Parkinson's disease electroencephalogram signal identification method and system based on multi-source mapping

Abstract

The invention provides a Parkinson's disease electroencephalogram signal recognition method and system based on multi-source mapping, and relates to the technical field of electroencephalogram signal recognition. Comprising the steps of constructing a tested data set and labeling category labels; dividing the tested data set into a training set and a test set, training a plurality of sub-classifiers by using the training set, and obtaining a set classifier comprising the plurality of sub-classifiers and weight combinations thereof based on the test effect of each sub-classifier on the test set; mapping data in the test set to the training set through semi-supervised learning, determining a final mapping path, and mapping the acquired electroencephalogram signal to be tested to corresponding mapping end point data in the training set; and inputting the corresponding mapping end point data in the training set into a set classifier to obtain a predicted classification label. When the to-be-tested electroencephalogram signals are identified, the to-be-tested electroencephalogram signals are mapped by using the mapping path and then identified and classified by using the set classifier, so that the accuracy of electroencephalogram classification of the Parkinson's disease can be effectively improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electroencephalogram (EEG) signal recognition, and in particular relates to a method and system for Parkinson's disease EEG signal recognition based on multi-source mapping. Background Art

[0002] Parkinson's disease (PD) affects 7-10 million people worldwide. EEG recognition and diagnosis of Parkinson's disease is a low-cost, non-invasive technology. Therefore, EEG recognition and diagnosis of Parkinson's disease patients through machine learning is currently the most effective and feasible method, which enables computers to identify the EEG characteristics of Parkinson's disease patients from EEG signals. In the past few decades, people have proposed a variety of classification methods and achieved good results.

[0003] However, due to individual differences in EEG features between patients, there will inevitably be obvious feature distribution mismatches between training EEG signals and test EEG signals. In this case, the classification accuracy of the classifier trained with the training set will decrease when encountering new subjects with different sample features. Therefore, a single classifier cannot further improve the classification effect of the model on various feature data and cannot ensure the effectiveness of classification.

[0004] Therefore, training a classifier that can effectively classify subjects with different characteristics is a challenging task. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a Parkinson's disease EEG signal recognition method and system based on multi-source mapping, which determines a set classifier based on the classification effect of multiple sub-classifiers on a test set, and determines a final mapping path based on the mapping of the test set to a training set. When identifying the EEG signal to be tested, it is first mapped using the mapping path, and then identified and classified using the set classifier, which can effectively improve the accuracy of Parkinson's disease EEG classification.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] The first aspect of the present invention provides a method for identifying Parkinson's disease electroencephalogram (EEG) signals based on multi-source mapping.

[0008] The method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping includes the following steps:

[0009] Obtain EEG signals from subjects with Parkinson's disease and healthy controls, construct a subject dataset and annotate category labels;

[0010] The test data set is divided into a training set and a test set, and multiple sub-classifiers are trained using the training set. Based on the test results of each sub-classifier on the test set, weights are assigned to the sub-classifiers to obtain a collective classifier containing multiple sub-classifiers and their weight combinations;

[0011] Map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and based on the final mapping path, map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set;

[0012] The corresponding mapping endpoint data in the training set is input into the ensemble classifier to obtain the predicted classification label, thus completing the recognition of the EEG signal to be tested.

[0013] The second aspect of the present invention provides a Parkinson's disease EEG signal recognition system based on multi-source mapping.

[0014] Parkinson's disease EEG signal recognition system based on multi-source mapping, including:

[0015] The dataset construction module is configured to: obtain EEG signals of subjects with Parkinson's disease and normal healthy controls, construct the subject dataset and annotate the category labels;

[0016] The ensemble classifier training module is configured to: divide the test data set into a training set and a test set, train multiple sub-classifiers using the training set, assign weights to the sub-classifiers based on the test effect of each sub-classifier on the test set, and obtain an ensemble classifier including multiple sub-classifiers and their weight combinations;

[0017] A mapping module is configured to: map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set based on the final mapping path;

[0018] The recognition module is configured to: input the corresponding mapping endpoint data in the training set into the set classifier, obtain the predicted classification label, and complete the recognition of the EEG signal to be tested.

[0019] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the method for identifying Parkinson's disease electroencephalogram signals based on multi-source mapping as described in the first aspect of the present invention.

[0020] The fourth aspect of the present invention provides an electronic device, comprising a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for identifying Parkinson's disease electroencephalogram signals based on multi-source mapping as described in the first aspect of the present invention are implemented.

[0021] One or more of the above technical solutions have the following beneficial effects:

[0022] The present invention provides a Parkinson's disease electroencephalogram (EEG) signal recognition method and system based on multi-source mapping. Different from the prior art which simply trains a single classifier, the present invention is based on a multi-source transfer learning method, trains a plurality of sub-classifiers, and then weightedly combines the sub-classifiers to construct an ensemble classifier, which is used for recognition prediction. In addition, the present invention maps test set data to training set data, changes the characteristics of the data, adapts the sub-classifiers to improve the classification effect of the classifier, and finally determines a mapping path. When recognizing the EEG signal to be tested, the final mapping path obtained above is first used to map it to the training set, and the corresponding mapping endpoint data in the training set is input into the ensemble classifier to obtain a predicted classification label, which can effectively improve the accuracy of Parkinson's disease EEG classification.

[0023] When constructing subjects, the present invention constructs the EEG signals of Parkinson's disease patients and normal healthy controls with similar EEG characteristics into one subject. The training set contains several different types of subjects and corresponding category labels to achieve the purpose of multi-source. The classifier learned by the subjects in the training set is used through multi-source transfer learning to obtain a set classifier, and the set classifier is used to classify the EEG of Parkinson's disease patients and non-patients.

[0024] When determining the mapping path for mapping the test set to the training set, the present invention first calculates a preliminarily determined specific mapping form, uses the preliminarily determined specific mapping form to map the test set to the training set, and iterates the above process based on the confidence level, corrects and updates the preliminarily determined specific mapping form until the final mapping path is obtained, and then uses the final mapping path to map the EEG signal to be tested, which can obtain a more accurate mapping result, thereby improving the accuracy of recognition and classification.

[0025] When iterating based on confidence, each preliminary mapping endpoint data after mapping is marked with confidence. The higher the confidence, the more accurate the unknown parameters are, which plays a greater role in the process of calibrating the unknown parameters and reduces the impact of data with low confidence.

[0026] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0028] Figure 1 This is a flow chart of the method of the first embodiment.

[0029] Figure 2 This is a comparison chart of the prediction accuracy of different classification models using the PD-CTL-ON data in the first embodiment.

[0030] Figure 3 This is a comparison chart of the prediction accuracy of different classification models using the PD-CTL-OFF data in the first embodiment. DETAILED DESCRIPTION

[0031] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0032] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0033] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0034] Embodiment 1

[0035] In view of the characteristics of Parkinson's disease and the problems existing in the prior art, the embodiment of the present invention provides a method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping, which has a higher recognition accuracy.

[0036] like Figure 1 As shown, the Parkinson's disease EEG signal recognition method based on multi-source mapping may include the following steps:

[0037] Obtain EEG signals of Parkinson's disease and normal healthy controls, construct subject data sets and annotate category labels;

[0038] The test data set is divided into a training set and a test set, and multiple sub-classifiers are trained using the training set. Based on the test results of each sub-classifier on the test set, weights are assigned to the sub-classifiers to obtain a collective classifier containing multiple sub-classifiers and their weight combinations;

[0039] Map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and based on the final mapping path, map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set;

[0040] The corresponding mapping endpoint data in the training set is input into the ensemble classifier to obtain the predicted classification label, thus completing the recognition of the EEG signal to be tested.

[0041] Specifically include:

[0042] (1) Obtain EEG data of Parkinson's disease as training set and test set respectively.

[0043] Obtain a certain number of EEG features of Parkinson's disease patients and normal healthy controls, pair the EEG data of Parkinson's disease patients with similar EEG features with normal healthy controls as a subject, and construct a subject data set. Each subject has a corresponding label as the EEG data marker of Parkinson's disease patients and normal healthy controls, which is used as the label of the classifier training data.

[0044] The subject data set is divided into a training set and a test set. The training set is used as data for classifier training. The training set contains several different types of Parkinson's disease EEG data subjects and corresponding category labels to achieve the purpose of multi-source. Each subject is composed of a Parkinson's patient with the same EEG characteristics and a normal healthy control paired.

[0045] At the same time, data fragments and labels of the test set are obtained to adapt the classification model to new data.

[0046] (2) Establish a collection classifier model.

[0047] Construct a training model for multi-source transfer learning: perform EEG feature classification learning between patients and non-patients for each subject, and train a sub-classifier for each subject; assign weights based on the classification effect of the sub-classifier on the test set data fragments, so that each sub-classification is combined into a collective classifier through weights.

[0048] Further:

[0049] The training set is used to train the SVM classifier of each subject as a sub-classifier. When the model is used in the test set, the model will extract the labeled fragments of the test set subjects, use the fragments to evaluate the classification accuracy of the sub-classifiers and assign weights, and combine them into a collective classifier that classifies the entire test set. The sub-classifiers with high classification accuracy for the test set have a larger weight in the collective classifier to adapt to the data characteristics of the test set.

[0050] (3) Test set data mapping.

[0051] Due to the differences in EEG features between subjects, the model first learns the features of the test set subject data fragments, and through semi-supervised feature mapping, makes the test set data closer to the training data features of each classifier.

[0052] Mapping the test data to the training data will further improve the accuracy of the classifier in classifying the test data. Mapping is divided into supervised learning mapping and semi-supervised learning mapping. Semi-supervised learning can make more full use of training data and achieve better mapping and classification results.

[0053] The training data is called the source domain, and the test data is called the target domain. Mapping the sample data of the target domain to the source domain can make the target domain data have the characteristics of the source domain, so that the source domain classifier can be used more effectively for classification. In the mapping process, the target domain is used as the origin of the mapping, and the source domain is used as the destination of the mapping.

[0054] The mapping endpoint is represented as:

[0055] D i ={d i =R m |i=1,……,n};

[0056] The mapping origin is represented as:

[0057] O={o i ∈R m |i=1,……,n};

[0058] Through the affine transformation formula, o i Transform to d i The process is called mapping:

[0059] d i =Ao i +b Formula (1)

[0060] Among them, A and b are unknown parameters; o i is the mapping origin in the test set; d i is the mapping endpoint in the training set.

[0061] 1) Semi-supervised learning determines the mapping path

[0062] Semi-supervised learning mapping, first extract part of the known label data of the training set to calculate the mapping paths A and b (preliminarily determined unknown parameters A0 and b0). i and d i It is known that A and b are parameters to be determined, namely, mapping paths. We will record the initially calculated mapping paths as A0 and b0.

[0063] In order to avoid excessive transfer or no transfer due to A0 and b0 being too large or too small, we determine the parameters A0 and b0 by weighted square error:

[0064]

[0065] in, is the Frobenius norm of the matrix; is the L2 norm of the vector.

[0066] The first item is used to correct the values ​​of A and b to prevent the transferred values ​​from being too different from the true values. The confidence f i Determines the impact of the data on the results;

[0067] The second term constrains A to prevent A from being too far away from the identity matrix I;

[0068] The third term constrains b to ensure that b is not far from 0, so that A and b remain between no transfer and over transfer.

[0069] Among them, β and γ can adjust A and b to improve accuracy. β and γ can be determined by cross-validation using the following formula:

[0070]

[0071] Where Tr() is the trace of the matrix, and the values ​​of β′ and γ′ can be set to 1 to 3.

[0072] Equation (2) can be solved by the following formulas:

[0073]

[0074]

[0075] After A0 and b0 are determined, the semi-supervised learning mapping will obtain a preliminary determined specific mapping form through formula (1). Based on the preliminary determined specific mapping form, all test set data are mapped through the mapping path to preliminarily obtain feature data, which is the data corresponding to the data in the training set and the test set.

[0076] The mapping end point D and the mapping origin O are known, and we use them as training data for retraining until the final mapping path is obtained.

[0077] 2) Determine the mapping endpoint through the prototype.

[0078] When mapping, we do not directly map the sample data in the target domain to the source domain, but first find some representative samples from the source domain as the mapping end point, and then map the target domain data to these sample representatives. We call these sample representatives mapping prototypes. This can prevent the forced establishment of inappropriate mapping connections due to the inability to find suitable source domain data during the mapping process.

[0079] There are two methods to obtain the prototype, the nearest prototype method and the Gaussian model method.

[0080] Closest prototype method: a clustering algorithm based on k-means.

[0081] Our dataset contains two labels, resulting in two different classes. In the source domain, we use k-means clustering to obtain the cluster center of each class. Each class can have multiple different cluster centers, and the number can be set based on the data scale. The appropriate number of cluster centers can improve the accuracy of mapping. In the target domain, we use the nearest distance method to replace the sample with the prototype closest to the target domain sample to obtain the mapping end point D, thereby obtaining A and b.

[0082] We calculate from o i The distances to all prototypes are sorted in descending order to get the closest to o i This process is called the nearest prototype method.

[0083] Gaussian model method.

[0084] In the Gaussian model, we set the mean of each class to μ i , define o i Projection on the contour surface of the i-th Mahalanobis distance:

[0085]

[0086] in, is the Mahalanobis distance of the i-th class, ρ can control the extended Mahalanobis distance profile and limit the deviation of the projection point from the class mean. Too large ρ may lead to no transfer, while too small ρ may lead to excessive transfer.

[0087] In this embodiment, we take the value of ρ as 50.

[0088] 3) Confidence setting.

[0089] After the original data of the test set is mapped into the feature space, each mapped data value is marked by the confidence. The higher the confidence, the more accurate the obtained parameters A and b are, and they play a greater role in the process of calibrating A and b. The lower the confidence, the smaller the impact on the final mapping. In formula (2), the confidence f is i As a weight.

[0090] When calculating A0 and b0, since we use the labeled data of the test set, there is no need to set the confidence level, so the confidence level is temporarily set to 1. When iteratively calculating A and b, the confidence level needs to be recalculated. Confidence level setting is an important step in the process of semi-supervised learning to include the test set in the training scope and to iteratively correct the mapping path.

[0091] The confidence is calculated using formula (3) (4), where the two parameters θ and τ in the formula can be determined by cross-validation of the data.

[0092] F(x i )=φ(d1-d2) (3)

[0093]

[0094] 4) Iterative learning update path

[0095] By calculating the confidence fi of the feature data, the data mapping can be accurate and better adapted to the classification of the set classifier. After obtaining the preliminary feature data through A0 and b0, each feature data is given a confidence and is re-used as the mapping origin. Then, a new prototype is found in the source domain to obtain the mapping end point, so as to recalculate the new mapping path A and b, and iterate the cycle.

[0096] After multiple cycles of iteration, new and more accurate parameters A and b are obtained, so that the data characteristics of the test set are closer to the training set, thereby completing the semi-supervised style transfer mapping and determining the final mapping path.

[0097] (4) Input the feature data into the classifier for classification and obtain the predicted label of the test set.

[0098] After the above steps determine the final mapping path:

[0099] Acquire the EEG signal to be tested, and map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set based on the final mapping path;

[0100] The corresponding mapping endpoint data in the training set is input into the ensemble classifier, and the predicted classification label is obtained based on the ensemble classifier to complete the recognition of the EEG signal to be tested.

[0101] To verify the effectiveness of the present invention, the following experiments were conducted to evaluate the proposed Parkinson's disease EEG signal recognition method based on multi-source mapping.

[0102] Our experimental data comes from the EEG Parkinson's disease (PD) database. The dataset is divided into two groups: PD patients (PD group) and healthy controls (CTL group). The PD group had two data extractions 7 days apart: the patients took treatment drugs every day, the first time, the data was collected on the 7th day, and the second time, the drugs were stopped and data was collected 7 days later. These datasets are called PD-ON and PD-OFF, respectively. The feature features of the PD group and the CTL group are similar, resulting in the formation of two combined datasets: PD-CTL-ON and PD-CTL-OFF, including 24 pairs of matched subjects. We extracted their EEG feature data for cross-validation to evaluate the classification accuracy of our model.

[0103] We conducted multiple experiments on the classification model and tested its performance by extracting different amounts of labeled data from the test set. The number of labeled data extracted from the test set was 5, 10, 15, and 20.

[0104] like Figure 2 As shown, where:

[0105] ON means: the first data collection during the patient's medication period;

[0106] ON(STM-1) means: STM by nearest prototype method;

[0107] ON(STM-2) means: STM is performed by Gaussian model method.

[0108] like Figure 2 As shown, the vertical axis is the prediction accuracy of the classification model. The prediction accuracy of the nearest prototype method using PD-CTL-ON data is 0.9700, 0.9893, 0.9947 and 0.9979 respectively. The prediction accuracy using the Gaussian model method is 0.9766, 0.9942, 0.9972 and 0.9992 respectively. The classification accuracy of data without migration mapping is 0.9062, 0.9089, 0.9147 and 0.9021 respectively.

[0109] Each accuracy value here is the average of 24 accuracy values ​​obtained from cross validation on 24 subjects.

[0110] like Figure 3 As shown, where:

[0111] OFF means: data collection was performed 7 days after the patient stopped taking the drug;

[0112] OFF(STM-1) means: STM is performed by the nearest prototype method;

[0113] OFF (STM-2) means: STM is performed by the Gaussian model method.

[0114] like Figure 3 As shown, the vertical axis is the prediction accuracy of the classification model. The prediction accuracy of the closest prototype method using the data PD-CTL-OFF is 0.9747, 0.9869, 0.9938 and 0.9972 respectively. The prediction accuracy using the Gaussian model method is 0.9809, 0.9894, 0.9966 and 0.9968 respectively. The classification accuracy without STM is 0.9182, 0.9259, 0.9238 and 0.9303 respectively.

[0115] Each accuracy value here is the average of 24 accuracy values ​​obtained from cross validation on 24 subjects.

[0116] In Table 1, we compare the existing methods with our method and show that our method has better accuracy. Figure 2 and Figure 3 It can be seen that the prediction accuracy of the model with mapping is significantly higher than that of the model without mapping, among which the Gaussian model method has the best mapping prediction effect when searching for prototypes.

[0117] Experimental results show that the Parkinson's disease local electroencephalographic discharge diagnosis method proposed in the present invention has achieved a high diagnostic accuracy rate.

[0118] Table 1 is a comparison between the existing method and the method of this embodiment

[0119]

[0120] In Table 1, we compare the classification performance of Logistic Regressive, J48 decision tree, SVM classifier, Random Forest, Convolutional Neural Network, Recurrent Neural Network, KNN, ASGCNN and our multivariate mapping method. The number of trained subjects, number of EEG data acquisition channels and prediction accuracy of each method are listed and compared. It shows that our multivariate mapping method only needs less training data and fewer data acquisition channels to achieve high-accuracy prediction.

[0121] This embodiment is based on a multi-source transfer learning method to solve the tasks in the above background technology from a new perspective. The method of this embodiment does not simply train a single classifier, but trains multiple sub-classifiers and then weights and combines the sub-classifiers to construct a set classifier; by mapping the test data, changing the features of the data to adapt to the sub-classifiers to improve the classification effect of the classifier, and then weighting to obtain the final classification result.

[0122] The classifier based on the training set subjects is used to classify the EEG of Parkinson's disease patients and non-patients through multi-source transfer learning. The experimental results show that the method of this embodiment can effectively improve the accuracy of Parkinson's disease EEG classification.

[0123] Embodiment 2

[0124] This embodiment discloses a Parkinson's disease EEG signal recognition system based on multi-source mapping.

[0125] Parkinson's disease EEG signal recognition system based on multi-source mapping, including:

[0126] The dataset construction module is configured to: obtain EEG signals of subjects with Parkinson's disease and normal healthy controls, construct the subject dataset and annotate the category labels;

[0127] The ensemble classifier training module is configured to: divide the test data set into a training set and a test set, train multiple sub-classifiers using the training set, assign weights to the sub-classifiers based on the test effect of each sub-classifier on the test set, and obtain an ensemble classifier including multiple sub-classifiers and their weight combinations;

[0128] A mapping module is configured to: map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set based on the final mapping path;

[0129] The recognition module is configured to: input the corresponding mapping endpoint data in the training set into the set classifier, obtain the predicted classification label, and complete the recognition of the EEG signal to be tested.

[0130] Embodiment 3

[0131] The purpose of this embodiment is to provide a computer-readable storage medium.

[0132] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the method for identifying Parkinson's disease electroencephalogram signals based on multi-source mapping as described in Example 1 of the present disclosure.

[0133] Embodiment 4

[0134] The purpose of this embodiment is to provide an electronic device.

[0135] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping as described in Example 1 of the present disclosure are implemented.

[0136] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0137] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0138] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A Parkinson's disease EEG signal recognition method based on multi-source mapping, characterized in that: The following steps are involved: Obtain EEG signals from subjects with Parkinson's disease and healthy controls, construct a subject dataset and annotate category labels; The test data set is divided into a training set and a test set, and multiple sub-classifiers are trained using the training set. Based on the test results of each sub-classifier on the test set, weights are assigned to the sub-classifiers to obtain a collective classifier containing multiple sub-classifiers and their weight combinations; Map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and based on the final mapping path, map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set; The corresponding mapping endpoint data in the training set is input into the ensemble classifier to obtain the predicted classification label, thus completing the recognition of the EEG signal to be tested.

2. The method for identifying Parkinson's disease EEG signals based on multi-source mapping as claimed in claim 1, characterized in that: The subject data set includes multiple subjects, each of which includes Parkinson's disease EEG signals and normal healthy control EEG signals with similar EEG characteristics; the training set includes several different types of subjects and corresponding category labels.

3. The method for identifying Parkinson's disease EEG signals based on multi-source mapping as claimed in claim 2, characterized in that: For each subject in the training set, a corresponding SVM classifier is trained as a sub-classifier; the sub-classifier with a higher classification accuracy for the test set has a greater weight in the ensemble classifier.

4. The method for identifying Parkinson's disease EEG signals based on multi-source mapping as claimed in claim 1, characterized in that: Map the data in the test set to the training set through semi-supervised learning to determine the final mapping path, including: Step 1: Setting a specific mapping form from the test set to the training set, wherein the specific mapping form includes unknown parameters; Step 2: Calculate the preliminarily determined unknown parameters through the weighted square error formula; Step 3: Based on the initially determined unknown parameters, a initially determined specific mapping form is obtained; Step 4: Map the test set to the training set according to the initially determined specific mapping form, and select appropriate initial mapping endpoint data from it; Step 5: Calculate the confidence of the preliminary mapping endpoint data; Step 6: Based on the calculated confidence level, iterate the process from step 2 to step 5 above to calibrate and update the unknown parameters initially determined, stop after the set number of iterations, and determine the final unknown parameters; Step 7: Based on the final unknown parameters, determine the final mapping path.

5. The method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping as claimed in claim 4, characterized in that: In step 1, the specific mapping form is set as follows: d i =Ao i +b Among them, A and b are unknown parameters; i is the mapping origin in the test set; d i is the mapping endpoint in the training set; In the step 2, the unknown parameters preliminarily determined are calculated by the weighted square error formula, specifically: in, is the Frobenius norm of the matrix; is the L2 norm of the vector; f i is the confidence level; β and γ are adjustment parameters; The solved A and b are used as the preliminarily determined unknown parameters A0 and b0.

6. The method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping as claimed in claim 5, characterized in that: In the step 4: Substitute A0 and b0 back into the specific mapping form to obtain a preliminarily determined specific mapping form; Map the test set to the training set using the initially determined specific mapping form; Appropriate preliminary mapping endpoint data were selected from the training set using the nearest prototype method and Gaussian model method.

7. The method for recognizing Parkinson's disease electroencephalogram signals based on multi-source mapping according to claim 4, characterized in that: In the step five and step six: Each preliminary mapping endpoint data after mapping is marked with confidence. The higher the confidence, the more accurate the unknown parameter is, and it plays a greater role in the process of calibrating the unknown parameter. The calculated confidence level is brought back into the weighted square error formula, and the unknown parameters are updated repeatedly until the set number of iterations is reached to obtain the final unknown parameters.

8. Parkinson's disease EEG signal recognition system based on multi-source mapping, characterized by: include: The dataset construction module is configured to: obtain EEG signals of Parkinson's disease and normal healthy controls, construct the subject dataset and annotate the category labels; The ensemble classifier training module is configured to: divide the test data set into a training set and a test set, train multiple sub-classifiers using the training set, assign weights to the sub-classifiers based on the test effect of each sub-classifier on the test set, and obtain an ensemble classifier including multiple sub-classifiers and their weight combinations; A mapping module is configured to: map the data in the test set to the training set through semi-supervised learning, determine the final mapping path, and map the acquired EEG signal to be tested to the corresponding mapping endpoint data in the training set based on the final mapping path; The recognition module is configured to: input the corresponding mapping endpoint data in the training set into the set classifier, obtain the predicted classification label, and complete the recognition of the EEG signal to be tested.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method for identifying Parkinson's disease electroencephalogram signals based on multi-source mapping as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the method for identifying Parkinson's disease electroencephalogram signals based on multi-source mapping as described in any one of claims 1 to 7 are implemented.

Images