively, other methods based on the membership of several prototype points may be used to solve the current sample class membership. For example, a feature extraction window with four scales can be used. A more detailed or coarse division can be implemented for a complete gait cycle to obtain a finer or coarser feature extraction window, thereby achieving multi-scale measurement of gait motion. In this embodiment of the present invention, the final membership of each sample class is obtained by linearly weighting the membership of the same type of sample classes at different scales (for example, with the weighting coefficient being 1). Other weighting coefficients or other weighting methods (such as exponential weighting) can also be used.

[0054] In order to further verify the effect of the present invention, an experimental verification was carried out, and the experimental process was evaluated using the Mini-BESTest scale. The Mini-BESTest scale is a classic balance ability and fall risk assessment tool commonly used in clinical practice. The higher the Mini-BESTest scale score, the better the subject's balance ability and the lower the risk of falling. Regarding the classification of fall risk levels, the clinical definition is as follows: subjects with a scale score of less than 19 points are classified as a high-risk group for falls; subjects with a scale score between 19-24 points are classified as a medium-risk group for falls; subjects with a scale score between 25-28 are classified as a low-risk group for falls. During the experimental verification process, based on the intelligent insole system, first, bilateral foot pressure data of 45 subjects over the age of 60 during daily straight walking were collected (the number of subjects with low / medium / high fall risk levels were 17 / 15 / 13, respectively). Subsequently, it was processed and classified based on the method of the present invention. The fall risk level classification results of the Gaussian-Mahalanobis distance fuzzy membership fall risk assessment method based on daily walking foot pressure data in the present invention and the classification method in the prior art are shown in Table 1 below.

[0055] Table 1 Comparison results between the present invention and the prior art

[0056] It should be noted that the classification results of the fall risk level classification algorithm presented in Table 1 are all the results under the optimal hyperparameters. The cross-individual leave-one-out method is a cross-validation method for scenarios where data differences between individuals are significant. Its core is to implement the leave-one-out verification process with "individual" as the basic unit. Specifically, in a data set containing N independent individuals, all the data of a certain individual is retained as the test set in each iteration, and all samples of the remaining N-1 individuals are integrated into the training set. The generalization ability of the model is evaluated by the prediction performance of the trained model on the retained individual data. After all individuals have completed a round of verification as test sets in turn, the performance indicators of each iteration (such as accuracy) are averaged, and this is used as the final performance evaluation result of the model in the cross-individual scenario.

[0057] In summary, the present invention filters the raw plantar pressure signal to remove interference and signal noise, ensuring subsequent calculations are accurate and unaffected by measurement noise. Furthermore, a Gaussian-Mahalanobis kernel is proposed to construct an unconstrained fuzzy membership function, which effectively models the relative proximity of samples to each class and supports one-to-many fuzzy membership representation. This structure is particularly suitable for classifying sample data in the fuzzy transition zone between different fall risk classes and with non-convex distribution structures, enhancing the robustness of recognition for atypical samples. Prototype points are no longer simply selected from class centers. Instead, representative samples from boundary-sensitive regions are preferentially extracted based on a clustering approach, improving the resolution of the fuzzy function at the discriminant boundary and effectively avoiding sample redundancy and generalized membership ambiguity. Furthermore, compared to traditional fuzzy functions (such as those based on hypercubes, hyperspheres, or regular polyhedra), the fuzzy membership function of the present invention does not require the distribution of pre-set sample clusters. It can fit any data sample distribution and adaptively characterize the covariance structure of sample distributions of different classes. It has stronger boundary direction perception capabilities and can accurately identify the actual boundary shape of non-spherical and non-convex classes. This invention integrates foot pressure statistical features extracted at multiple time scales to form a multi-scale feature nested membership evaluation mechanism. This can simultaneously perceive both short-term center of gravity shifts and long-term gait trends ("short-term" corresponds to pressure data during the gait phase, and "long-term" corresponds to data from the entire gait cycle). Based on performance at different time granularities, it can provide comprehensive fall risk assessment results. In short, this invention has low complexity, high execution efficiency, low computing power requirements, and can respond and process in real time at the edge terminal, making it more widely applicable.

[0058] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0059] A computer-readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure within a groove on which instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse passing through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0060] Various aspects of the present invention are described herein with reference to flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0061] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device is generated that implements the functions / actions specified in one or more blocks in the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium, where these instructions cause the computer, programmable data processing device, and / or other device to operate in a specific manner. Thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing various aspects of the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0062] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device, so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more blocks in the flowchart and / or block diagram.

[0063] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of an instruction, and the module, program segment or part of the instruction contains one or more executable instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0064] While various embodiments of the present invention have been described above, the foregoing description is intended to be illustrative, non-exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or technological improvements in the marketplace, or to enable others skilled in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A fall risk classification method based on daily walking foot pressure data, characterized in that: The following steps are involved: Collect foot pressure data during the target's daily walking; Preprocessing and feature extraction are performed on the foot pressure data to obtain foot pressure feature data; The foot pressure feature data is used as a sample to be classified, and a trained fuzzy classifier is used to obtain a fall risk classification result. The fuzzy classifier obtains the fall risk classification result by calculating the membership degree between samples for the foot pressure feature data.

2. The method according to claim 1, characterized in that During the fuzzy classifier training process, the membership between samples is calculated by the following steps: Get the prototype points in each type of sample, expressed as: in, is a category Samples in Does not belong to the category but belong to other categories; is the Euclidean norm, It is the prototype point In the current The index in the class sample, , K is the current The total number of prototype points of the class, Representation category q After being split into K groups, the sample set of group i is, Indicates that it does not belong to q The sample set of the class; Calculate any sample point in the training set With class The degree of membership between .

3. The method according to claim 2, characterized in that The sample points With class The degree of membership between Calculated according to the following formula: in, is the sample point With class The degree of membership between is a category The covariance matrix of all prototype points in , is a parameter that controls the attenuation steepness, is a category The prototype point in Is the prototype point in the category The index in K Is the current class The total number of prototype points.

4. The method according to claim 2, characterized in that The sample points With class The degree of membership between Calculated according to the following formula: in ， is the sample point With class The degree of membership between is a category The covariance matrix of all prototype points in , is a parameter that controls the attenuation steepness, is a class The prototype point in The prototype point is in the class The index in K Is the current class The total number of prototype points.

5. The method according to claim 1, wherein The fall risk classification results are obtained based on the following formula: in: Among them, the sample points The category is , is the sample point In the class The membership degree of Indicates category, is the category index, k is the index of the feature extraction window, is the number of feature extraction windows.

6. The method according to claim 2, characterized in that The prototype points are points at the boundaries of each sample class identified based on a clustering method, or multiple sample points in the center of the sample are used as prototype points, or all sample points in the current sample class are used as prototype points.

7. The method according to claim 5, characterized in that The number of feature extraction windows is set to 4, which are: feature extraction based on bilateral pressure data of a complete single gait cycle; feature extraction based on pressure data during the left single support phase and pressure data during the right single support phase; Feature extraction is performed based on pressure data during the left rising period of the double support phase and based on pressure data during the right rising period of the double support phase; Feature extraction is performed based on the pressure data during the right descent of the double support phase and the pressure data during the left descent of the double support phase.

8. The method according to claim 6, characterized in that The clustering method is K-means clustering, hierarchical clustering, density clustering or spectral clustering.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of being run on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images