Depression electroencephalogram signal classification method for screening deep forest based on cross-layer confidence coefficient

Abstract

The invention discloses a depression electroencephalogram signal classification method based on cross-layer confidence coefficient screening depth forest, which is characterized by comprising the following steps of: S1, acquiring depression electroencephalogram signal data of Jidai courtyard, carrying out time domain and frequency domain feature conversion, carrying out preprocessing operation on the data, and dividing the data into a training set and a test set; and S2, constructing a depression electroencephalogram signal classification model with a cross-layer confidence screening function based on a deep forest technology. The model is composed of two main modules of multi-granularity scanning and cascade forests, and each cascade forest comprises two random forests and two complete random forests; s3, in the training process, training data are sent into the cross-layer confidence screening electroencephalogram signal classification model, the model carries out detailed sliding window analysis on the data by utilizing multi-granularity scanning, and time domain and frequency domain features of the processed electroencephalogram signals obtained during data preprocessing are spliced to obtain enhanced features; s4, the enhanced features screen a cascade forest layer by layer through cross-layer confidence, output of each layer is a prediction result, operation of crossing cascade layers is achieved through confidence screening and comparison, and finally a trained depression electroencephalogram signal classification model is formed. Compared with the prior art, time domain and frequency domain feature splicing is added in the multi-granularity scanning stage, a cross-layer confidence coefficient screening mechanism is added in the cascade forest stage, and the feature extraction capacity of the model for complex electroencephalogram data is enhanced, so that key information is captured and analyzed more accurately, the performance of the model is effectively optimized, and the method is suitable for being popularized and applied. The classification accuracy and efficiency are improved, and powerful support is provided for the development of the real-time electroencephalogram signal analysis and application field.

Description

Technical Field

[0001] The present invention relates to the field of depression EEG signal classification, and in particular to a depression EEG signal classification method based on improved deep forest. Background Art

[0002] Depression EEG signal classification is an important research direction that has attracted much attention in the field of neuroscience and medicine in recent years. Depression is a serious mental health disorder that affects millions of people worldwide. As a non-invasive biosignal recording method, electroencephalogram (EEG) can capture the spatiotemporal characteristics of brain electrical activity and is therefore widely used in the study and diagnosis of depression.

[0003] In the past, EEG-based depression research mainly relied on traditional data analysis methods and expert judgment. However, these methods are often limited by subjectivity and low standardization. With the advancement of technology and the improvement of computing power, it has become possible to automatically classify depression EEG signals using machine learning and deep learning techniques.

[0004] The background of depression EEG signal classification includes many challenges and innovations. First, high-dimensional EEG data requires effective feature extraction and selection methods to reduce data dimensions and improve classification accuracy. Second, the diversity and individual differences of depression require the classification model to have a certain generalization ability to adapt to the EEG signal patterns of different individuals. In addition, how to overcome noise interference and the lack of labeled samples during data collection is also an important issue in current research.

[0005] In recent years, with the maturity and application of deep learning technology, researchers have begun to explore the use of deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN) and attention mechanisms to extract more representative and discriminative features from complex EEG data. The development of these technologies has not only improved the accuracy of depression EEG signal classification, but also provided new possibilities for early diagnosis and personalized treatment of depression based on EEG. Overall, the background of depression EEG signal classification shows the characteristics of multidisciplinary integration, involving cross-applications in neuroscience, psychology, computer science and other fields, aiming to deeply understand and respond to the global challenge of depression through advanced technical means. Summary of the invention

[0006] The present invention aims to provide a method for improving the accuracy and efficiency of depression EEG signal classification. The method involves collecting and preprocessing remote sensing data, and dividing it into training and test data sets. Then, an improved depression EEG signal classification model is constructed using a technique based on feature combination and cross-layer confidence screening. The model consists of a multi-granularity scan and a cascade forest module, wherein each cascade forest layer includes two random forests and two completely random forests. By performing a multi-granularity scan on the training set and combining the spliced ​​time domain and frequency domain features, the subsamples are processed by each cross-layer confidence screening cascade forest, and the output results of each layer are used to optimize the final classification model. The characteristic of this method is that the time domain and frequency domain features are combined in the multi-granularity scan stage, and a hierarchical confidence screening module is added to the cascade forest to ensure that the overall system can effectively and comprehensively utilize information from different levels and feature sources, thereby optimizing the performance and efficiency of the model, while reducing the computational complexity.

[0007] The present invention adopts the following technical solution: a method for classifying depression EEG signals based on cross-layer confidence screening deep forest, comprising:

[0008] Obtain EEG signal data for preprocessing, including removing or correcting possible noise and artifacts, normalizing the signal, reducing the noise in the spectrum data through spectrum smoothing technology, and finally dividing the preprocessed EEG signal data into a training set and a test set;

[0009] A depression EEG signal classification model is constructed based on an improved deep forest; the depression EEG signal classification model includes a feature splicing multi-granularity scanning module and a confidence screening cascade forest module; each cascade layer of the cascade forest module includes two random forests and two completely random forests;

[0010] The training set is input into a depression EEG signal classification model for training. The depression EEG signal classification model performs sliding window scanning on the training set through a feature splicing multi-granularity scanning module to generate sub-samples, and splices the time domain and frequency domain features of the EEG signal samples;

[0011] After obtaining the samples of time domain and frequency domain splicing, they are input into the cascade forest with hierarchical confidence screening.

[0012] The subsample passes through each cascade layer of the cascade forest module, and the classification result is output according to the confidence of each cascade layer to obtain a trained depression EEG signal classification model.

[0013] Furthermore, before the training set is input into the depression EEG signal classification model for training, the following steps are also included:

[0014] By performing Fourier transform on the EEG signal data in the training set, the time domain features are converted into frequency domain features.

[0015] Furthermore, after the original sample is input into the multi-granularity scanning module of the deep forest, the input features are processed and slid through sliding windows of different sizes to obtain time domain and frequency domain feature vectors of different sizes.

[0016] The time domain feature vector and frequency domain feature vector of the same sliding window are further concatenated:

[0017] Time domain and frequency domain feature splicing formula:

[0018] x concat =[x time ,x freq ]

[0019] where x time represents the time domain feature vector, x freq represents the frequency domain eigenvector, x concat It means that they are spliced ​​into a new vector, and each granularity size is spliced ​​in turn. The time domain and frequency domain splicing data obtained by multi-granularity scanning through a sliding window are input into a random forest and a completely random forest, and finally a feature vector is constructed.

[0020] The resulting feature vector is input into a cross-layer confidence-filtered cascade forest.

[0021] In the design of Cascade Forest, we first calculate the average precision after completing the first layer training. This accuracy evaluation becomes the key decision metric for determining the processing path for each instance.

[0022] The trained model of the first layer is applied to all instances and their prediction accuracy is calculated. Next, we set a threshold and use this average accuracy as the standard. Those instances with accuracy below the threshold are pushed to the next layer (i.e., layer n+1) for analysis and correction.

[0023] Those instances with accuracy below the threshold are pushed to the next layer (i.e., layer n+1) for analysis and correction. Those instances with accuracy above the threshold are pushed to the next stage (i.e., layer n+2) for further in-depth processing and decision making.

[0024] This strategy can not only dynamically adjust the processing path according to the prediction accuracy of the instance, but also optimize the performance and efficiency of the overall system.

[0025] The method for obtaining the average precision is:

[0026]

[0027] The model method for confidence screening is:

[0028] F i={pre(A i )|P i ≥ω}

[0029] Where pre(A i ) represents the i-th sample A i The predicted value of ω is the confidence screening threshold.

[0030] Furthermore, when the cascade forest module increases the cascade layer and transmits the subsample to the next cascade layer, it also includes:

[0031] The accuracy of each cascade layer is obtained in turn. When the accuracy of the next cascade layer does not change, the cascade forest module stops growing the cascade layer, and the classification results of the corresponding cascade layer output sub-samples whose current accuracy has not changed.

[0032] Furthermore, the accuracy rate is calculated as follows:

[0033] The accuracy is the ratio of correctly classified samples to the total number of samples, that is,

[0034]

[0035] Among them, n correct is the number of samples correctly classified, n total is the total number of samples.

[0036] Combined with the confusion matrix above, the formula can also be written as follows:

[0037]

[0038] Accuracy is the simplest and most intuitive evaluation indicator in classification problems.

[0039] The calculation method of the F1 score is:

[0040]

[0041] Precision reflects the model's ability to distinguish negative samples. The higher the Precision, the stronger the model's ability to distinguish negative samples. Recall reflects the model's ability to recognize positive samples. The higher the Recall, the stronger the model's ability to recognize positive samples. F1 score is a combination of the two. The higher the F1 score, the more robust the model.

[0042] The beneficial effects of the present invention are as follows: The main innovation of this study is that the time domain and frequency domain features of the EEG signal are effectively spliced ​​in the multi-granularity scanning module of the deep forest model. This improvement aims to enhance the model's feature extraction capability for complex EEG data, thereby more accurately capturing and analyzing key information. In addition, a cross-layer confidence screening mechanism is introduced in the cascade forest stage of the gcForest model. By evaluating the classification confidence of each instance, the model can selectively pass specific instances to the next level model in a targeted manner, thereby optimizing the overall prediction performance. This strategy can not only effectively transfer and filter information between different levels of the model, but also reduce computational costs and resource consumption while maintaining high classification accuracy. By evaluating the classification confidence of each instance, the performance of the model is effectively optimized, the classification accuracy and efficiency are improved, and strong support is provided for the progress in the field of real-time EEG signal analysis and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 The present invention is a flowchart of a method for classifying depression EEG signals based on cross-layer confidence screening deep forest.

[0044] Figure 2 The invention discloses a time domain and frequency domain splicing multi-granularity scanning module for a depression EEG signal classification method based on cross-layer confidence screening deep forest.

[0045] Figure 3 The present invention discloses a confidence screening cascade forest module for a depression EEG signal classification method based on a cross-layer confidence screening deep forest. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] A flowchart of a method for classifying depression EEG signals based on cross-layer confidence screening deep forests according to an embodiment of the present invention is shown in FIG. Figure 1 As shown, including:

[0048] Obtaining EEG signal data for preprocessing, and dividing the preprocessed EEG signal data into a training set and a test set;

[0049] The method of obtaining EEG signal data for preprocessing is as follows: First, extract the EEG signal data from MATLAB, which involves loading and parsing the raw data file, containing the voltage time series collected from the EEG device, recorded at a specific sampling frequency (e.g., several hundred samples per second).

[0050] After loading the data, perform basic data preprocessing, including removing or correcting possible noise, artifacts, or motion artifacts, etc. Split the data into a data set for model training and a data set for testing.

[0051] The next step is to split the data into time windows of fixed length. Here we use 500 and 1000 data points (or time units) as the size of a sliding window to obtain the original EEG signal dataset.

[0052] At the same time, for the EEG signals in the same time window, frequency domain features are extracted. This can be done by applying methods such as Fourier transform or wavelet transform to convert the signal from the time domain to the frequency domain. Frequency domain features include power spectral density, frequency characteristics and coherence of different frequency bands (such as delta wave, theta wave, alpha wave, beta wave, gamma wave).

[0053] A depression EEG signal classification model is constructed based on a deep forest; the depression EEG signal classification model includes a time domain and frequency domain feature splicing multi-granularity scanning module and a cross-layer confidence screening cascade forest module; each cascade layer of the cascade forest module includes two random forests and two completely random forests; the present invention constructs a depression EEG signal classification network model based on the deep forest model, and the depression EEG signal classification network model includes two parts: time domain and frequency domain feature splicing multi-granularity scanning and cross-layer confidence screening cascade forest, and each cascade layer of the cascade forest includes two rotation forests and two completely random forests as classifiers.

[0054] The training set is input into a depression EEG signal classification model for training. The depression EEG signal classification model performs a sliding window scan on the training set through a feature splicing multi-granularity scanning module to generate sub-samples, and splices the time domain and frequency domain features of the EEG signal samples to obtain a new time domain and frequency domain splicing feature;

[0055] After obtaining the samples of time domain and frequency domain splicing, they are input into the cascade forest with hierarchical confidence screening.

[0056] The subsample passes through each cascade layer of the cascade forest module, and the classification result is output according to the confidence of each cascade layer to obtain a trained depression EEG signal classification model.

[0057] Furthermore, before the training set is input into the depression EEG signal classification model for training, the following steps are also included:

[0058] By performing Fourier transform on the EEG signal data in the training set, the time domain features are converted into frequency domain features.

[0059] Furthermore, after the original sample is input into the multi-granularity scanning module of the deep forest, the input features are processed and slid through sliding windows of different sizes to obtain time domain and frequency domain feature vectors of different sizes.

[0060] The time domain feature vector and frequency domain feature vector of the same sliding window are further concatenated:

[0061] Time domain and frequency domain feature splicing formula:

[0062] x concat =[x time ,x freq ]

[0063] where x time represents the time domain feature vector, x freq represents the frequency domain eigenvector, x concat It means that they are spliced ​​into a new vector, and each granularity size is spliced ​​in turn. The time domain and frequency domain splicing data obtained by multi-granularity scanning through a sliding window are input into a random forest and a completely random forest, and finally a feature vector is constructed.

[0064] The resulting feature vector is input into a cross-layer confidence-filtered cascade forest.

[0065] In the design of Cascade Forest, we first calculate the average precision after completing the first layer training. This accuracy evaluation becomes the key decision metric for determining the processing path for each instance.

[0066] The trained model of the first layer is applied to all instances and their prediction accuracy is calculated. Next, we set a threshold and use this average accuracy as the standard. Those instances with accuracy below the threshold are pushed to the next layer (i.e., layer n+1) for analysis and correction.

[0067] Those instances with accuracy below the threshold are pushed to the next layer (i.e., layer n+1) for analysis and correction. Those instances with accuracy above the threshold are pushed to the next stage (i.e., layer n+2) for further in-depth processing and decision making.

[0068] This strategy can not only dynamically adjust the processing path according to the prediction accuracy of the instance, but also optimize the performance and efficiency of the overall system.

[0069] The method for obtaining the average precision is:

[0070]

[0071] Where pre(A i ) represents the i-th sample A i The predicted value of ω is the confidence screening threshold.

[0072] From this, the model of confidence screening cascade forest can be obtained by the formula:

[0073] F i ={pre(A i )|P i ≥ω}

[0074] A precision threshold ω is set as a standard. All instances with prediction accuracy below this threshold will be pushed to the next layer (i.e., layer n+1) for further analysis and correction. Those instances with prediction accuracy above the threshold are pushed to the next stage (i.e., layer n+2) for more in-depth processing and decision making.

[0075] In other words, this process further processes instances with insufficient precision through a layered strategy to ensure that the final processing stage (n+2th layer) can process high-quality data or decisions.

[0076] At the same time, when the cascade forest module of the present invention increases the cascade layer and transmits the subsample to the next cascade layer, it also includes: obtaining the accuracy of each cascade layer in turn, and when the accuracy of the next cascade layer does not change, the cascade forest module stops increasing the cascade layer, and the prediction result of the subsample output corresponding to the cascade layer that does not change the current accuracy. The calculation method of the accuracy is:

[0077] The accuracy is the ratio of correctly classified samples to the total number of samples, that is,

[0078]

[0079] Among them, n correct is the number of samples correctly classified, n total is the total number of samples.

[0080] Combined with the confusion matrix above, the formula can also be written as follows:

[0081]

[0082] Accuracy is the simplest and most intuitive evaluation indicator in classification problems.

[0083] Furthermore, the F1 score is calculated to evaluate the weighted rotation deep forest model.

[0084] The calculation method of F1 score is:

[0085]

[0086] Precision reveals the effectiveness of the model in identifying and excluding non-target (negative) samples; a high precision means that the model performs well in distinguishing non-target samples. Recall maps the model's ability to capture target (positive) samples; a high recall means that the model can effectively detect target samples. The F1 score is an indicator for evaluating the stability of the model, combining the advantages of precision and recall. A high F1 score represents the robustness of the model's combination of these two capabilities.

[0087] In the present invention, the time domain and frequency domain features of the EEG signal are effectively spliced ​​in the multi-granularity scanning module of the deep forest model. This improvement aims to enhance the model's feature extraction capability for complex EEG data, thereby more accurately capturing and analyzing key information. The present invention also introduces a cross-layer confidence screening mechanism in the cascade forest stage of the gcForest model. By evaluating the classification confidence of each instance, it can selectively pass specific instances to the next level model in a targeted manner, thereby optimizing the overall prediction performance. This strategy can not only effectively transfer and filter information between different levels of the model, but also reduce computing costs and resource consumption while maintaining high classification accuracy. By evaluating the classification confidence of each instance, the performance of the model is effectively optimized, the classification accuracy and efficiency are improved, and strong support is provided for the progress in the field of real-time EEG signal analysis and application.

[0088] It should be clear that these are only recommended implementations of the present invention, and the present invention is not limited to these examples. The spirit of the present invention lies in innovation and improvement, and any modification, substitution or improvement made under the guidance of the present invention should be deemed to fall within the protection scope of the present invention.

Claims

1. A depression EEG signal classification method based on cross-layer confidence screening deep forest, characterized in that: The method comprises the following specific steps: The EEG signal data of depressed patients in the First Hospital of Jilin University were obtained, and the obtained EEG signal data were preprocessed, and the data were divided into a data set for model training and a data set for testing. Construct a depression EEG signal classification model based on deep forest; The depression EEG signal classification model includes a multi-granularity scanning module and a cascade forest module; each cascade layer of the cascade forest module includes two completely random forests and two random forests; The processed training data is input into the constructed model for training. The model first uses sliding windows of different sizes of the multi-granularity scanning module to perform multi-scale segmentation on the data, thereby generating time domain and frequency domain splicing sub-data samples of various scales. The sub-samples scanned by the sliding window will be input into the confidence filtering cascade forest of the deep forest in turn. Each cross-layer confidence filtering cascade forest layer in the cascade forest model will be used to further refine the classification results through the confidence filtering strategy, greatly shortening the running time, and finally forming a trained hierarchical confidence deep forest model for the classification of depressive EEG signals.

2. According to claim 1, a method for classifying depression EEG signals based on cross-layer confidence screening deep forest, the steps are: before inputting the training set into the deep forest model for training, it also includes: By preprocessing the EEG signal data in the training set, first, extract the EEG signal data from MATLAB, which involves loading and parsing the raw data file, which contains the voltage time series collected from the EEG device and recorded at a specific sampling frequency (such as hundreds of samples per second). After loading the data, basic data preprocessing is performed, including removing or correcting possible noise and artifacts, normalizing the signal, reducing the noise in the spectrum data through spectral smoothing technology, and splitting the data into a data set for model training and a data set for testing. The next step is to split the data into time windows of fixed length. Here we use 500 and 1000 data points (or time units) as the size of a sliding window to obtain the original EEG signal dataset. At the same time, for the EEG signals in the same time window, frequency domain features are extracted. This can be done by applying methods such as Fourier transform or wavelet transform to convert the signal from the time domain to the frequency domain. Frequency domain features include power spectral density, frequency characteristics and coherence of different frequency bands (such as delta wave, theta wave, alpha wave, beta wave, gamma wave).

3. The method for classifying depression EEG signals based on cross-layer confidence screening deep forest according to claim 1, characterized in that: The superposition feature method of the multi-granularity scanning is: Perform multi-scale analysis on EEG signals and extract time domain and frequency domain features in different time windows. These features can capture the changing patterns of signals at different times and frequencies, providing basic data for subsequent analysis and processing. In each time window, the time domain and frequency domain features are combined into a feature vector. This feature vector contains the integration of time domain and frequency domain information, so that the signal in each time window can be represented in the form of a comprehensive feature. x concat =[x time ,x freq ] By taking the time domain features as the first half of the vector and the frequency domain features as the second half of the vector, the feature vector retains the timing information of the original signal and contains the results of frequency domain analysis.

4. The method for classifying depression EEG signals based on cross-layer confidence deep forest according to claim 1, characterized in that: In the design of Cascade Forest, we first calculate the average precision after completing the first layer training. This accuracy evaluation becomes the key decision metric for determining the processing path for each feature. Specifically, we apply the first layer trained model to all instances and calculate their prediction accuracy. Then, we set a threshold and use this average accuracy as the standard. Those instances with accuracy below the threshold are pushed to the next layer (i.e., layer n+1) for analysis and correction. Those instances with accuracy above the threshold are pushed to the next stage (i.e., layer n+2) for further and more in-depth processing and decision-making. This strategy can not only dynamically adjust the processing path according to the prediction accuracy of the instance, but also optimize the performance and efficiency of the overall system. The method for obtaining the sample average precision is: The model method for confidence screening is: F i ={pre(A i )|P i ≥ω} Where pre(A i ) represents the i-th sample A i Instances above the threshold ω are pushed to the n+2 layer (the third layer), which means they receive deeper processing and decision-making, which may involve more complex models or more feature combinations. Instances below the threshold enter the n+1 layer (the second layer), which can continue to be optimized and analyzed based on the original features to improve their predictive ability.

5. The method for classifying depression EEG signals based on cross-layer confidence screening deep forest according to claim 1, characterized in that: When the cascade forest module cascades layers and transmits subsamples to the next cascade layer, it also includes: The feature vector extracted using the sliding window of the initial size is first input into the first layer of the cascade forest for training, thereby obtaining a category distribution vector. Next, this category distribution vector is combined with the original feature vector to form a new feature set for use in the next cascade forest layer. In each subsequent layer, the category distribution vector obtained in this layer is also combined with the original feature vector extracted from the sliding window of the corresponding size obtained from the multi-granularity scan. In addition, instances with confidence levels higher than the average confidence level in the first two layers (n-2 layers) must also be considered to ensure that the overall system can effectively and comprehensively utilize information from different levels and feature sources, thereby optimizing the performance and efficiency of the model. This merging and training step continues until a stable point of accuracy is reached. That is, when the accuracy of a certain cascade layer no longer improves, stop building more cascade layers and use the prediction results of this layer for the sample.

6. The method for classifying depression EEG signals based on cross-layer confidence screening deep forest according to claim 5, characterized in that: The accuracy rate is calculated as follows: The accuracy is the ratio of correctly classified samples to the total number of samples, that is, Among them, n correct is the number of samples correctly classified, n total is the total number of samples. Combined with the confusion matrix above, the formula can also be written as: Accuracy is the simplest and most intuitive evaluation indicator in classification problems.

7. The method for classifying depression EEG signals based on cross-layer confidence screening deep forest according to claim 5, characterized in that: The calculation method of the F1 score is: Precision measures the model's ability to correctly identify non-target categories. A higher precision indicates a lower misjudgment rate. Recall measures the model's ability to detect target categories. A higher recall rate indicates a more comprehensive recognition of target categories. The F1 score is the harmonic average of these two indicators. A higher F1 score means that the model maintains a higher precision while maintaining the recall rate, indicating that the overall performance of the model is relatively balanced.

Images