CN111597991A - Rehabilitation detection method based on channel state information and BilSTM-Attention - Google Patents
Rehabilitation detection method based on channel state information and BilSTM-Attention Download PDFInfo
- Publication number
- CN111597991A CN111597991A CN202010412872.4A CN202010412872A CN111597991A CN 111597991 A CN111597991 A CN 111597991A CN 202010412872 A CN202010412872 A CN 202010412872A CN 111597991 A CN111597991 A CN 111597991A
- Authority
- CN
- China
- Prior art keywords
- csi
- bilstm
- layer
- rehabilitation
- attention
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 28
- 230000009471 action Effects 0.000 claims abstract description 54
- 238000013528 artificial neural network Methods 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 22
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 238000001914 filtration Methods 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 11
- 238000000513 principal component analysis Methods 0.000 claims abstract description 9
- 238000007781 pre-processing Methods 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 25
- 239000013598 vector Substances 0.000 claims description 18
- 230000007246 mechanism Effects 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 11
- 238000000034 method Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 6
- 230000009467 reduction Effects 0.000 claims description 6
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000003213 activating effect Effects 0.000 claims description 3
- 239000000969 carrier Substances 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 claims description 3
- 238000013136 deep learning model Methods 0.000 claims description 3
- 238000009499 grossing Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 description 6
- 238000000605 extraction Methods 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000004880 explosion Methods 0.000 description 3
- 208000013409 limited attention Diseases 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 210000000323 shoulder joint Anatomy 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 210000004556 brain Anatomy 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000001934 delay Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 210000003414 extremity Anatomy 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000474 nursing effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 210000001364 upper extremity Anatomy 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- 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
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
- G06V40/23—Recognition of whole body movements, e.g. for sport training
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Social Psychology (AREA)
- Psychiatry (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Pathology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a rehabilitation detection method based on channel state information and BilSTM-Attention, which comprises the following implementation steps: collecting CSI data of rehabilitation actions in an indoor environment, and extracting amplitude information; preprocessing steps such as low-pass filtering, normalization and principal component analysis are carried out on the CSI data; segmenting the preprocessed signals, detecting the starting point and the ending point of the action, and dividing segmented data segments into a training set and a test set; the training set is input into a BiLSTM-Attention-based deep neural network for motion recognition model training to obtain a rehabilitation motion recognition model, and the model can be used for classifying the collected CSI test set data to achieve the purposes of rehabilitation motion recognition and grading the rehabilitation degree. The invention adopts a deep neural network based on BilSTM-Attention to automatically learn and select characteristics, thereby realizing the identification of ten rehabilitation degrees of three different actions.
Description
Technical Field
The invention relates to the technical field of human body perception and behavior recognition, in particular to a rehabilitation detection method based on channel state information and BilSTM-Attention, which is used for solving the problems of extracting human body behavior characteristics by utilizing WiFi signals in an indoor environment, realizing rehabilitation action recognition and grading rehabilitation degree.
Background
With the rapid development of indoor wireless network application technology, the widespread use of wireless access equipment and the gradual maturity of artificial intelligence technology, the human body perception and behavior recognition technology based on the WiFi signal becomes an important topic, provides convenience for a considerable number of people, and has very important research value and significance in numerous fields such as human body detection, human-computer interaction, medical monitoring, indoor positioning, safety monitoring and the like.
Rehabilitation detection in clinical nursing work is an important work, the difference between the obstacle degree and the normal standard of a patient can be analyzed by scoring and evaluating the limb rehabilitation degree, a basis is provided for formulating a rehabilitation treatment scheme, and an objective index is provided for judging the rehabilitation treatment effect. However, most of the traditional rehabilitation detection methods are evaluated by doctors through protractors or visual observation and rely on manual monitoring. The method is greatly influenced by the subjectivity of doctors, and has the problems of human error, low efficiency and the like. Therefore, a rehabilitation detection method based on channel state information and a BilSTM-Attention neural network is developed, and the robustness of a rehabilitation detection system is enhanced.
The long-short term memory network (LSTM) is a variant of the Recurrent Neural Network (RNN), is suitable for processing and predicting important events with relatively long intervals and delays in a time sequence, effectively solves the problems of gradient disappearance, gradient explosion, long-term dependence and the like of the traditional recurrent neural network, and improves the learning capacity and classification accuracy of the neural network. During LSTM training, the influence of the input information before the mutual coordination and adjustment of the three logic gates on the current input information is realized, the state of the current memory module is updated, and the correlation of the information on the time sequence is fully considered, so that the method is very suitable for processing time sequence type data. And BilSTM is formed by combining forward LSTM and backward LSTM, and CSI measurement values in both the past direction and the future direction can be considered.
The Attention mechanism is the core idea that limited Attention resources are concentrated on key information points in a large amount of information, so that a large amount of calculation and low efficiency caused by average force are avoided. The Attention mechanism sequentially carries out similarity calculation on the query data and the input data through a similarity calculation function to obtain a similarity score. And then, carrying out weighted summation on the query data and the corresponding similarity scores to obtain a final key information point.
Disclosure of Invention
The invention aims to provide a rehabilitation detection method based on channel state information and BilSTM-Attention, which solves the problems of human error and low efficiency in the prior art.
In order to achieve the purpose, the invention adopts the technical scheme that:
a rehabilitation detection method based on channel state information and BilSTM-Attention specifically comprises the following steps:
step 1: collecting CSI data of rehabilitation actions in an indoor environment, and extracting amplitude information;
step 2: preprocessing the CSI data acquired in the step 1;
and step 3: acquiring key information of the action from the signal preprocessed in the step 2, detecting the starting point and the ending point of the action, and dividing the segmented data segment into a training set and a test set;
and 4, step 4: inputting the training set into a BiLSTM-Attention-based deep neural network for motion recognition model training to obtain a rehabilitation motion recognition model, and classifying the collected CSI test set data by adopting the model to achieve the purposes of rehabilitation motion recognition and grading the rehabilitation degree.
Further, in step 1, CSI data of the rehabilitation action is collected in the indoor environment, and amplitude information is extracted, specifically:
the WiFi signals of different degrees of rehabilitation actions are collected by a PC (personal computer) and a router which are provided with an Intel5300 wireless network card, and a three-dimensional matrix with a CSI matrix of 1 × 3 × 30 is extracted, wherein 1 × 3 represents a transmitting antenna and three receiving antennas, the three-dimensional matrix comprises 1 × 3-3 channels, 30 represents that each channel has 30 subcarriers, and 1 × 3 × 30-90 CSI streams are shared; the CSI is channel information of a physical layer, the performance of a wireless channel is reflected, and the amplitude of the CSI is obtained after the absolute value of the CSI is taken.
Further, in step 2, the CSI data acquired in step 1 is subjected to preprocessing steps such as low-pass filtering, normalization, and principal component analysis, specifically:
step 2-1: smoothing the original data extracted in the step 1 by a Hampel filter, filtering abnormal values and eliminating noise interference; then, filtering most high-frequency components in the CSI amplitude by using a Butterworth low-pass filter;
step 2-2: using min-max to perform standardization processing on the CSI subcarrier amplitude sequence denoised in the step 2-1, namelyWherein xnewRepresenting a new feature, x, after normalizationoldRepresenting old features before normalization, xmaxAnd xminRespectively representing the maximum value and the minimum value of all samples before the feature is processed;
step 2-3: tracking the correlation of the sample after min-max standardization through principal component analysis, performing orthogonal transformation by means of a characteristic vector of an original data covariance matrix, converting a random vector related to an original component into a new random vector unrelated to the component, converting multiple indexes into a few comprehensive indexes, namely principal components, determining the principal components to be reserved, and discarding other components, thereby realizing the dimensionality reduction of data; a first principal component is selected.
Further, in step 3, the specific steps are as follows:
step 3-1: the start of the action is judged by a dynamic threshold algorithm, and firstly, the calculation is carried outMean absolute deviation D of amplitude of jth sliding window CSIj: Wherein a isi(k) Is the amplitude of the sub-carrier i of the data packet k, S is the index set of all data packets of the sliding window, P is the total number of sub-carriers, and ω is the length of the sliding window;
step 3-2: updating the noise level, N, using an exponential moving average algorithmj=(1-αn)Nj-1+αn×DjCoefficient αnIs set to 0.15;
step 3-3: d obtained in the step 3-1 and the step 3-2jAnd NjBy comparison, if in a sliding window, DjGreater than noise level NjFour times, the starting point of the action will be detected; the end point of the operation is detected in the same manner.
Further, in step 4, a deep learning model based on the BilSTM is constructed by using a deep learning tool, the CSI training samples of the actions segmented in the step 3 are input, model training is carried out on the action recognition problem, an Attention mechanism is introduced, key information points are further focused in data characteristics, and finally, a detection result is obtained by activating function classification actions to obtain an action recognition model; and then, classifying the CSI test sample input model to obtain an action identification classification prediction result.
Further, the model architecture established in step 4 is composed of 5 parts: the device comprises a CSI signal input layer, a BilSTM layer, an attention mechanism layer, a leveling layer and a Softmax classification layer;
the CSI signal input layer receives an input CSI subcarrier amplitude sequence;
the BilSTM layer comprises a forward layer and a backward layer, and when the t-th sequence data is processed, the hidden states of the forward LSTM layer and the backward LSTM layer are respectively expressed asAndcomplete hidden state h of BilSTMtEqual to concatenation of hidden states of the forward and backward layers, i.e.Outputting the features learned from BilSTM to the next layer;
the attention mechanism layer is designed as a softmax regression layer, and uses the features learned from the BilSTM as input to obtain an attention matrix representing the importance of the features and the sequence data; the formula adopted when the power mechanism layer performs data processing is as follows:wherein h istIs the output vector of the upper BilSTM neural network layer, W is the model coefficient, b is the offset;
the leveling layer combines the learned characteristics with the attention moment matrix by using element-by-element multiplication to obtain an attention-based characteristic matrix, and then levels the characteristic matrix into a characteristic vector through the leveling layer;
and the Softmax classification layer finally identifies different actions by receiving the characteristic vectors obtained after the dimensionality reduction of the leveling layer.
The invention has the beneficial effects that:
(1) the CSI signal collected by the present invention is a new sampling index obtained by Orthogonal Frequency Division Multiplexing (OFDM) decoding, and compared with the conventional Received Signal Strength Indication (RSSI), the CSI signal can distinguish multipath components to a certain extent. Meanwhile, the CSI contains the amplitude and phase information of each subcarrier, so that richer time-frequency domain information can be provided.
(2) The BiLSTM neural network adopted by the invention increases the consideration of the CSI measurement value in the future direction on the basis of the LSTM neural network, and effectively solves the problems of gradient disappearance, gradient explosion, long-term dependence and the like of the traditional cyclic neural network. Meanwhile, compared with the traditional feature extraction method, the BilSTM can automatically learn and select features, and manual extraction is avoided. The BilSTM is combined with an attention mechanism, so that the learning ability of the neural network and the classification precision of rehabilitation actions are improved more effectively.
(3) The attention mechanism adopted by the invention has the characteristic of simulating the attention of the human brain, the limited attention resources are concentrated on important contents in a large amount of information, and less attention is allocated to other contents, so that a large amount of calculation and low efficiency caused by average force are avoided. Therefore, the weight distribution can be carried out on the features learned from the BilSTM neural network, the selective analysis can be carried out according to the importance degree of the feature matrix and the sequence data, and the rehabilitation action detection is more deep and accurate.
Drawings
Fig. 1 is a schematic flow chart of a rehabilitation detection method according to an embodiment of the invention.
FIG. 2 is a neural network architecture based on BilSTM-Attention in an embodiment of the present invention.
Fig. 3 is a table for scoring the degree of shoulder joint rehabilitation in an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
A rehabilitation detection method based on channel state information and BilSTM-Attention comprises the following implementation steps: collecting CSI data of rehabilitation actions in an indoor environment, and extracting amplitude information; preprocessing steps such as low-pass filtering, normalization and principal component analysis are carried out on the CSI data; segmenting the preprocessed signals, detecting the starting point and the ending point of the action, and dividing segmented data segments into a training set and a test set; the training set is input into a BiLSTM-Attention-based deep neural network for motion recognition model training to obtain a rehabilitation motion recognition model, and the model can be used for classifying the collected CSI test set data to achieve the purposes of rehabilitation motion recognition and grading the rehabilitation degree. The invention adopts a deep neural network based on BilSTM-Attention to automatically learn and select characteristics, thereby realizing the identification of ten rehabilitation degrees of three different actions.
A rehabilitation detection method based on channel state information and BilSTM-Attention is disclosed, the flow is shown in figure 1, and the method specifically comprises the following steps:
step 1: collecting CSI data of rehabilitation actions in an indoor environment, and extracting amplitude information, wherein the method specifically comprises the following steps: WiFi signals of different degrees of rehabilitation actions are collected by using a PC and a router provided with an Intel5300 wireless network card, and a three-dimensional matrix with a CSI matrix of 1 × 3 × 30 is extracted, where 1 × 3 denotes a transmitting antenna and three receiving antennas, so that 1 × 3 is 3 channels, 30 denotes 30 subcarriers in each channel, and 1 × 3 × 30 is 90 CSI streams in total. The CSI is channel information of a physical layer, the performance of a wireless channel is reflected, and the amplitude of the CSI is obtained after the absolute value of the CSI is taken.
Step 2: the method comprises the following steps of preprocessing the CSI data acquired in the step 1, such as low-pass filtering, normalization, principal component analysis and the like, and specifically comprises the following steps:
step 2-1: and (3) smoothing the original data extracted in the step (1) by a Hampel filter, filtering abnormal values and eliminating noise interference. Then, filtering most high-frequency components in the CSI amplitude by using Butterworth low-pass filter (Butterworth low-pass filter);
step 2-2: and (4) carrying out min-max standardization processing on the CSI subcarrier amplitude sequence subjected to denoising in the step 2-1. Namely, it isWherein xnewRepresenting a new feature, x, after normalizationoldRepresenting old features before normalization, xmaxAnd xminRespectively representing the maximum and minimum values of all samples before the feature is processed.
Step 2-3: the method comprises the steps of tracking the correlation of a sample subjected to min-max standardization through Principal Component Analysis (PCA), performing orthogonal transformation by means of a characteristic vector of an original data covariance matrix, converting a random vector related to an original Component into a new random vector unrelated to the Component, converting multiple indexes into a few comprehensive indexes (namely Principal components), determining Principal components needing to be reserved, and discarding other components, thereby realizing the dimensionality reduction of data. The invention selects a first principal component.
And step 3: acquiring the key information of the action from the signal preprocessed in the step 2, detecting the starting point and the ending point of the action, and dividing the segmented data segment into a training set and a test set, wherein the method specifically comprises the following steps:
step 3-1: judging the start of action through a dynamic threshold algorithm, firstly calculating the average absolute deviation D of the amplitude values of the CSI of the jth sliding windowj: Wherein a isi(k) Is the amplitude of the sub-carrier i of packet k, S is the index set of all packets of the sliding window, P is the total number of sub-carriers, and ω is the length of the sliding window.
Step 3-2: updating the noise level, N, using an exponential moving average algorithmj=(1-αn)Nj-1+αn×DjIn the present invention, the coefficient αnIs set to 0.15.
Step 3-3 comparing D obtained in step 3-1 and step 3-2jAnd NjBy comparison, if in a sliding window, DjGreater than noise level NjFour times, then the starting point of the action will be detected. The end point of the operation is detected in the same manner.
And 4, step 4: inputting a training set into a BiLSTM-Attention-based deep neural network to perform action recognition model training to obtain a rehabilitation action recognition model, classifying collected CSI test set data by adopting the model to achieve rehabilitation action recognition and the purpose of scoring the rehabilitation degree, and specifically comprising the following steps of: and (3) constructing a deep learning model based on the BilSTM by using a deep learning tool, inputting the CSI training samples of the actions segmented in the step (3), performing model training on the action recognition problem, introducing an Attention mechanism, further focusing key information points in data characteristics, and finally obtaining a detection result by activating function classification actions to obtain the action recognition model. And then, classifying the CSI test sample input model to obtain an action identification classification prediction result.
The model architecture established in this step is shown in fig. 2.
The model consists of 5 parts: a CSI Signal input Layer (CSI Signal Layer), a BilSTM Layer, an Attention mechanism Layer (Attention Layer), a flattening Layer (Flatten Layer), and a Softmax classification Layer (SoftmaxClassification Layer), the functions of which are as follows.
CSI Signal input Layer (CSI Signal Layer): an input CSI subcarrier amplitude sequence is received.
BilsTM layer: the BilSTM layer includes a forward layer and a backward layer. When processing the t-th sequence data, the hidden states of the forward LSTM layer and the backward LSTM layer are respectively expressed asAndcomplete hidden state h of BilSTMtEqual to concatenation of hidden states of the forward and backward layers, i.e.The features learned from BilSTM are output to the next layer.
Attention Layer (Attention Layer): in the present invention, the attention model is designed as a softmax regression layer, and the features learned from BilSTM are used as input to obtain an attention matrix representing the importance of the features and sequence data. The formula adopted when the power mechanism layer performs data processing is as follows:wherein h istIs the output vector of the upper BilSTM neural network layer, W is the model coefficient, and b is the offset.
Planarization Layer (flat Layer): the attention-based feature matrix is obtained by combining the learned features with the attention moment matrix using element-by-element multiplication. The feature matrix is then flattened into feature vectors by a flattening layer.
Softmax Classification Layer (Softmax Classification Layer): and the Softmax classification Layer finally identifies different actions by receiving the characteristic vector obtained after the dimensionality reduction processing of the flattening Layer (Flatten Layer).
And 5: taking the shoulder joint rehabilitation test as an example, according to the 'Shangtian hemiplegic upper limb function evaluation method', the shoulder joint abduction action can effectively evaluate the rehabilitation degree, and the rehabilitation degree scoring table is shown in fig. 3. Therefore, after the rehabilitation action is identified based on the channel state information and the BilSTM-Attention, the rehabilitation degree of the patient can be accurately identified according to the rehabilitation degree scoring table.
According to the rehabilitation detection method based on the channel state information and the BilSTM-Attention, the collected CSI signals are new sampling indexes obtained by Orthogonal Frequency Division Multiplexing (OFDM) decoding. Compared to the conventional Received Signal Strength Indication (RSSI), the CSI can distinguish multipath components to some extent. Meanwhile, the CSI contains the amplitude and phase information of each subcarrier, so that richer time-frequency domain information can be provided.
According to the rehabilitation detection method based on the channel state information and the BilSTM-Attention, the BilSTM neural network is used on the basis of the LSTM neural network, consideration on future direction CSI measurement values is increased, and the problems of gradient disappearance, gradient explosion, long-term dependence and the like of the traditional circulating neural network are effectively solved. Meanwhile, compared with the traditional feature extraction method, the BilSTM can automatically learn and select features, and manual extraction is avoided. The BilSTM is combined with an attention mechanism, so that the learning ability of the neural network and the classification precision of rehabilitation actions are improved more effectively.
According to the rehabilitation detection method based on the channel state information and the BilSTM-Attention, the adopted Attention mechanism has the characteristic of simulating the Attention of the human brain, limited Attention resources are focused on important contents in a large amount of information, less Attention is distributed to other contents, and therefore large amount of calculation and low efficiency caused by average force are avoided. Therefore, the weight distribution can be carried out on the features learned from the BilSTM neural network, the selective analysis can be carried out according to the importance degree of the feature matrix and the sequence data, and the rehabilitation action detection is more accurate.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.
Claims (6)
1. A rehabilitation detection method based on channel state information and BilSTM-Attention is characterized in that: the method specifically comprises the following steps:
step 1: collecting CSI data of rehabilitation actions in an indoor environment, and extracting amplitude information;
step 2: preprocessing the CSI data acquired in the step 1;
and step 3: acquiring key information of the action from the signal preprocessed in the step 2, detecting the starting point and the ending point of the action, and dividing the segmented data segment into a training set and a test set;
and 4, step 4: inputting the training set into a BiLSTM-Attention-based deep neural network for motion recognition model training to obtain a rehabilitation motion recognition model, and classifying the collected CSI test set data by adopting the model to achieve the purposes of rehabilitation motion recognition and grading the rehabilitation degree.
2. The rehabilitation detection method based on channel state information and BilSTM-Attention according to claim 1, characterized in that: in step 1, collecting CSI data of rehabilitation action in an indoor environment, and extracting amplitude information, specifically:
the WiFi signals of different degrees of rehabilitation actions are collected by a PC (personal computer) and a router which are provided with an Intel5300 wireless network card, and a three-dimensional matrix with a CSI matrix of 1 × 3 × 30 is extracted, wherein 1 × 3 represents a transmitting antenna and three receiving antennas, the three-dimensional matrix comprises 1 × 3-3 channels, 30 represents that each channel has 30 subcarriers, and 1 × 3 × 30-90 CSI streams are shared; the CSI is channel information of a physical layer, the performance of a wireless channel is reflected, and the amplitude of the CSI is obtained after the absolute value of the CSI is taken.
3. The rehabilitation detection method based on channel state information and BilSTM-Attention according to claim 1, characterized in that: in step 2, preprocessing steps such as low-pass filtering, normalization and principal component analysis are performed on the CSI data acquired in step 1, specifically:
step 2-1: smoothing the original data extracted in the step 1 by a Hampel filter, filtering abnormal values and eliminating noise interference; then, filtering most high-frequency components in the CSI amplitude by using a Butterworth low-pass filter;
step 2-2: using min-max to perform standardization processing on the CSI subcarrier amplitude sequence denoised in the step 2-1, namelyWherein xnewRepresenting a new feature, x, after normalizationoldRepresenting old features before normalization, xmaxAnd xminRespectively representing the maximum value and the minimum value of all samples before the feature is processed;
step 2-3: tracking the correlation of the sample after min-max standardization through principal component analysis, performing orthogonal transformation by means of a characteristic vector of an original data covariance matrix, converting a random vector related to an original component into a new random vector unrelated to the component, converting multiple indexes into a few comprehensive indexes, namely principal components, determining the principal components to be reserved, and discarding other components, thereby realizing the dimensionality reduction of data; a first principal component is selected.
4. The rehabilitation detection method based on channel state information and BilSTM-Attention according to claim 1, characterized in that: in the step 3, the concrete steps are as follows:
step 3-1: judging the start of action through a dynamic threshold algorithm, firstly calculating the average absolute deviation D of the amplitude values of the CSI of the jth sliding windowj: Wherein a isi(k) Is the amplitude of the sub-carrier i of the data packet k, S is the index set of all data packets of the sliding window, P is the total number of sub-carriers, and ω is the length of the sliding window;
step 3-2: updating the noise level, N, using an exponential moving average algorithmj=(1-αn)Nj-1+αn×DjCoefficient αnIs set to 0.15;
step 3-3: d obtained in the step 3-1 and the step 3-2jAnd NjBy comparison, if in a sliding window, DjGreater than noise level NjFour times, the starting point of the action will be detected; the end point of the operation is detected in the same manner.
5. The rehabilitation detection method based on channel state information and BilSTM-Attention according to claim 1, characterized in that: in step 4, a deep learning model based on the BilSTM is constructed by using a deep learning tool, CSI training samples of the actions segmented in the step 3 are input, model training is carried out on the action recognition problem, an Attention mechanism is introduced, key information points are further focused in data characteristics, and finally, a detection result is obtained by activating function classification actions to obtain an action recognition model; and then, classifying the CSI test sample input model to obtain an action identification classification prediction result.
6. The rehabilitation detection method based on channel state information and BilSTM-Attention of claim 5, characterized in that: the model architecture established in step 4 consists of 5 parts: the device comprises a CSI signal input layer, a BilSTM layer, an attention mechanism layer, a leveling layer and a Softmax classification layer;
the CSI signal input layer receives an input CSI subcarrier amplitude sequence;
the BilSTM layer includes a forward layer and a backward layer, and when processing the t-th sequence data, the forward LS layer is usedThe hidden states of the TM layer and the backward LSTM layer are respectively expressed asAndcomplete hidden state h of BilSTMtEqual to concatenation of hidden states of the forward and backward layers, i.e.Outputting the features learned from BilSTM to the next layer;
the attention mechanism layer is designed as a softmax regression layer, and uses the features learned from the BilSTM as input to obtain an attention matrix representing the importance of the features and the sequence data; the formula adopted when the power mechanism layer performs data processing is as follows:wherein h istIs the output vector of the upper BilSTM neural network layer, W is the model coefficient, b is the offset;
the leveling layer combines the learned characteristics with the attention moment matrix by using element-by-element multiplication to obtain an attention-based characteristic matrix, and then levels the characteristic matrix into a characteristic vector through the leveling layer;
and the Softmax classification layer finally identifies different actions by receiving the characteristic vectors obtained after the dimensionality reduction of the leveling layer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010412872.4A CN111597991A (en) | 2020-05-15 | 2020-05-15 | Rehabilitation detection method based on channel state information and BilSTM-Attention |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010412872.4A CN111597991A (en) | 2020-05-15 | 2020-05-15 | Rehabilitation detection method based on channel state information and BilSTM-Attention |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111597991A true CN111597991A (en) | 2020-08-28 |
Family
ID=72182845
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010412872.4A Pending CN111597991A (en) | 2020-05-15 | 2020-05-15 | Rehabilitation detection method based on channel state information and BilSTM-Attention |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111597991A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112131972A (en) * | 2020-09-07 | 2020-12-25 | 重庆邮电大学 | Method for recognizing human body behaviors by using WiFi data based on attention mechanism |
CN112423265A (en) * | 2020-11-13 | 2021-02-26 | 武汉理工大学 | CSI-based dual-cycle neural network shipborne environment indoor positioning method |
CN113288172A (en) * | 2021-05-24 | 2021-08-24 | 山东师范大学 | Epilepsia electroencephalogram signal identification method and system |
CN113723221A (en) * | 2021-08-11 | 2021-11-30 | 西安交通大学 | Indoor behavior real-time identification method and system based on WiFi channel state information |
CN113935373A (en) * | 2021-10-11 | 2022-01-14 | 南京邮电大学 | Human body action recognition method based on phase information and signal intensity |
CN114222202A (en) * | 2021-11-22 | 2022-03-22 | 上海数川数据科技有限公司 | Environment self-adaptive activity detection method and system based on WiFi CSI |
CN116304915A (en) * | 2023-05-16 | 2023-06-23 | 山东科技大学 | WiFi-based contactless action recognition method, system and laboratory device |
CN117951614A (en) * | 2024-03-21 | 2024-04-30 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Object activity recognition method and device based on channel state information |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109635837A (en) * | 2018-11-10 | 2019-04-16 | 天津大学 | A kind of carefree fall detection system of scene based on commercial wireless Wi-Fi |
CN109846472A (en) * | 2019-01-30 | 2019-06-07 | 郑州大学 | Beat classification method based on BiLSTM-Attention deep neural network |
-
2020
- 2020-05-15 CN CN202010412872.4A patent/CN111597991A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109635837A (en) * | 2018-11-10 | 2019-04-16 | 天津大学 | A kind of carefree fall detection system of scene based on commercial wireless Wi-Fi |
CN109846472A (en) * | 2019-01-30 | 2019-06-07 | 郑州大学 | Beat classification method based on BiLSTM-Attention deep neural network |
Non-Patent Citations (1)
Title |
---|
BIYUN SHENG等: "Deep Spatial–Temporal Model Based Cross-Scene Action Recognition Using Commodity WiFi", 《IEEE INTERNET OF THINGS JOURNAL》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112131972A (en) * | 2020-09-07 | 2020-12-25 | 重庆邮电大学 | Method for recognizing human body behaviors by using WiFi data based on attention mechanism |
CN112423265A (en) * | 2020-11-13 | 2021-02-26 | 武汉理工大学 | CSI-based dual-cycle neural network shipborne environment indoor positioning method |
CN113288172A (en) * | 2021-05-24 | 2021-08-24 | 山东师范大学 | Epilepsia electroencephalogram signal identification method and system |
CN113723221A (en) * | 2021-08-11 | 2021-11-30 | 西安交通大学 | Indoor behavior real-time identification method and system based on WiFi channel state information |
CN113723221B (en) * | 2021-08-11 | 2023-09-08 | 西安交通大学 | Indoor behavior real-time identification method and system based on WiFi channel state information |
CN113935373A (en) * | 2021-10-11 | 2022-01-14 | 南京邮电大学 | Human body action recognition method based on phase information and signal intensity |
CN114222202A (en) * | 2021-11-22 | 2022-03-22 | 上海数川数据科技有限公司 | Environment self-adaptive activity detection method and system based on WiFi CSI |
CN114222202B (en) * | 2021-11-22 | 2023-09-22 | 上海数川数据科技有限公司 | Environment self-adaptive activity detection method and system based on WiFi CSI |
CN116304915A (en) * | 2023-05-16 | 2023-06-23 | 山东科技大学 | WiFi-based contactless action recognition method, system and laboratory device |
CN116304915B (en) * | 2023-05-16 | 2023-08-29 | 山东科技大学 | WiFi-based contactless action recognition method, system and laboratory device |
CN117951614A (en) * | 2024-03-21 | 2024-04-30 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Object activity recognition method and device based on channel state information |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111597991A (en) | Rehabilitation detection method based on channel state information and BilSTM-Attention | |
Zhang et al. | Human activity recognition based on motion sensor using u-net | |
CN108629380B (en) | Cross-scene wireless signal sensing method based on transfer learning | |
CN110309861B (en) | Multi-modal human activity recognition method based on generation of confrontation network | |
CN106658590B (en) | Design and implementation of multi-person indoor environment state monitoring system based on WiFi channel state information | |
CN111698258B (en) | WiFi-based environmental intrusion detection method and system | |
CN114564982A (en) | Automatic identification method for radar signal modulation type | |
CN116343284A (en) | Attention mechanism-based multi-feature outdoor environment emotion recognition method | |
CN111901028B (en) | Human body behavior identification method based on CSI (channel State information) on multiple antennas | |
CN114330572A (en) | Anomaly detection method and system based on contrast learning and computer storage medium | |
CN114048773A (en) | Behavior identification method and system based on transfer learning and WiFi | |
CN111310719B (en) | Unknown radiation source individual identification and detection method | |
CN114781463A (en) | Cross-scene robust indoor tumble wireless detection method and related equipment | |
CN114626419B (en) | Action recognition method based on channel state information in WIFI and improved convolutional neural network | |
CN115392333A (en) | Equipment fault diagnosis method based on improved end-to-end ResNet-BilSTM dual-channel model | |
CN113935373A (en) | Human body action recognition method based on phase information and signal intensity | |
CN116701871A (en) | Rolling bearing composite fault feature selection method based on causal feature weighting network | |
CN116956222A (en) | Multi-complexity behavior recognition system and method based on self-adaptive feature extraction | |
Tang et al. | Specific emitter identification for IoT devices based on deep residual shrinkage networks | |
Jakkala et al. | Deep CSI learning for gait biometric sensing and recognition | |
CN117606801A (en) | Cross-domain bearing fault diagnosis method based on multi-characterization self-adaptive network | |
CN114757237B (en) | Speed-independent gait recognition method based on WiFi signals | |
CN115204238B (en) | PPG signal identity recognition method for wearable equipment and wearable equipment | |
CN111053552A (en) | QRS wave detection method based on deep learning | |
CN116386847A (en) | DB-LSTM neural network-based intelligent prediction algorithm for mobile medical Internet of things spectrum |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200828 |
|
RJ01 | Rejection of invention patent application after publication |