Fall risk classification method, medium, and device based on daily ambulatory foot pressure data

Abstract

The application discloses a fall risk classification method, medium and equipment based on daily walking foot pressure data, and belongs to the technical field of data classification and prediction. The method comprises the following steps: collecting foot pressure data in a target daily walking process; pre-processing and extracting features of the foot pressure data to obtain foot pressure feature data; taking the foot pressure feature data as a sample to be classified, and using a trained fuzzy classifier to obtain a fall risk classification result; and calculating the membership degree between samples to obtain the fall risk classification result for the foot pressure feature data. The application has low complexity, high execution efficiency, small required computing power, can be processed in real time on an edge terminal, and is more widely applied.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data classification prediction in medical health monitoring, and more particularly, to a fall risk classification method based on daily walking foot pressure data, a medium and an apparatus. BACKGROUND

[0002] Fall is a common syndrome of the elderly. After falling, some old people are limited in activity or cannot move, and need additional medical care costs. Accurate assessment of the fall risk level of the elderly can make the elderly understand their own risks and make caregivers clear about the direction of intervention, so as to layout fall prevention measures in advance, so as to curb the harm in the embryonic stage.

[0003] At present, the main methods for judging the fall risk level assessment include fall risk assessment schemes based on clinical assessment tools, fall risk assessment schemes based on video means and fall risk assessment schemes based on wearable systems. The clinical assessment tool mainly assesses the fall risk through the standardization scale (such as Morse fall assessment scale, Berg balance scale) and systematic function test (such as timed up and go test TUG, five times sit-to-stand test) dominated by medical staff, and quantitatively scores in combination with multiple factors such as fall history, mobility, drug influence, etc., which is suitable for hospital and nursing home scenes, but it relies on artificial experience and lacks dynamic monitoring ability, and it is difficult to cover real-time risks in home environment. In the fall risk assessment scheme based on video means, the video analysis system based on computer vision technology captures the action sequence through the camera and identifies the abnormal posture (such as sudden falling), and improves the detection robustness in combination with audio data (impact sound), but its application is limited by privacy disputes (such as low acceptance of bedroom or bathroom scenes), high deployment cost of multiple cameras and misjudgment caused by light / shading. In the fall risk assessment scheme based on wearable system, the wearable device uses foot pressure, electromyography and inertial sensors to collect gait and balance data, and realizes real-time prediction and fall event detection of fall risk through machine learning algorithm, which is portable and suitable for home and outdoor scenes, but it faces the problems of poor robustness and low accuracy.

[0004] In the prior art, patent application CN110367991A discloses an old person fall risk assessment method, which includes four steps of data preparation, model construction, parameter range estimation and fall risk assessment. Although this scheme can assess the fall risk level of the elderly, it needs to collect additional physical data, increasing the cost of fall risk assessment. Patent application CN119235299A discloses an old person fall risk real-time assessment system based on intelligent wearable device data driving. This system realizes the assessment of fall risk level by integrating multiple modal sensor signals, but the system device is complex and requires more types of sensors, which is not convenient for use outside the hospital. Moreover, the system uses a linear support vector machine model to make a "hard cut" of the sample space with a hyperplane / decision boundary, which is prone to misjudgment for motion data samples in the overlapping and fuzzy transition zone between classes.

[0005] Through analysis, the prior art mainly has the following defects:

[0006] 1) The assessment method based on clinical tools needs to rely on manual operation and subjective judgment, and has the limitation that the assessment result is affected by the experience level of medical staff. Moreover, static scales and tests cannot monitor fall risk in dynamic activities (such as walking at home) in real time, and lack the monitoring ability of sudden balance disorders, making it difficult to cover complex daily environments.

[0007] 2) The fall risk assessment scheme based on video means needs video acquisition equipment assistance and can only work in specific areas covered by video acquisition equipment. Moreover, through image processing, only typical human imbalance behaviors can have a high recognition rate, and there is a recognition blind area. In addition, this way requires more equipment, the hardware device is bulky, the installation and use are cumbersome, and the cost is high, making it inconvenient to use in common life scenes.

[0008] 3) The prior art usually increases the types and number of worn sensors to realize fall risk assessment. Although this scheme can capture more comprehensive human motion signals through more sensors, this way only starts from the data modal, the classification model performance is not robust, and the sample data in the overlapping, non-convex structure and fuzzy transition zone between classes is not solved. In addition, in actual application, it significantly increases the complexity and cost of fall risk assessment, making it difficult to use in out-of-hospital or home scenarios.

[0009] 4) The prior art usually focuses on optimizing the "quality" of the sensor signal features in the fall risk classification scheme, for example, some new combined features are proposed based on different modal sensor signals in order to improve the assessment effect. This scheme does not solve the fundamental problem from the perspective of classification strategy or logic, and does not solve the sample data in the overlapping, non-convex structure and fuzzy transition zone between classes.

[0010] 5) Threshold-based fall risk assessment scheme, due to the relatively fixed threshold, the robustness or generalization of the unknown sample is poor, and it is difficult to use in actual scene. SUMMARY

[0011] The purpose of the present application is to overcome the defects of the prior art described above, and to provide a fall risk classification method, medium and equipment based on daily walking foot pressure data.

[0012] According to a first aspect of the present application, a fall risk classification method based on daily walking foot pressure data is provided. The method comprises the following steps:

[0013] Collecting foot pressure data during the target daily walking process;

[0014] Preprocessing and feature extraction of the foot pressure data to obtain foot pressure feature data;

[0015] 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 calculates the membership degree between samples to obtain the fall risk classification result.

[0016] According to a second aspect of the present application, a computer readable storage medium is provided, which stores a computer program, wherein the computer program is executed by a processor to implement the steps of the fall risk classification method based on daily walking foot pressure data of the first aspect.

[0017] According to a third aspect of the present application, a computer device is provided, comprising a memory and a processor, and a computer program capable of running on the processor is stored on the memory, wherein the processor executes the computer program to implement the steps of the fall risk classification method based on daily walking foot pressure data of the first aspect.

[0018] Compared with the prior art, the present application has the advantages that a fall risk classification scheme based on daily walking foot pressure data is provided, which does not require any additional information about the subjects (such as height, age and health status, etc.) and does not require any additional specific action to be performed, and only the foot pressure information during the daily walking process can realize the fall risk assessment. With the present application, the subjects only need to walk a few steps after wearing the data collection device without time and space constraints, and the fall risk level of the subjects can be automatically and accurately classified, solving the problem of low fall risk assessment accuracy caused by the complexity of the fall risk assessment process and the difficulty in correctly classifying the sample data in the fuzzy transition zone between different fall risk classes and with non-convex structure distribution.

[0019] Other features of the present application, its nature and advantages will become apparent from the following detailed description of the exemplary embodiments of the application, taken in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings incorporated in and forming a part of the specification, illustrate embodiments of the application and, together with the description, serve to explain the principles of the application.

[0021] Figure 1 is a flow chart of a fall risk classification method based on daily walking foot pressure data according to an embodiment of the present application;

[0022] Figure 2 is a schematic diagram of segmentation of continuous foot pressure data according to gait cycle and gait phase definition according to an embodiment of the present application;

[0023] Figure 3 is a schematic diagram of the process of membership function change with ( = 0.5) according to an embodiment of the present application;

[0024] Figure 4 is a schematic diagram of the process of membership function change with ( = 2.0) according to an embodiment of the present application;

[0025] Figure 5 is a schematic diagram of the process of membership function change with ( = 5.0) according to an embodiment of the present application;

[0026] Figure 6 is a schematic diagram of the membership of the class according to an embodiment of the present application;

[0027] Figure 7 is a schematic diagram of the process of a fall risk classification method based on daily walking foot pressure data according to an embodiment of the present application. DETAILED DESCRIPTION

[0028] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of the components and steps set forth in the examples, numerical expressions, and numerical values are not limiting to the scope of the present application unless otherwise specifically stated.

[0029] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way limiting to the scope of the application or its applications or uses.

[0030] Techniques, methods, and apparatus known to those of ordinary skill in the relevant art(s) can not be discussed in detail herein. However, where appropriate, the described techniques, methods, and apparatus should be considered as being part of the specification.

[0031] In all of the compositions and methods shown and discussed herein, any particular value is to be construed merely as an example, and not a limitation. Other examples of the exemplary embodiments can therefore have different values.

[0032] It should be noted that like reference numerals and letters refer to like items throughout the attached drawings, and thus once an item is defined in one drawing, it is not necessary that it be further discussed in further detail in the subsequent drawings.

[0033] 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 itself, but also on a large amount of sample data, and deep learning algorithms are not interpretable. In the evaluation of fall risk, it is difficult to understand the decision-making process of the algorithm from a medical point of view, and it is difficult to ensure the safety of the diagnosis or treatment recommendations. In addition, the deep learning model has a large number of parameters and very high complexity, which is difficult to deploy in general embedded chips and form a wearable intelligent evaluation system. Moreover, the deep learning model is sensitive to data distribution. If the test data is significantly different from the training data (such as individual differences between different subjects' gaits), there may be a significant performance decline. Currently, machine learning models used for fall risk classification tasks do not perform well, which is mainly because:

[0034] 1) Lack of adaptability to class boundary ambiguity. There is a natural "fuzzy boundary" phenomenon among different fall risk populations. For example, some subjects' movement characteristics are highly similar to those of fall risk level 1 and highly similar to those of risk level 2 (such as slight imbalance or moderate instability), which leads to a high overlap between classes in the feature space region, making it difficult to accurately classify such gray area samples, and thus prone to misjudgment, affecting classification accuracy.

[0035] 2) Not suitable for non-convex sample distribution structure. The distribution of movement characteristics often presents a non-linear, non-convex structure, for example, some high-risk samples may be distributed in the "surrounding zone" between low-risk areas. Traditional classifiers based on linear or convex assumptions, such as SVM (Support Vector Machine) and LDA (Linear Discriminant Analysis), are difficult to accurately fit such complex boundaries.

[0036] 3) In actual data, some areas may be occupied by a certain class of samples, while other classes are sparse or even missing, forming the so-called "area exclusivity" phenomenon. Existing machine learning models are prone to overfitting in these concentrated areas, and unstable performance in the boundary or sparse area, thereby affecting the classification performance.

[0037] To overcome the defects in the prior art, the present application provides a fall risk classification method based on daily walking foot pressure data. Referring to Figure 1 The method comprises the following steps:

[0038] Step S110, collecting foot pressure data during daily walking of human body.

[0039] The foot pressure data (i.e. plantar pressure data) collection device can use an intelligent insole system or other foot pressure collection system during walking.

[0040] For example, an intelligent foot pressure insole system is used to collect the pressure signals of both feet during daily walking of human body. The intelligent foot pressure insole system is composed of a pair of pressure insoles and a smart phone (with an app installed with pressure data display and saving function), and the pressure insoles and the smart phone transmit data through Bluetooth. The pressure insoles and normal insoles have the same appearance and size and usage method, and can be directly embedded in shoes. In an embodiment, a single pressure insole has 32 pressure sensing unit channels, and a pair of pressure insoles has 64 pressure sensing unit channels. The 64-channel foot pressure data is collected by the intelligent foot pressure insole system at a frequency of 25Hz, and then displayed and saved on the smart phone app. In addition, the body weight information of each subject can also be collected synchronously.

[0041] Step S120, pre-processing and feature extraction of the collected foot pressure data to obtain foot pressure feature data set, and then constructing training set, validation set and test set.

[0042] For example, the pressure data of each channel is filtered by a 4th order low-pass Butterworth filter with a cut-off frequency of 15Hz. After filtering, the pressure data of each channel is divided by the body weight of the corresponding subject for normalization. Then, the continuous pressure data of the left 32 channels is accumulated, and the accumulated pressure sum is recorded as Sum_L_pre; similarly, the continuous pressure data of the right 32 channels is accumulated, and the accumulated pressure sum is recorded as Sum_R_pre. Figure 2The continuous foot pressure data is divided according to a complete gait cycle and according to gait phase data segmentation operation, the abscissa (x-axis) represents the index of the sampling point number, and the ordinate represents the pressure sum (for example, the unit is Pa), wherein the curves of Sum_L_pre and Sum_R_pre changing with time during daily walking are shown, the red curve is Sum_L_pre, and the blue curve is Sum_R_pre. The left pressure data between x+2 and t+1 is the pressure data during the left single support phase. The left pressure data between x+1 and x+2 is the pressure data during the left side of the double support phase. The left pressure data between t and x is the pressure data during the left side of the double support phase. The right pressure data between x and x+1 is the pressure data during the right single support phase. The right pressure data between t and x is the pressure data during the right side of the double support phase. The right pressure data between x+1 and x+2 is the pressure data during the right side of the double support phase. The continuous foot pressure data is cut into several segments according to a single complete gait cycle, each segment only contains the continuous foot pressure data of the right side and the corresponding continuous foot pressure data of the left side in a single complete gait cycle. In addition, in order to capture the gait change more finely, the left and right continuous foot pressure data in a single gait cycle can be further cut according to the gait phase parameters defined by medicine (left single support phase, right single support phase, rising stage of double support phase and descending stage of double support phase), respectively into several foot pressure data segments related to the left side (left single support phase data segment, left side rising segment of double support phase and left side descending segment of double support phase) and several foot pressure data segments related to the right side (right single support phase data segment, right side rising segment of double support phase and right side descending segment of double support phase).

[0043] In the above manner, the continuous bilateral foot pressure data can be divided into several data segments with a single gait cycle. Then, for each data segment with a single gait cycle, further extract the pressure data during the left single support phase, the pressure data during the left side of the double support phase, the pressure data during the left side of the double support phase, the pressure data during the right single support phase, the pressure data during the right side of the double support phase, and the pressure data during the right side of the double support phase. The purpose of dividing the continuous foot pressure data and further extracting the gait phase pressure data segment is to smoothly carry out the subsequent foot pressure feature extraction work.

[0044] In one embodiment, four feature extraction windows are defined, respectively, feature extraction based on bilateral pressure data of a complete single gait cycle, feature extraction based on pressure data during left single support phase and right single support phase, feature extraction based on pressure data during left side rise of double support phase and right side rise of double support phase, and feature extraction based on pressure data during right side fall of double support phase and left side fall of double support phase. Specifically, the same series of feature extraction operations are implemented based on the four feature extraction windows (assuming that the extracted feature dimension is d). Assuming that n foot pressure data segments with complete gait cycles are segmented from continuous daily gait data of a plurality of subjects, the four feature extraction windows of different scales will form four data sets with a size of n*d, respectively. The labels of the samples in the data sets are defined as follows: if n_j foot pressure data segments with complete gait cycles are segmented from the daily gait data of the jth subject, and the fall risk level of the current subject is p_j, then the labels of the n_j samples are all p_j. Through this design, the foot pressure feature data set can be obtained, and then the training set, the validation set and the test set can be constructed for the training of the classifier, for example, the training set reflects the corresponding relationship between the foot pressure feature data samples and the fall risk level.

[0045] In step S130, the fuzzy classifier is trained using the training set and the validation set, and the fuzzy classifier obtains the fall risk classification result by calculating the membership degrees between the samples for the foot pressure feature data.

[0046] Based on the obtained foot pressure feature data set, the membership degrees between the samples and the membership degrees of the samples to the entire category samples can be further calculated. For example, the Gaussian-Mahalanobis distance fuzzy membership calculation method or other membership function calculation methods can be used.

[0047] 1) Unconstrained Gaussian-Mahalanobis distance fuzzy membership calculation method

[0048] In order to better understand the method, the definition of the prototype point is given. The prototype point refers to a set of sample points in each category that represent the distribution boundary features of the category. For example, the prototype point can be obtained by the K-means clustering method. For the current category , the prototype point The specific calculation process is as follows:

[0049] (1)

[0050] wherein, is the sample in the category ; and is the sample that does not belong to the category , but belongs to other categories; is the Euclidean norm, also called L2 distance; is the current prototype point of the current class. K is the index of the sample in the current class. is the total number of prototype points of the current q class. is the sample set of the i-th group after the data set is split into K groups.

[0051] After obtaining the prototype point of each class, the membership between any sample point in the data set and the class is calculated. The concept of membership is the core concept used to describe the relationship between elements and sets, which quantifies the degree to which an element belongs to a certain set. For example, the membership The calculation process is represented as:

[0052] (2)

[0053] (3)

[0054] (4)

[0055] where, is the membership between the sample point and the class , is the sample point to be calculated the membership with the class . is the covariance matrix of all prototype points in the class , is the parameter to control the decay steepness, T represents the transpose, is the prototype point in the class . is the index of the prototype point in the class . K is the total number of prototype points of the current class. Both formula (2) and (3) are the calculation functions of , and any one can be selected in actual application. Formula (4) is the membership function between and the prototype point.

[0056] It should be noted that the general formula of the Gaussian function is , d which represents the distance measure (Euclidean distance) between the sample x and the prototype point (such as the cluster center or other reference point). The general formula of the Mahalanobis distance is​​ The conventional membership function is usually a global formula, not a result of accumulation of several membership functions. The conventional membership function has hypercube, hypersphere or regular polyhedron functions. If the distribution of data clusters is initially a hypercube, hypersphere or regular polyhedron, the result of conventional membership calculation is good. However, if the distribution of data clusters is not the assumed hypercube, hypersphere or regular polyhedron, for example, the distribution of data clusters is horseshoe-shaped or banana-shaped, the error of membership calculation of the sample x to the whole class is very large. In view of the distribution characteristics of foot pressure data, the membership calculation formula used in the present application is accumulated from the membership of the sample x to several prototype points, and thus the distribution of data clusters does not need to be preset in advance. That is, no matter how the distribution of data clusters is, the Gaussian-Mahalanobis distance fuzzy membership function can well fit the distribution of clusters.

[0057] It should be noted that the membership function with constraint is like pouring plastic into a fixed mold. Taking a "spherical mold" or "cubic mold", pouring plastic into it, the obtained membership region can only be a sphere or a cube. If the data cluster is really spherical, the effect is good, but if the cluster is like "horseshoe-shaped" or "banana-shaped", it is difficult to fit. The membership function without constraint is like using many small mud balls to make a puzzle, and a pile of small mud balls (each mud ball is a local membership function, for example, a small Gaussian peak) is placed at the boundary of the cluster to freely assemble. The shape of the assembled puzzle can be arbitrarily curved, bifurcated and with holes, and can be determined according to data.

[0058] In summary, the calculation process of the membership does not assume that the cluster is spherical, cubic or any regular body, but is assembled by a plurality of "local Gaussian peaks". The position and shape (covariance matrix Sigma) of each peak are derived from the data itself. Therefore, the calculation process function of the membership is called the unconstrained Gaussian-Mahalanobis distance fuzzy membership function.

[0059] It should be noted that is the covariance matrix of all prototype points in the class , reflecting the correlation between the characteristics of foot pressure and the main direction of the cluster. is a parameter for controlling the steepness of decay. In order to facilitate the understanding of the influence of parameter on membership, the process of the unconstrained Gaussian-Mahalanobis membership function varying with is presented in the form of a graph. For example, Figure 3 , Figure 4 and Figure 5 present =0.5, =2.0 and The change in the membership function at 5.0, MF1 and MF2 represent different values. The Gaussian-Markov membership function of the value Figure 3 , Figure 4 and Figure 5 The x-axis represents the sample point. x， The vertical axis represents the sample points. x membership degree .

[0060] To further illustrate the concept of Gaussian-Mahalanx distance fuzzy membership, the following explanation is provided. Assume the current foot pressure feature is a two-dimensional feature, see [link to relevant documentation]. Figure 6 As 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 (labeled as...). x 1 and x 2) If both the x-axis and y-axis are within the range [0,1], then a point in this 1×1 coordinate plane is denoted as point t. Assume the category... q All samples are prototype points (all samples can be directly used as prototype points, or the boundary points can be used as prototype points through clustering. Compared with directly using all samples as prototype points, fewer prototype points are obtained through clustering, which can significantly reduce the computation of membership degree. Therefore, it is recommended to use clustering to obtain prototype points. In this example, all sample points are used as prototype points). Then, the membership degree from t to category q is calculated using formula (2), and marked with color in this 1×1 coordinate plane, such as Figure 6 As shown in (b), the deeper the red color, the higher the membership degree. It can be seen that the color is deep red where the original sample clustering is very high. To further illustrate this, t is assigned to a category. q The membership degree is displayed in three-dimensional space, such as Figure 6 The three-dimensional coordinate form shown in (c) is used, where the x-axis and y-axis are the planar coordinates of t. x 1 and x 2 indicates that the z-axis represents the membership degree. Similarly, it can be seen that the further away from the prototype point, the more t corresponds to the category. q The lower the membership degree, the higher it is when they are closer.

[0061] 2) Fall Risk Classification Method Based on Gaussian-Mahalanx Distance Fuzzy Membership Degree

[0062] By using the Gaussian-Mahalanx distance fuzzy membership degree calculation method, fall risk classification and prediction can be achieved.

[0063] Assume the sample to be classified is And the existing dataset contains Class samples, respectively kind, Class, Class, Class.

[0064] In each feature extraction window , the following operation is performed once:

[0065] Step 11: Obtain the boundary prototype points in each class of samples by clustering method where, denotes the class, ranging from 1 to c, is the index of the current prototype point in the class sample.

[0066] Step 12: Calculate the membership of samples in each class where, denotes the class, ranging from 1 to c.

[0067] Step 13: The membership of each class in the feature extraction window is denoted as where, denotes the class, ranging from 1 to c, k is the index of the feature extraction window, is the number of feature extraction windows (for example, m is 4).

[0068] After performing the above three steps in each feature extraction window, the following operations are performed:

[0069] Step 21: Multi-scale fusion. Linearly weight the membership of samples in each feature extraction scale window to the class Finally, the final membership of sample to the class is:

[0070] (5)

[0071] where, denotes the class, ranging from 1 to c.

[0072] Step 22: Find the class of samples with the maximum membership to sample , denoted as:

[0073] (6)

[0074] wherein, represents a class, the range of c is between 1 and c.

[0075] Step 23: obtain the final prediction result, i.e. the sample belongs to the class .

[0076] Step S140, for the actual collected foot pressure data, using the trained fuzzy classifier, obtain the fall risk classification result.

[0077] 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 foot pressure data in the target daily walking process; pre-processing 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 result.

[0078] To further understand the training and actual application process of the fuzzy classifier, Figure 7 a process 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; pre-processing the foot pressure data; extracting features from the foot pressure data; normalizing the data to obtain a training set; determining the fuzzy degree estimation of the sample x for each class in each feature extraction window; determining an optimal parameter in each feature extraction window using a validation set; linearly weighting the membership of the sample x in each class in different feature extraction windows; obtaining a trained fuzzy classifier. Finally, using the trained fuzzy classifier, fall risk classification can be performed.

[0079] It should be noted that the above embodiments can be appropriately changed or modified by those skilled in the art without departing from the spirit and scope of the present application. For example, in the filtering process, the filtering algorithm can be replaced by other filtering algorithms. In calculating the prototype points of each sample class, the clustering algorithm can use K-means clustering or hierarchical clustering, density clustering, spectral clustering, etc. For example, in addition to the points at the boundary of each sample class identified by the clustering method as the prototype points, the sample points in the central zone of the sample obtained by clustering or other means can also be used as the prototype points, or all sample points in the current sample class can be used as the prototype points. The fuzzy membership function can be an unconstrained fuzzy membership function based on Gaussian-Mahalanobis distance, or other membership functions can be used, but the classification effect may be reduced. The current sample class membership is obtained by accumulating the membership of the prototype points in the current class or by taking the maximum prototype point membership, and other ways based on the membership of several prototype points to solve the current sample class membership can also be used. For example, a four-scale feature extraction window is used, and more detailed or more rough division of a complete gait cycle can also be used to obtain a more fine-grained or more coarse-grained feature extraction window to measure the gait movement in more scales. In the embodiments of the present application, the final membership of each sample class is obtained by linearly weighting the membership of the same type of sample class under different scales (for example, the weighting coefficients are all 1), and other weighting coefficients or other weighting methods (such as exponential weighting) can also be used.

[0080] In order to further verify the effect of the present application, an experimental verification is carried out, and the Mini-BESTest scale is used for evaluation in the experimental process. The Mini-BESTest scale is a classic balance ability and fall risk evaluation tool commonly used in clinical practice. The higher the score of the Mini-BESTest scale, the better the balance ability of the subject, 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 less than 19 are classified as high-risk fallers; subjects with a scale score between 19 and 24 are classified as medium-risk fallers; subjects with a scale score between 25 and 28 are classified as low-risk fallers. In the experimental verification process, based on the intelligent insole system, first, the bilateral foot pressure data of 45 subjects over the age of 60 during daily straight walking (the number of subjects in the low / medium / high fall risk level is 17 / 15 / 13, respectively) is collected. Subsequently, the method of the present application is used for processing and classification, and the fall risk level classification results of the Gaussian-Mahalanobis distance fuzzy membership fall risk evaluation method based on daily walking foot pressure data in the present application and the classification method in the prior art are shown in Table 1 below.

[0081] Table 1 Comparison results of the present application and the prior art

[0082]

[0083] It should be noted that the classification results of the fall risk level classification algorithm presented in Table 1 are all under the optimal hyperparameters. Cross-individual leave-one-out is a cross-validation method for scenarios where data differences between individuals are significant. The core is to implement the leave-one-out validation process based on "individual" as the basic unit. Specifically, in a dataset containing N independent individuals, each iteration retains all data of a certain individual as the test set, and integrates all samples of the remaining N-1 individuals into the training set. The prediction performance of the retained individual data is used to evaluate the model generalization ability. When all individuals are sequentially used as the test set to complete a round of validation, the performance indicators (such as accuracy) of each iteration are averaged to serve as the final performance evaluation result of the model in the cross-individual scenario.

[0084] In summary, the present application filters the original plantar pressure signal, removes interference and signal noise to ensure accurate subsequent calculation and prevent measurement noise interference. A Gaussian-Mahalanobis kernel is proposed to construct an unconstrained fuzzy membership function, which can effectively model the relative closeness of samples to each class, supporting one-to-many fuzzy membership expression. This structure is particularly suitable for classifying samples in the fuzzy transition zone between different fall risk classes and with non-convex structure distribution, enhancing the recognition robustness of atypical samples. Instead of selecting class centers indiscriminately, the prototype points are preferentially extracted from representative samples in the boundary sensitive area based on clustering, improving the resolution of the fuzzy function at the discrimination boundary and effectively avoiding sample redundancy and membership generalization. In addition, compared with traditional fuzzy functions (such as hypercube, hypersphere or regular polyhedron functions), the fuzzy membership function of the present application does not need to implement the distribution of preset sample clusters, can fit any form of data sample distribution and adaptively depict the covariance structure of different class sample distribution, has stronger boundary direction perception ability and can accurately identify the actual boundary shape of non-spherical and non-convex classes. The present application combines multiple time-scale extracted foot pressure statistical features to form a nested membership evaluation mechanism, which can simultaneously perceive short-term center of gravity deviation and long-term gait trend ("short-term" corresponds to pressure data in the gait phase, "long-term" corresponds to data in the entire gait cycle), and can give a comprehensive fall risk evaluation result based on the performance under different time granularity. In summary, the present application has low complexity, high execution efficiency, small required computing power, can be processed in real time on edge terminals, and has wider application.

[0085] The present application can be a system, a method, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present application.

[0086] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium can be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, 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 mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0087] Various aspects of the disclosure can be described in the context of methods, apparatus (systems) and computer program products according to embodiments of the present application. It is to be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0088] These computer readable program instructions can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions can also be stored in a computer readable storage medium that can be coupled to a computer or other programmable data processing apparatus, such that the computer readable storage medium can provide instructions to the computer or other programmable data processing apparatus, which execute the instructions to produce a computer implemented process, such that the instructions, which execute via the computer or other programmable data processing apparatus, implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0089] The computer readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable data processing apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0090] The computer program product of the present application can be implemented by a variety of means. For example, the present application can be implemented on hardware, software, firmware, or combinations thereof. In this regard, software or program can include routines, programs, objects, components, data structures, etc., that perform particular tasks. The software or program can be implemented in a desired programming language (e.g., Objective-C, Java, C, C++, etc.). It will be appreciated that a variety of programming platforms and / or operating systems can be used to implement the software or program of the present application. For example, the software or program of the present application can be implemented on a Windows® platform or a UNIX® platform. Further, it will be appreciated that the software or program of the present application can be implemented on a variety of platforms and operating systems consistent with this disclosure.

[0091] Embodiments of the present application have been described above, with examples of the description being exemplary and not exhaustive, and are not limited to the disclosed embodiments. Numerous modifications and adaptations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The scope of the present application is defined by the appended claims.

Claims

1. A fall risk classification method based on daily ambulatory foot pressure data, characterized by, The method comprises the following steps: Collecting foot pressure data during a target daily walking process; 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, wherein the fuzzy classifier obtains the fall risk classification result by calculating the membership degree between samples for the foot pressure feature data; During the training of the fuzzy classifier, the membership degree between samples is calculated by the following steps: Obtaining a prototype point in each sample class, denoted as: wherein, is a sample in the class ; is a sample not in the class , but in other classes; is the Euclidean norm, is a prototype point in the current class sample, , K is the total number of prototype points of the current class, denotes the sample set of the i-th group after the class q is split into K groups, denotes the sample set not in the class q ; Compute membership between any sample point in the training set and class ;​ wherein the sample points between the classes membership are calculated according to the following formula: in, Sample points With class Membership degree between them It is a category The covariance matrix of all prototype points in the matrix. It is a parameter that controls the steepness of the decay. It is a category The prototype point in the middle, Is the prototype point in the category? Index in K Is the current class The total number of prototype points; wherein, or the sample points between the classes membership is calculated according to the following formula: wherein ， is a sample point and a membership degree between the class , is a covariance matrix of all prototype points in the class , is a parameter to control the steepness of the decay, is a prototype point in the class , is an index of the prototype point in the class , K is a total number of prototype points of the current class .

2. The method of claim 1, wherein, The fall risk classification result is obtained based on the following formula: Wherein: wherein the sample points The class to which the sample point belongs is , is the sample point The membership of the class , denotes the class, is the class index, k is the index of the feature extraction window, is the number of feature extraction windows.

3. The method of claim 1, wherein, The prototype point is a point at the boundary of each sample class identified based on a clustering method, or multiple sample points in the central zone of the sample are used as the prototype point, or all sample points in the current sample class are used as the prototype point.

4. The method of claim 2, wherein, 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 a left single support phase and pressure data during a right single support phase; Feature extraction based on pressure data during a left side rise in a double support phase and feature extraction based on pressure data during a right side rise in a double support phase; Feature extraction based on pressure data during a right side descent in a double support phase and pressure data during a left side descent in a double support phase.

5. The method of claim 3, wherein, The clustering method is K-means clustering, hierarchical clustering, density clustering or spectral clustering.

6. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method according to any one of claims 1 to 5.

7. A computer device comprising a memory and a processor, having stored on the memory a computer program capable of running on the processor, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

Images