CN114495152A - Gait data classification method, computer readable storage medium and device - Google Patents
Gait data classification method, computer readable storage medium and device Download PDFInfo
- Publication number
- CN114495152A CN114495152A CN202111547334.7A CN202111547334A CN114495152A CN 114495152 A CN114495152 A CN 114495152A CN 202111547334 A CN202111547334 A CN 202111547334A CN 114495152 A CN114495152 A CN 114495152A
- Authority
- CN
- China
- Prior art keywords
- data
- gait data
- classification model
- gait
- classification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a gait data classification method, a computer readable storage medium and a device, wherein the method comprises the following steps: acquiring gait data to be classified; the gait data to be classified is knee joint six-degree-of-freedom motion data; building a classification model based on decision fusion and training the classification model to obtain a trained classification model; building the sub-models of the classification model comprises the following steps: the system comprises a multi-parameter convolution neural network, a multi-parameter residual error network, an attention-based long-time and short-time memory network and a gate control cycle unit network; and inputting the gait data to be classified into the trained classification model to obtain a classification result. According to the method, the integrated learning network based on the self-training algorithm is built and used, the six-degree-of-freedom dynamic data of the Knee joint acquired by the Opti-Knee three-dimensional motion analysis system can be analyzed, the prediction result is output in real time, and the gait data can be accurately classified.
Description
Technical Field
The present invention relates to the technical field of gait recognition, and in particular, to a gait data classification method, a computer-readable storage medium, and a device.
Background
Gait analysis is a common mode for studying walking laws, a knee joint is used as an important joint and has important influence on gait, and due to the complex structure of the knee joint and the six-degree-of-freedom motion mode of the knee joint, the obtained six-degree-of-freedom motion data of the knee joint is complex, good classification is difficult to realize by adopting a data processing method in the prior art, and accurate gait analysis cannot be realized.
Therefore, a method for classifying gait data is needed.
Disclosure of Invention
In view of the deficiencies of the prior art, the present invention is directed to a gait data classification method, a computer-readable storage medium and a device.
The technical scheme adopted by the invention is as follows:
in a first aspect, a gait data classification method comprises:
acquiring gait data to be classified; the gait data to be classified is knee joint six-degree-of-freedom motion data;
building a classification model based on decision fusion and training the classification model to obtain a trained classification model; building the sub-models of the classification model comprises the following steps: the system comprises a multi-parameter convolution neural network, a multi-parameter residual error network, an attention-based long-time and short-time memory network and a gate control cycle unit network;
and inputting the gait data to be classified into the trained classification model to obtain a classification result.
Optionally, the gait data classification method, wherein the knee joint six-degree-of-freedom motion data includes: abduction and adduction, anteflexion and retroextension, internal rotation and external rotation, forward and backward displacement, internal and external displacement and up and down displacement.
Optionally, the gait data classification method, wherein the training of the classification model specifically includes the following steps:
s200, acquiring a plurality of gait data, and forming a training set by using part of the gait data as training data;
s201, dividing the gait data in the training set into a first labeled data subset and a non-labeled data subset, and training a classification model by using the gait data in the first labeled data subset;
s202, predicting class labels of the gait data in the unlabeled data subset by adopting a trained classification model, setting a prediction probability threshold as a judgment condition, and recording the class labels as first labels when the prediction probability of the gait data in the unlabeled data subset exceeds the prediction probability threshold;
s203, merging the gait data with the first label into the first labeled data subset to obtain a second labeled data subset, and training the classification model again by using the gait data in the second labeled data subset; repeating the step S202;
and S204, when the class label meeting the judgment condition is not obtained any more after the step S202 is repeated, outputting the classification model trained for the last time as a trained classification model.
Optionally, the gait data classification method further includes, after step S204: performing performance evaluation on the classification model trained for the last time by adopting test data; when the evaluation result is lower than the expected result, performing secondary training on the classification model; the test data is the gait data left after the training data is removed from the plurality of gait data.
Optionally, the gait data classification method, wherein the multi-parameter convolutional neural network comprises: six convolution channels corresponding to the six degrees of freedom of the gait data, wherein each convolution channel comprises seven convolution layers, three pooling layers and one flattening layer, the three pooling layers are arranged in an inserting mode, and the flattening layer is connected with the adjacent pooling layer; the activation function of each convolution layer is a linear rectifying unit.
Optionally, the gait data classification method, wherein the multi-parameter residual network includes: and each convolution channel comprises three residual modules and three jump connection layers, each residual module comprises three convolution layers and batch regularization layers which are inserted in the three convolution layers, and each jump connection layer comprises one convolution layer and one batch regularization layer.
Optionally, the gait data classification method, wherein the attention-based long and short term memory network includes: two layers stacked with each other are a long-time memory network layer and an attention layer; the number of hidden neurons is respectively 50 and 20; the attention layer is realized by a full connection layer and a softmax activation function.
Optionally, the gait data classification method, wherein the gated loop unit network includes: two layers of gate control circulation unit network layers are stacked, and the number of hidden neurons is 30 and 20 respectively.
In a second aspect, a computer readable storage medium stores one or more programs, which are executable by one or more processors to implement the steps in the gait data classification method as described above.
In a third aspect, a gait data classification device includes: a processor and a memory; the memory has stored thereon a computer readable program executable by the processor; the processor, when executing the computer readable program, implements the steps in the gait data classification method as described above.
Has the advantages that: the invention provides a gait data classification method, wherein a classification model based on decision fusion is built and trained to obtain a trained classification model; building the sub-models of the classification model comprises the following steps: and inputting the acquired gait data to be classified into a trained classification model to obtain a classification result. Compared with the prior art, the gait data classification method provided by the invention has the characteristics of good repeatability and high prediction accuracy.
Drawings
Fig. 1 is a flowchart of a gait data classification method according to this embodiment.
Fig. 2 is six-degree-of-freedom single-cycle mean data of the knee joint in one walking cycle according to an embodiment of the present invention.
Fig. 3 is a flow chart of a self-training algorithm according to an embodiment of the present invention.
Fig. 4(a) is a structural block diagram of an MCNN convolutional neural network, and fig. 4(b) is a structural block diagram of an MResNet convolutional neural network.
Fig. 5(a) is a structural block diagram of an alsm recurrent neural network, and fig. 5(b) is a structural block diagram of a GRU recurrent neural network.
Fig. 6 is a block diagram of an integrated network architecture based on self-training.
Fig. 7 is a schematic structural diagram of a terminal according to this embodiment.
Detailed Description
The present invention provides a gait data classification method, medium and terminal, and in order to make the objects, technical solutions and effects of the present invention clearer and clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. As used herein, the term "and/or" includes all or any element and all combinations of one or more of the associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The invention will be further explained by the description of the embodiments with reference to the drawings. The present embodiment provides a method, as shown in fig. 1, including:
s10, acquiring gait data to be classified; the gait data to be classified is knee joint six-degree-of-freedom motion data;
s20, building a classification model based on decision fusion and training the classification model to obtain a trained classification model; building the sub-models of the classification model comprises the following steps: the system comprises a multi-parameter convolution neural network, a multi-parameter residual error network, an attention-based long-time and short-time memory network and a gate control cycle unit network;
and S30, inputting the gait data to be classified into the trained classification model to obtain a classification result.
In order to realize the mining and analysis of the dynamic data of the six degrees of freedom of the knee joint and the accurate classification of the gait data, the invention adopts four sub-networks of a Multi-parameter Convolutional Neural Network (MCNN), a Multi-parameter Residual error Network (MResNet), an Attention-based Long Short-term Memory Network (ALSTM) and a Gated cycle Unit Network (GRU), and the integrated learning Network based on decision-level fusion is built together to train and predict the gait data. And realizing accurate classification of gait data.
In this embodiment, the Multi-parameter Convolutional Neural Network (MCNN) has a structure as shown in fig. 4(a), and includes six convolution channels corresponding to six degrees of freedom of the gait data, each data sample is respectively input into a convolution channel corresponding to each parameter, each convolution channel includes a total of 7 convolution layers (Conv1D), 3 max pooling layers (MaxPool1D), and 1 flattening layer (Flatten), and each convolution layer uses a Linear rectifying Unit (Rectified lu) as an activation function. Convolution operation can realize the extraction of the features to obtain feature vectors. The maximum pooling layer can realize down-sampling operation and compress the input feature vectors, so that the feature vectors are shortened, and the calculation complexity of the model is simplified; and on the other hand, feature compression is carried out to extract main features. Through the final flattening layer, the features of the corresponding parameters output by the convolution channel can be output in a vector form. The activation function is to determine whether the output of each neuron reaches a threshold, that is, whether the feature intensity of a certain part of the data reaches a certain standard, and if not, the feature intensity is set to 0, which indicates that the features extracted from the part of the data do not have obvious effect on classification, so that the features are not output. ReLU is used as a popular activation function nowadays, and can not only enable partial neurons to be set to be 0 so as to improve the sparsity of the network and reduce the interdependency among parameters, but also relieve the occurrence of overfitting to a certain extent; and the ReLU can reduce the amount of computation of the network.
After the characteristics are extracted, the 6 characteristic vectors are spliced (connected) along the axial direction of the channel to obtain the characteristic vector of the multi-parameter data, and then the characteristic vector is input into a full connection layer (Dense) with 1000 neurons, and the activation function is ReLU. The output vector of the full-connection layer is input into the dropout layer for regularization, and the overfitting condition during training is relieved to a certain extent. And inputting the final feature vector into the last full-connected layer, and performing nonlinear classification by using a softmax classifier to obtain a final analysis result.
MCNN uses cross entropy as a loss function. And calculating a cross entropy loss function through the prediction probability obtained by feedforward propagation, realizing the back propagation of the error by using a chain rule of derivatives, and calculating the derivatives of all parameters of the model. Finally, the derivative of the model parameters in back propagation is updated by using an Adam optimizer to minimize the loss function and obtain more accurate prediction performance.
In this embodiment, the Multi-parameter Residual Network (MResNet) has a structure as shown in fig. 4(b), where the feature extraction paths of MResNet and MCNN are similar, and the biggest difference is that the convolution channels are no longer simple convolution layer and maximum pooling layer stacks, but a jump connection is introduced, so that each convolution channel has a structure similar to that of the Residual Network. Each convolution channel consists of 3 residual blocks, each containing 3 convolution layers and 3 batch regularization layers (batchnormaize), with the activation function of ReLU. In the jump connection, the invention does not directly map the input and Add the feature vector output after convolution, but simultaneously sets 1 convolution layer and 1 batch regularization layer in the jump connection, extracts the features of the input with different scales from the main path, and finally adds (Add) the feature vector elements obtained by the two paths for output.
After feature extraction is carried out on 6 channels, feature vectors are processed the same as MCNN, the channels are spliced along the axial direction, then a global average pooling layer (GlobalatAveragePool 1D) is input for data dimension reduction, the data dimension reduction is input into a full connection layer, and a softmax classifier is used for nonlinear classification, so that a final analysis result is obtained. The loss function is the same as the choice of optimizer and MCNN.
Compared with the traditional CNN stacking convolution structure mode, the jump connection in the residual error network can avoid overfitting caused by excessive extraction of redundant features on the basis of increasing the depth to improve the classification accuracy. The MResNet designed in the invention integrates a traditional residual error network and a multi-parameter mechanism in the MCNN, is more suitable for data size and can better extract effective characteristics.
In this embodiment, the Attention-based Long-Short-term Memory Network (ALSTM) is stacked by two LSTM layers, and the number of hidden neurons is 50 or 20. The output of the LSTM layer is input into the attention layer for analysis. After the feedforward network training is completed, the timing related features can be obtained from the LSTM layer. The attention layer is implemented by a fully connected layer through which the weight of each timing feature is calculated, together with the softmax activation function, as indicated by the box in fig. 5 (a). And normalizing the weights obtained by the full connection layer through a softmax function to ensure that the sum of the attention weights is 1. And then, performing point multiplication (Multiply) on the attention weight output by the full connection layer and the input feature vector, so that the whole classification network can pay more attention to the feature with larger weight, namely the feature more related to classification. Meanwhile, because the data comprises data of 6 parameters, namely multidimensional data, each dimension of data is designed to independently calculate a group of attention weight vectors, so that the network can better mine the relation between different parameters and gait. The output of the attention layer is finally input into the full connection layer after being processed by the flattening layer, and is subjected to nonlinear classification by using a softmax classifier to obtain a final analysis result. The loss function is the same as the choice of optimizer and MCNN. According to the method, ALSTM is used, and the time sequence information characteristics are captured and the time period in the gait cycle is acquired to have a better classification effect in a mode of combining LSTM and an attention mechanism, so that the negative influence of the redundant characteristics generated in certain time periods with low correlation on the model performance is reduced.
In this embodiment, the Gated Recurrent Unit (GRU) has a structure as shown in fig. 5(b), and the LSTM and the GRU are used as Recurrent neural networks commonly used in various applications, and the important timing features are extracted and stored through various gating functions, and the GRU has one less gating function than the LSTM, so the calculation amount is reduced. The GRU in the invention is composed of two Gru stacked layers, and the number of hidden neurons is 30 and 20 respectively. And then extracting to obtain a feature vector, inputting the feature vector into the full-connected layer, and performing nonlinear classification by using a softmax classifier to obtain a final analysis result. The loss function is the same as the choice of optimizer and MCNN.
In an implementation manner of the embodiment, through an Opti-Knee three-dimensional motion analysis system, six-degree-of-freedom motion data of an individual single-leg Knee joint can be captured, including: abduction/adduction (unit: degree), anteversion/hindextension (unit: degree), pronation/supination (unit: degree), anteroposterior displacement (unit: cm), internal and external displacement (unit: cm), and up and down displacement (unit: cm). As shown in fig. 2, the present invention collects the mean data of six degrees of freedom single cycle of knee joint, which is obtained when each subject completes one walking cycle on flat ground, and 96 continuous time sampling points are collected, that is, each subject will obtain a group of 12 cases of dynamic time sequence data containing six degrees of freedom of both legs after completing the designated action, and the length of each case is 96. In the invention, six-degree-of-freedom data of the knee joint of the single-leg test is taken as one data sample, namely, each subject acquires 2 data samples in total, and the research related to the invention has 410 data samples in total. Meanwhile, each data sample has a tag data, 0 represents normal, and 1 represents damage. Therefore, for data output by the Opti-Knee three-dimensional motion analysis system, the size of the data is modified into a 6-case vector set of 96 × 6, and the data is input into a deep learning network in cooperation with corresponding labels to be trained and tested.
In one implementation of the embodiment, a self-trained semi-supervised algorithm is utilized in the deep learning network to improve the performance of each sub-network in the integrated network. The self-training algorithm trains an initial classification network by using a small amount of labeled data, and predicts unlabeled samples by using the classification network, so that unlabeled data with higher prediction probability is obtained and is expanded into a label set. This procedure is actually the classification network using its own prediction to improve itself. As one of common methods in the semi-supervised algorithm, the self-training algorithm does not need to estimate parameters like a generative model, does not need to construct a complex graph model like a semi-supervised learning algorithm based on a graph regularization frame, and does not need to establish specific assumed conditions like cooperative training. The self-training algorithm can complete a complex semi-supervised learning task only by 1 classification model, a small amount of labeled data and a certain amount of unlabeled data.
The self-training algorithm comprises the following specific steps:
the method comprises the following steps: dividing training data into a training set and a test set, dividing labeled data and unlabeled data in the training set, and training a sub-network by using the labeled training set data;
step two: predicting class labels of the unlabeled data by using the trained sub-network, setting a prediction probability threshold as a judgment condition, and taking the prediction labels as 'pseudo labels' (first labels) of the unlabeled data when the prediction probability of the unlabeled data exceeds the threshold;
step three: merging the data with the pseudo labels into a label training data set, retraining the sub-network by using the data, and repeating the step two;
step four: and when the step two is repeated, the data of the pseudo labels meeting the judgment condition is not obtained any more, the sub-network trained for the last time is output as a final network, the final network is used for predicting the class labels of the test data set, and the performance of the sub-network is further evaluated.
Self-training is used as a traditional semi-supervised algorithm, and can be used for carrying out multiple iterations by generating a pseudo label and adding a training mode to obtain a better decision boundary and realize better network performance. Meanwhile, the phenomenon that the training is easy to overfit when the data is less is well relieved. The specific algorithm flow is shown in fig. 3.
Since it is inevitable that data is mis-labeled during the self-training process, the mis-labeled "pseudo-label" data becomes noisy data in the training set. The effectiveness of the samples in the training set is often not questioned in specific tasks, but the noise data in the training set directly destroys the real structural information of the data, thereby affecting the training effect of the classification model. Meanwhile, as the model iteration process continues, the classification performance of the model will be continuously weakened. However, different networks face the same data and can often learn the characteristics of different viewing angles, so in order to reduce the influence of noise data on the performance of the classification model, the invention adopts a multi-network integration mode in the subsequent design to reduce the influence of noise data on a single classifier, and realizes the improvement of the overall diagnosis performance and improves the generalization capability of the diagnosis model through the integration mode.
Meanwhile, although the method in the invention alleviates noise interference in an integrated manner and does not directly delete noise data in the self-training iteration process of the classification model, the arrangement is just the way, so that the time consumption of the algorithm in the self-training iteration and the program execution is obviously reduced, which means that the method has a practical application prospect and also reflects the excellent performance of the classification model in the invention from another aspect.
In the embodiment, the self-training algorithm is utilized, and the four deep learning networks designed above are combined to jointly build an integrated learning network based on decision-level fusion, so as to train and predict the data extracted by the Opti-Knee system. The classification is realized through different characteristics learned in the same data by a plurality of classification networks, so that the generalization of the algorithm is improved. The structure and input-output process of the integrated network are shown in fig. 6. In the invention, firstly, the output of each sub-network is obtained by respectively training and learning four sub-networks in a self-training way (namely, the output of each sub-network is obtained
Based on a decision fusion mode, soft voting (soft voting) mechanism is used for averaging the prediction probabilities of the same class output by each sub-network to obtain the output of the final integrated network
And
namely:
and outputting the category with higher prediction probability as the prediction class label of the data sample by the integrated network.
The invention aims to realize the classification of normal gait and abnormal gait for the gait condition of a subject, and the classification result is evaluated by five indexes, including Accuracy (Accuracy, Acc), Specificity (Spe), Sensitivity (Sensitivity, Senn), F1 score (F1-score), Area Under ROC curve (AUC). Acc represents the overall classification accuracy of the invention, Spe and Sen represent the classification accuracy of the invention for negative samples and positive samples respectively, F1-score gives consideration to the accuracy and recall of the classifier, and is also an index for defining the accuracy of the classifier. The ROC curve is drawn through the true positive rate and the false positive rate, the performance of the classifier can be reflected, and the AUC is a quantitative index for quantifying the ROC curve. The larger the above five index values are, the better the performance of the classifier is represented. Acc, Spe, Sen, F1-score are specifically defined as follows:
wherein, tp (true positive), tn (true negative), fp (false positive), fn (false negative) represent the number of true positive, true negative, false positive, false negative samples in the classification result, respectively.
Based on the gait data classification method, the invention also provides a computer-readable storage medium, which stores one or more programs, wherein the one or more programs can be executed by one or more processors to realize the steps of the gait data classification method according to the above embodiment.
Based on the above gait data classification method, the invention also provides a gait data classification device, as shown in fig. 7, which comprises at least one processor (processor) 30; a display screen 31; and a memory (memory)32, which may also include a communication interface (communication interface)33 and a bus 34. The processor 30, the display 31, the memory 32 and the communication interface 33 can communicate with each other through the bus 34. The display screen 31 is configured to display a user guidance interface preset in the initial setting mode. The communication interface 33 may transmit information. The processor 30 may call logic instructions in the memory 32 to perform the methods in the embodiments described above. Furthermore, the logic instructions in the memory 32 may be implemented in software functional units and stored in a computer readable storage medium when sold or used as a stand-alone product. The memory 32, which is a computer-readable storage medium, may be configured to store a software program, a computer-executable program, such as program instructions or modules corresponding to the methods in the embodiments of the present disclosure. The processor 30 executes the functional application and data processing by executing the software program, instructions or modules stored in the memory 32, i.e. implements the method in the above-described embodiments. The memory 32 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. Further, the memory 32 may include high speed random access memory and may also include non-volatile memory. For example, a variety of media that can store program codes, such as a usb disk, a removable hard disk, a Read-only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, may also be transient storage media. In addition, the specific processes loaded and executed by the instruction processors in the storage medium and the terminal are described in detail in the method, and are not stated herein.
Furthermore, by utilizing the gait data classification device, the early accurate diagnosis of the ACL fracture damage of the knee joint can be realized by analyzing the gait data of the patient with the ACL fracture damage of the knee joint Anterior Cruciate Ligament (ACL), and the real-time automatic diagnosis result output of the ACL damage can be realized. The device is used for carrying out a plurality of experiments on the data of 205 subjects, and the results show that the device can still output the prediction result in a short time and keep a high accuracy rate on the basis of no wound, short time and low cost, thereby providing important technical assistance for the automatic diagnosis of ACL damage.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; while the invention has been described in detail with reference to the foregoing examples, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A gait data classification method is characterized by comprising the following steps:
acquiring gait data to be classified; the gait data to be classified is knee joint six-degree-of-freedom motion data;
building a classification model based on decision fusion and training the classification model to obtain a trained classification model; building the sub-models of the classification model comprises the following steps: the system comprises a multi-parameter convolution neural network, a multi-parameter residual error network, an attention-based long-time and short-time memory network and a gate control cycle unit network;
and inputting the gait data to be classified into the trained classification model to obtain a classification result.
2. The gait data classification method according to claim 1, wherein the knee joint six-degree-of-freedom motion data includes: abduction, adduction, anteversion, retroversion, pronation, supination, pronation, anteroposterior displacement, medial and lateral displacement, and up and down displacement.
3. The gait data classification method according to claim 1, characterized in that the training of the classification model comprises the following steps:
s200, acquiring a plurality of gait data, and forming a training set by using part of the gait data as training data;
s201, dividing the gait data in the training set into a first labeled data subset and a non-labeled data subset, and training a classification model by using the gait data in the first labeled data subset;
s202, predicting class labels of the gait data in the unlabeled data subset by adopting a trained classification model, setting a prediction probability threshold as a judgment condition, and recording the class labels as first labels when the prediction probability of the gait data in the unlabeled data subset exceeds the prediction probability threshold;
s203, merging the gait data with the first label into the first labeled data subset to obtain a second labeled data subset, and training the classification model again by using the gait data in the second labeled data subset; repeating the step S202;
and S204, when the class label meeting the judgment condition is not obtained any more after the step S202 is repeated, outputting the classification model trained for the last time as a trained classification model.
4. The gait data classification method according to claim 3, further comprising, after said step S204: performing performance evaluation on the classification model trained for the last time by adopting test data; when the evaluation result is lower than the expected result, performing secondary training on the classification model; the test data is the gait data left after the training data is removed from the plurality of gait data.
5. The gait data classification method according to claim 1, characterized in that the multi-parameter convolutional neural network comprises: six convolution channels corresponding to the six degrees of freedom of the gait data, wherein each convolution channel comprises seven convolution layers, three pooling layers and one flattening layer, the three pooling layers are arranged in an inserting mode, and the flattening layer is connected with the adjacent pooling layer; the activation function of each convolution layer is a linear rectifying unit.
6. The gait data classification method according to claim 1, characterized in that the multi-parameter residual network comprises: and each convolution channel comprises three residual modules and three jump connection layers, each residual module comprises three convolution layers and batch regularization layers which are inserted in the three convolution layers, and each jump connection layer comprises one convolution layer and one batch regularization layer.
7. The gait data classification method according to claim 1, characterized in that the attention-based long-short memory network comprises: two layers stacked with each other are a long-time memory network layer and an attention layer; the number of hidden neurons is respectively 50 and 20; the attention layer is realized by a full connection layer and a softmax activation function.
8. The gait data classification method according to claim 1, characterized in that the gated-cycle cell network comprises: two layers of gate control circulation unit network layers are stacked mutually, and the number of hidden neurons is respectively 30 and 20.
9. A computer readable storage medium storing one or more programs, the one or more programs being executable by one or more processors to perform the steps of the gait data classification method according to any one of claims 1 to 8.
10. A gait data classification apparatus, comprising: a processor and a memory; the memory has stored thereon a computer readable program executable by the processor; the processor, when executing the computer readable program, implements the steps in the gait data classification method according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111547334.7A CN114495152A (en) | 2021-12-16 | 2021-12-16 | Gait data classification method, computer readable storage medium and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111547334.7A CN114495152A (en) | 2021-12-16 | 2021-12-16 | Gait data classification method, computer readable storage medium and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114495152A true CN114495152A (en) | 2022-05-13 |
Family
ID=81493581
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111547334.7A Pending CN114495152A (en) | 2021-12-16 | 2021-12-16 | Gait data classification method, computer readable storage medium and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114495152A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115171886A (en) * | 2022-07-25 | 2022-10-11 | 北京戴来科技有限公司 | Frozen gait detection method and device based on random forest algorithm and storage medium |
| CN116821730A (en) * | 2023-08-30 | 2023-09-29 | 北京科锐特科技有限公司 | Fan fault detection method, control device and storage medium |
| CN117893575A (en) * | 2024-03-15 | 2024-04-16 | 青岛哈尔滨工程大学创新发展中心 | Ship motion prediction method and system with self-attention mechanism integrated by graph neural network |
| CN117893575B (en) * | 2024-03-15 | 2024-05-31 | 青岛哈尔滨工程大学创新发展中心 | Ship motion prediction method and system with self-attention mechanism integrated by graph neural network |
-
2021
- 2021-12-16 CN CN202111547334.7A patent/CN114495152A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115171886A (en) * | 2022-07-25 | 2022-10-11 | 北京戴来科技有限公司 | Frozen gait detection method and device based on random forest algorithm and storage medium |
| CN116821730A (en) * | 2023-08-30 | 2023-09-29 | 北京科锐特科技有限公司 | Fan fault detection method, control device and storage medium |
| CN116821730B (en) * | 2023-08-30 | 2024-02-06 | 北京科锐特科技有限公司 | Fan fault detection method, control device and storage medium |
| CN117893575A (en) * | 2024-03-15 | 2024-04-16 | 青岛哈尔滨工程大学创新发展中心 | Ship motion prediction method and system with self-attention mechanism integrated by graph neural network |
| CN117893575B (en) * | 2024-03-15 | 2024-05-31 | 青岛哈尔滨工程大学创新发展中心 | Ship motion prediction method and system with self-attention mechanism integrated by graph neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | A directed acyclic graph network combined with CNN and LSTM for remaining useful life prediction | |
| Xing et al. | Diagnosing deep learning models for high accuracy age estimation from a single image | |
| CN114495152A (en) | Gait data classification method, computer readable storage medium and device | |
| AU2021101172A4 (en) | Design and implementation of convolution neural network on edge computing smart phone for human activity recognition | |
| CN113673244B (en) | Medical text processing method, medical text processing device, computer equipment and storage medium | |
| Feng et al. | One-dimensional VGGNet for high-dimensional data | |
| Nutter et al. | Design of novel deep learning models for real-time human activity recognition with mobile phones | |
| Biswas et al. | Hybrid expert system using case based reasoning and neural network for classification | |
| CN114333049A (en) | Pig attack behavior identification method, pig attack behavior identification system, computer equipment and storage medium | |
| Merenda et al. | A novel fitness tracker using edge machine learning | |
| CN117154256A (en) | Electrochemical repair method for lithium battery | |
| CN113408721A (en) | Neural network structure searching method, apparatus, computer device and storage medium | |
| Ishwarya et al. | Performance-enhanced real-time lifestyle tracking model based on human activity recognition (PERT-HAR) model through smartphones | |
| Gill et al. | Human Action Detection using EfficientNetB3 Model | |
| Sree et al. | Optimized conversion of categorical and numerical features in machine learning models | |
| Jiang et al. | Unsupervised deep learning for data-driven reliability and risk analysis of engineered systems | |
| Thulare et al. | An empirical analysis and application of the expectation-maximization and matrix completion algorithms for varying degrees of missing data | |
| ALSHARİ et al. | Machine Learning Model to Diagnose Diabetes Type 2 Based on Health Behavior | |
| Sarakon et al. | An individual model for human activity recognition using transfer deep learning | |
| Latha et al. | Human action recognition using deep learning methods (CNN-LSTM) without sensors | |
| Khan et al. | Suspicious activities recognition in video sequences using DarkNet-NasNet optimal deep features | |
| Sserwadda et al. | Topological Similarity and Centrality Driven Hybrid Deep Learning for Temporal Link Prediction | |
| Islamadina et al. | Performance of deep learning benchmark models on thermal imagery of pain through facial expressions | |
| Praveen et al. | An In-depth Exploration of ResNet-50 for Complex Emotion Recognition to Unraveling Emotional States | |
| Tian et al. | Learning semantic concepts from noisy media collection for automatic image annotation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |