CN116821790A - Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy - Google Patents
Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy Download PDFInfo
- Publication number
- CN116821790A CN116821790A CN202310665156.0A CN202310665156A CN116821790A CN 116821790 A CN116821790 A CN 116821790A CN 202310665156 A CN202310665156 A CN 202310665156A CN 116821790 A CN116821790 A CN 116821790A
- Authority
- CN
- China
- Prior art keywords
- sample
- training
- gait
- data set
- integrated learning
- 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
- 238000000034 method Methods 0.000 title claims abstract description 38
- 230000005021 gait Effects 0.000 title claims description 28
- 230000008014 freezing Effects 0.000 title description 7
- 238000007710 freezing Methods 0.000 title description 7
- 238000012549 training Methods 0.000 claims abstract description 54
- 206010056242 Parkinsonian gait Diseases 0.000 claims abstract description 36
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 26
- 230000003044 adaptive effect Effects 0.000 claims abstract description 7
- 238000007781 pre-processing Methods 0.000 claims abstract description 6
- 238000005070 sampling Methods 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000002759 z-score normalization Methods 0.000 claims description 4
- 230000011218 segmentation Effects 0.000 claims description 3
- 230000007717 exclusion Effects 0.000 claims description 2
- 230000004044 response Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 4
- 238000013135 deep learning Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 208000018737 Parkinson disease Diseases 0.000 description 5
- 208000027089 Parkinsonian disease Diseases 0.000 description 4
- 206010034010 Parkinsonism Diseases 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000000875 corresponding effect Effects 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 208000024891 symptom Diseases 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 206010006100 Bradykinesia Diseases 0.000 description 1
- 208000006083 Hypokinesia Diseases 0.000 description 1
- 206010052904 Musculoskeletal stiffness Diseases 0.000 description 1
- 206010071390 Resting tremor Diseases 0.000 description 1
- 206010044565 Tremor Diseases 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006735 deficit Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000003340 mental effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004770 neurodegeneration Effects 0.000 description 1
- 208000015122 neurodegenerative disease Diseases 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000001144 postural effect Effects 0.000 description 1
- 230000003449 preventive effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/103—Detecting, measuring or recording devices for testing the shape, pattern, colour, size or movement of the body or parts thereof, for diagnostic purposes
- A61B5/11—Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb
- A61B5/112—Gait analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7271—Specific aspects of physiological measurement analysis
- A61B5/7275—Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
-
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Surgery (AREA)
- Veterinary Medicine (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Software Systems (AREA)
- Physiology (AREA)
- Animal Behavior & Ethology (AREA)
- Heart & Thoracic Surgery (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Computing Systems (AREA)
- Psychiatry (AREA)
- Computational Linguistics (AREA)
- Dentistry (AREA)
- Fuzzy Systems (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a frozen gait prediction method based on an adaptive enhanced integrated learning strategy, which comprises the steps of obtaining an original data set, and preprocessing to obtain the data set; dividing the preprocessed data set; the processed training samples are endowed with the same preset sample weight, and are input into a multi-scale convolutional neural network constructed based on a self-adaptive enhancement algorithm, a plurality of weak classifiers are obtained through multiple sample training and sample weight adjustment, and the weak classifiers are combined to finally obtain a strong classifier for frozen gait prediction. According to the method, a plurality of weak classifiers are trained by constructing a multi-scale convolutional neural network, and the weak classifiers are combined into one strong classifier by utilizing an integrated learning strategy, so that a frozen gait prediction method is established. Therefore, the problem that the prediction model has poor prediction effect on few categories is solved, and the prediction accuracy of the model on the few categories is improved.
Description
Technical Field
The application relates to the technical field of health state index prediction, in particular to a frozen gait prediction method based on an adaptive enhanced integrated learning strategy.
Background
Parkinson's Disease (PD) is a typical neurodegenerative disease, which occurs mainly in the elderly. As the aged population grows, the incidence of PD steadily increases over time. Motor symptoms characteristic of parkinson patients include resting tremor, bradykinesia, frozen gait (FoG), muscle stiffness and postural balance impairment. Frozen gait is a severe motor symptom of parkinsonism, commonly seen in late parkinsonism, parkinsonism and parkinsonism. The main manifestation of a frozen gait is that a patient may want to move forward but not move their legs, feel their feet standing on the ground, and this situation often occurs when the patient starts, turns around or passes through a narrow road, especially when the patient is under mental stress. When a frozen gait occurs, the patient's feet cannot move, but the body still has forward inertia, which can lead to frequent falls, injuries, and even more serious consequences.
The detection of frozen gait is that an event can be detected after the occurrence of the frozen gait, and real-time detection cannot be achieved due to the delay of calculation time. If the occurrence of frozen gait can be predicted, preventive measures or external stimulus can be taken to prevent the parkinsonian patient from freezing in gait to some extent, so that the quality of daily life of the patient can be better improved. Because gait signals have the characteristic of class imbalance, the common solution is based on the processing mode of undersampling or oversampling of data samples, and although the number difference among sample classes can be reduced, the problems of sample number reduction, sample repetition or sample authenticity reduction and the like can be caused, and the influence caused by class imbalance can not be completely overcome, so that the prediction accuracy of few classes is low, and the prediction effect of a model is influenced.
The prior art scheme is as follows: data enhancement [ short C, KHOSHGOFTAAR T M.A survey on Image Data Augmentation for Deep Learning [ J ]. Journal of Big Data,2019,6 (1) ] is a series of technical means and methods performed to increase the diversity and number of data samples. One is a sampling method. The sampling method is to change an unbalanced data set into a balanced data set by processing a training set, and is mainly divided into undersampling and oversampling. Sample sampling by reducing the number of most classes of samples or increasing the number of few classes of samples, the number gap between classes is reduced, and common methods are random undersampling, random oversampling, and SMOTE oversampling [ cha wla N V, BOWYER K W, HALL LO, et al SMOTE: synthetic Minority Over-sampling Technique [ J ]. The Journal of Artificial Intelligence Research,2002,16 (0): 321-57 ]. The sampling method aims at balancing the proportion of various types of samples from the data layer, and can improve the prediction accuracy of a few types to a certain extent. The other is image transformation. By performing a series of transformations and extensions on the raw data, such as gaussian noise, scale transformation, translational transformation, flipping, etc., more data can be obtained while also preventing model overfitting.
The traditional manual features refer to features which are designed manually in a machine learning model and can maximally distinguish different objects or scenes, and the extraction basis is based on priori knowledge and experience, so that people can learn through observation. The method [ ORPHANDOU N K, HUSSAIN A, KEIGHT R, et al, predicting Freezing of Gait in Parkinsons Disease Patients using Machine Learning; proceedings of the IEEE Congress on Evolutionary Computation (IEEE CEC) as part of the IEEE World Congress on Computational Intelligence (IEEE WCCI), rio de Janeiro, BRAZIL, F Jul 08-13,2018[ C ]. Ieee: NEW YORK,2018 ] is one of the most fundamental ways to predict gait freezing, with the aim of creating robust and well-differentiated feature descriptors. Traditional manual features represent expert knowledge of physical movement symptoms and data-driven features extracted from original signals and time-frequency characterization thereof, and have strong interpretability and easy debugging.
The deep learning has the advantages over traditional machine learning in that features in data can be automatically extracted and learned, no human need for feature calculation and extraction is required, and non-linear relationships and large-scale data can be processed. Secondly, the deep learning method can acquire the features with stronger semantic information by using a deep convolutional neural network. The CNN-LSTM algorithm has the advantages of multidimensional input, advanced feature extraction, long-term dependency capture and the like, and is used in a plurality of fields such as video analysis, human behavior analysis, medical image analysis and the like. CNN-LSTM algorithm [ ZHANG Y, GU D.A deep convolutional-recurrent neural network for freezing of gait detection in patients with Parkinson's disease; proceedings of the 2019 12th International Congress on Image and Signal Processing,BioMedical Engineering and Informatics (CISP-BMEI), F,2019[ C ]. IEEE ] uses a convolution layer to automatically extract features, form a feature map, uses a cyclic LSTM layer to model the time dynamic process of the feature map, and takes into account both time information and depth features.
The prior art has the disadvantages that the data is enhanced: data enhancement can increase noise and inaccuracy, resulting in bias in model training; the enhanced data may be different from the real data, and some samples which do not provide effective information are generated, so that the training complexity of the model is increased, and the generalization capability of the model is influenced; computing resources and training time may be increased.
The method based on the traditional manual characteristics comprises the following steps: for the calculation and selection of the traditional manual characteristics, the optimal characteristics are not easy to find, and when a data set is changed, the characteristics may need to be adjusted and updated, a new task cannot be self-adapted, and a great deal of time and labor cost are required; it is difficult to take all factors and conditions into account when creating manual features, and to process high-dimensional data, and feature scope may limit the model's performance and generalization ability.
Deep learning-based method: first, deep learning requires a large amount of data to train, otherwise it may result in insufficient generalization ability of the model, i.e., poor performance on new data; secondly, the training time is long because it requires repeated iterative training of a large amount of data, and if the model is too complex, the training time is longer; the interpretability is relatively poor, and the deep learning trains multiple layers of weights and deviations, so that the internal neurons are difficult to effectively interpret; the adaptability to sudden events and data changes is weak.
Disclosure of Invention
The application aims to overcome the defects in the prior art, and in order to achieve the purposes, the frozen gait prediction method based on the self-adaptive enhanced integrated learning strategy is adopted to solve the problems in the background art.
A frozen gait prediction method based on an adaptive enhanced integrated learning strategy comprises the following steps:
step S1, acquiring an original data set of gait signals, and preprocessing to obtain a preprocessed data set of gait signals;
step S2, dividing the preprocessed data set into a training sample and a test sample according to a preset proportion based on layered sampling;
step S3, endowing the processed training samples with the same preset sample weight, inputting the training samples into a multi-scale convolutional neural network constructed based on a self-adaptive enhancement algorithm, and obtaining a first weak classifier and a classification result through multiple sample training;
step S4, according to the first weak classifier obtained by training and the classification result, adjusting the weight of the sample, retraining to obtain a new weak classifier, and repeating the steps S3 and S4 to obtain a plurality of weak classifiers;
and S5, combining the obtained weak classifiers based on an integrated learning algorithm to finally obtain a strong classifier for frozen gait prediction.
As a further aspect of the application: the specific steps in the step S1 include:
data exclusion: acquiring an original data set of gait signals and removing data of non-experimental parts;
and (3) standardization treatment: performing a patient-specific Z-score normalization process on each patient gait signal in the dataset;
and (3) signal segmentation: the 5 seconds before each onset of the sample class FoG is re-labeled as sample class pre-FoG, wherein the sample classes are Walk, foG, and pre-FoG, the gait signals of all patients are integrated, and the gait signals are divided into signal segments with the duration of 1 second.
As a further aspect of the application: the specific steps of dividing the data set into the training sample and the test sample according to the preset proportion in the step S2 are as follows:
the method comprises the steps of adopting hierarchical sampling, selecting samples of 80% of the preprocessed data set as training samples for model training, and selecting samples of 20% of the preprocessed data set as test samples for model testing.
As a further aspect of the application: the specific steps in the step S3 include:
obtaining a training sample divided according to a preset proportion after processing;
giving the same sample weight to each sample in the training samples, and inputting each sample with the sample weight into a multi-scale convolutional neural network training network model constructed based on an adaptive enhancement algorithm, wherein a backbone network of the multi-scale convolutional neural network is composed of residual networks formed by multi-scale convolutional modules;
then, a first weak classifier is obtained through multi-layer convolution;
and then according to the classification result of the previous weak classifier, the method is used for training the next weak classifier by improving the sample weight of a minority class sample which is easy to be misclassified in the data set.
As a further aspect of the application: the specific steps of the multi-scale convolutional neural network constructed based on the adaptive enhancement algorithm in the step S3 include:
inputting the preprocessed gait signals to a one-dimensional convolution layer with the convolution kernel size of 7, extracting signal characteristics, and sequentially inputting the extracted characteristics to a batch normalization layer and a pooling layer;
and finally, extracting signal characteristics by adopting a main network formed by a plurality of convolution modules, wherein a residual error network is formed by adopting a multi-scale convolution module.
As a further aspect of the application: the specific steps of extracting signal features by adopting a main network consisting of a plurality of convolution modules comprise:
the signal features are extracted by using one-dimensional convolution layer with the convolution kernel size of 1, then the signal features are respectively sent to the one-dimensional convolution layers with the convolution kernels of 1 and 3 to further extract the signal features with different sizes, and finally the signal features with different scales are fused.
As a further aspect of the application: the specific steps in the step S5 include:
the sample weight updating strategy formula based on the self-adaptive enhanced integrated learning algorithm is as follows:
wherein ,is the weight of the ith training sample used by the mth convolutional neural network, alpha is the learning rate, K is the total number of sample categories, gamma i Is the label vector corresponding to the ith training sample, P m (x i ) Is the output vector of the mth convolutional neural network in response to the ith training sample, and N is the total number of samples.
Compared with the prior art, the application has the following technical effects:
by adopting the technical scheme, the freezing gait prediction method is established by training a plurality of weak classifiers through a multi-scale convolutional neural network based on a self-adaptive enhancement algorithm and then combining the plurality of weak classifiers into one strong classifier by adopting an integrated learning strategy. Therefore, the problem that the frozen gait prediction model has poor prediction effect on few categories due to the fact that frozen gait signals have the characteristics of unbalanced categories is solved, weights are continuously adjusted based on an integrated learning network algorithm to learn gait characteristics better, and the prediction accuracy of the model on the few categories is improved.
Drawings
The following detailed description of specific embodiments of the application refers to the accompanying drawings, in which:
FIG. 1 is a schematic diagram illustrating steps of a frozen gait prediction method according to an embodiment of the disclosure;
FIG. 2 is a general flow diagram of a frozen gait prediction method in accordance with an embodiment of the disclosure;
FIG. 3 is a schematic diagram of an overall framework of a network model of a multi-scale convolutional neural network in accordance with an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of a multi-scale convolution module according to an embodiment of the present disclosure.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Referring to fig. 1 and 2, in an embodiment of the present application, a frozen gait prediction method based on an adaptively enhanced integrated learning strategy includes the following steps:
step S1, acquiring an original data set of gait signals, and preprocessing to obtain a preprocessed data set of gait signals, wherein the steps specifically comprise:
data preprocessing:
step S11, excluding non-experimental parts in the original data set:
in a specific embodiment, the application uses Daphnet data set, the original data marked as "0" in the data set is not part of the experiment, and the corresponding activity mode does not conform to the experiment protocol, so that the irrelevant data are excluded;
step S12, standardization processing:
performing a per-patient specific Z-score normalization (person-specific Z-score normalization) on each patient gait signal in the dataset;
step S13, signal segmentation:
in this example, the pre-FoG was re-labeled 5 seconds before each FoG episode, i.e., the sample class was: walk, foG, pre-FoG. The gait signals of all patients were then integrated and divided into signal segments of 1 second duration, with the last segment overlapping 50% of the next segment.
Step S2, dividing the preprocessed data set into a training sample and a test sample according to a preset proportion based on layered sampling:
in this embodiment, the data set after the data preprocessing is divided. Specifically, hierarchical sampling is adopted, samples of 80% of the preprocessed data set are selected as training samples for model training, and samples of 20% of the preprocessed data set are selected as test samples for model testing.
Step S3, endowing the processed training samples with the same preset sample weight, inputting the training samples into a multi-scale convolutional neural network constructed based on a self-adaptive enhancement algorithm, and obtaining a first weak classifier and a classification result through multiple sample training;
in this embodiment, a multi-scale convolutional neural network based on an adaptive enhancement algorithm is adopted, as shown in fig. 3, which is a schematic diagram of a network model overall framework of the multi-scale convolutional neural network, and the specific steps include:
step S31, obtaining training samples which are divided according to a preset proportion after processing;
step S32, giving the same sample weight to each sample in the training samples, and inputting each sample with the sample weight into a multi-scale convolutional neural network training network model constructed based on a self-adaptive enhancement algorithm, wherein a backbone network of the multi-scale convolutional neural network is composed of residual networks formed by multi-scale convolutional modules;
step S33, obtaining a first weak classifier through multi-layer convolution;
step S34, according to the classification result of the previous weak classifier, the sample weight of a few types of samples which are easy to be misclassified in the data set is improved to be used for training the process of the next weak classifier.
The multi-scale convolutional neural network constructed based on the self-adaptive enhancement algorithm comprises the following specific steps:
in the embodiment, firstly, the preprocessed gait signal is input into a one-dimensional convolution layer with the convolution kernel size of 7, signal characteristics are extracted, and then the extracted characteristics are sequentially input into a batch normalization layer (Batch Normalization, BN) and a Pooling layer (Pooling);
then, a main network composed of a plurality of convolution modules is adopted to extract signal characteristics, and residual network connection is introduced due to the problems of gradient disappearance, gradient explosion and the like possibly caused by the too deep convolution neural network.
In this embodiment, the network part uses a Multi-scale convolution module (Multi-scale convolution module) to form a residual network, firstly, one-dimensional convolution layer with a convolution kernel size of 1 is used to extract signal features, then the signal features are respectively sent to the one-dimensional convolution layers with convolution kernels of 1 and 3 to further extract signal features with different sizes, and finally, the signal features with different scales are fused, as shown in fig. 4, which is a schematic structural diagram of the Multi-scale convolution module.
Step S4, according to the first weak classifier obtained by training and the classification result, adjusting the weight of the sample, retraining to obtain a new weak classifier, and repeating the steps S3 and S4 to obtain a plurality of weak classifiers;
step S5, combining the obtained weak classifiers based on an ensemble learning algorithm to finally obtain a strong classifier for frozen gait prediction, wherein the specific steps comprise:
the sample weight updating strategy formula based on the self-adaptive enhanced integrated learning algorithm is as follows:
wherein ,is the weight of the ith training sample used by the mth convolutional neural network, alpha is the learning rate, K is the total number of sample categories, gamma i Is the label vector corresponding to the ith training sample, P m (x i ) The output vector of the mth convolutional neural network responding to the ith training sample is the nth convolutional neural network, and N is the total number of samples;
according to the above step strategy, we can train a number of different weak classifiers on the data set updating the sample weights. And obtaining a strong classifier from the trained weak classifier through an integration strategy, and predicting the frozen gait signals acquired in real time through the obtained strong classifier.
Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.
Claims (7)
1. The frozen gait prediction method based on the self-adaptive enhanced integrated learning strategy is characterized by comprising the following steps of:
step S1, acquiring an original data set of gait signals, and preprocessing to obtain a preprocessed data set of gait signals;
step S2, dividing the preprocessed data set into a training sample and a test sample according to a preset proportion based on layered sampling;
step S3, endowing the processed training samples with the same preset sample weight, inputting the training samples into a multi-scale convolutional neural network constructed based on a self-adaptive enhancement algorithm, and obtaining a first weak classifier and a classification result through multiple sample training;
step S4, according to the first weak classifier obtained by training and the classification result, adjusting the weight of the sample, retraining to obtain a new weak classifier, and repeating the steps S3 and S4 to obtain a plurality of weak classifiers;
and S5, combining the obtained weak classifiers based on an integrated learning algorithm to finally obtain a strong classifier for frozen gait prediction.
2. The frozen gait prediction method based on the adaptively enhanced integrated learning strategy according to claim 1, wherein the specific steps in the step S1 include:
data exclusion: acquiring an original data set of gait signals and removing data of non-experimental parts;
and (3) standardization treatment: performing a patient-specific Z-score normalization process on each patient gait signal in the dataset;
and (3) signal segmentation: the 5 seconds before each onset of the sample class FoG is re-labeled as sample class pre-FoG, wherein the sample classes are Walk, foG, and pre-FoG, the gait signals of all patients are integrated, and the gait signals are divided into signal segments with the duration of 1 second.
3. The frozen gait prediction method based on the self-adaptive enhanced integrated learning strategy according to claim 1, wherein the specific steps of dividing the data set into the training sample and the test sample according to the preset proportion in the step S2 are as follows:
the method comprises the steps of adopting hierarchical sampling, selecting samples of 80% of the preprocessed data set as training samples for model training, and selecting samples of 20% of the preprocessed data set as test samples for model testing.
4. The frozen gait prediction method based on the adaptively enhanced integrated learning strategy according to claim 1, wherein the specific steps in the step S3 include:
obtaining a training sample divided according to a preset proportion after processing;
giving the same sample weight to each sample in the training samples, and inputting each sample with the sample weight into a multi-scale convolutional neural network training network model constructed based on an adaptive enhancement algorithm, wherein a backbone network of the multi-scale convolutional neural network is composed of residual networks formed by multi-scale convolutional modules;
then, a first weak classifier is obtained through multi-layer convolution;
and then according to the classification result of the previous weak classifier, the method is used for training the next weak classifier by improving the sample weight of a minority class sample which is easy to be misclassified in the data set.
5. The frozen gait prediction method based on the self-adaptive enhancement integrated learning strategy according to claim 4, wherein the specific steps of constructing the multi-scale convolutional neural network based on the self-adaptive enhancement algorithm in the step S3 include:
inputting the preprocessed gait signals to a one-dimensional convolution layer with the convolution kernel size of 7, extracting signal characteristics, and sequentially inputting the extracted characteristics to a batch normalization layer and a pooling layer;
and finally, extracting signal characteristics by adopting a main network formed by a plurality of convolution modules, wherein a residual error network is formed by adopting a multi-scale convolution module.
6. The method for predicting frozen gait based on the adaptively enhanced integrated learning strategy of claim 5, wherein the specific step of extracting signal features by using a backbone network consisting of a plurality of convolution modules comprises:
the signal features are extracted by using one-dimensional convolution layer with the convolution kernel size of 1, then the signal features are respectively sent to the one-dimensional convolution layers with the convolution kernels of 1 and 3 to further extract the signal features with different sizes, and finally the signal features with different scales are fused.
7. The frozen gait prediction method based on the adaptively enhanced integrated learning strategy according to claim 1, wherein the specific steps in the step S5 include:
the sample weight updating strategy formula based on the self-adaptive enhanced integrated learning algorithm is as follows:
wherein ,is the weight of the ith training sample used by the mth convolutional neural network, alpha is the learning rate, K is the total number of sample categories, gamma i Is the label vector corresponding to the ith training sample, P m (x i ) Is the output vector of the mth convolutional neural network in response to the ith training sample, and N is the total number of samples.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310665156.0A CN116821790A (en) | 2023-06-05 | 2023-06-05 | Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310665156.0A CN116821790A (en) | 2023-06-05 | 2023-06-05 | Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116821790A true CN116821790A (en) | 2023-09-29 |
Family
ID=88117755
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310665156.0A Pending CN116821790A (en) | 2023-06-05 | 2023-06-05 | Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116821790A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117786682A (en) * | 2024-02-28 | 2024-03-29 | 厦门理工学院 | Physical challenge attack resisting method, device, equipment and medium based on enhanced framework |
-
2023
- 2023-06-05 CN CN202310665156.0A patent/CN116821790A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117786682A (en) * | 2024-02-28 | 2024-03-29 | 厦门理工学院 | Physical challenge attack resisting method, device, equipment and medium based on enhanced framework |
CN117786682B (en) * | 2024-02-28 | 2024-05-14 | 厦门理工学院 | Physical challenge attack resisting method, device, equipment and medium based on enhanced framework |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109345508B (en) | Bone age evaluation method based on two-stage neural network | |
Gupta et al. | Wearable sensors for evaluation over smart home using sequential minimization optimization-based random forest | |
CN111783534A (en) | Sleep staging method based on deep learning | |
Keivani et al. | Automated analysis of leaf shape, texture, and color features for plant classification. | |
CN111387975B (en) | Electroencephalogram signal identification method based on machine learning | |
CN116821790A (en) | Freezing gait prediction method based on self-adaptive enhanced integrated learning strategy | |
Yang et al. | Classification of diabetic retinopathy severity based on GCA attention mechanism | |
Mampitiya et al. | Efficient and low-cost skin cancer detection system implementation with a comparative study between traditional and CNN-based models | |
Ganeshbabu | Glaucoma image classification using discrete orthogonal stockwell transform | |
Simanjuntak et al. | Cataract Classification Based on Fundus Images Using Convolutional Neural Network | |
Sallam et al. | Diabetic retinopathy grading using resnet convolutional neural network | |
Bharathi et al. | Diabetes diagnostic method based on tongue image classification using machine learning algorithms | |
CN113705670A (en) | Brain image classification method and device based on magnetic resonance imaging and deep learning | |
Redhya et al. | Refining PD classification through ensemble bionic machine learning architecture with adaptive threshold based image denoising | |
Ali et al. | Blood vessel segmentation from color retinal images using K-means clustering and 2D gabor wavelet | |
Nurrahmadayeni et al. | Analysis of deep learning methods in diabetic retinopathy disease identification based on retinal fundus image | |
CN116361723A (en) | Bearing fault diagnosis and classification method based on multi-scale characteristics and attention | |
CN111860949B (en) | Prediction method and device based on time sequence image | |
Ebin et al. | A Variant Binary Classification Model for No-DR Mild-DR Detection using CLAHE Images With Transfer Learning | |
Ahmadieh et al. | Visual image reconstruction based on EEG signals using a generative adversarial and deep fuzzy neural network | |
Sengar et al. | An efficient artificial intelligence-based approach for diagnosis of media haze disease | |
Bian et al. | Automatic optic disc/cup segmentation and glaucoma classification and fovea localization | |
Mavaddati | Sparse structured principal component analysis and model learning for classification and quality detection of rice grains | |
Karan et al. | Deep Neural Network for Parkinson Disease Prediction Using SPECT Image | |
Anupama et al. | Performance analysis of learning algorithms for automated detection of glaucoma |
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 |