信号处理--基于混合CNN和transfomer自注意力的多通道脑电信号的情绪分类的简单应用

关于

工具

数据集

数据集简述

方法实现

数据读取

编辑数据预处理

传统机器学习模型(逻辑回归，支持向量机，随机森林)

多层感知机模型

CNN+transfomer模型

代码获取

关于

本实验利用结合了卷积神经网络 (CNN) 和 Transformer 组件的混合架构，实现基于 EEG 的有效情绪分类。
尝试各种机器学习模型，包括逻辑回归、支持向量机 (SVM)、随机森林分类器和多层感知器 (MLP) 神经网络，以比较不同模型的性能。

图片来自于： https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9991178

工具

数据集

数据集简述

脑电图数据是从两名受试者（1 名男性、1 名女性，年龄 20-22 岁）收集的，针对特定电影剪辑引发的六种情绪状态（积极、消极、中性）中的每一种状态。该数据集包括从脑电波中收集的 324,000 个数据点，这些数据点被重新采样到 150Hz。还收集了中性脑电波数据，作为代表受试者静息情绪状态的第三类数据。从四个电极（TP9、AF7、AF8、TP10）记录 EEG 数据，并进行处理以生成通过 1 秒滑动窗口提取的统计特征数据集。

图片来源：https://www.researchgate.net/figure/This-figure-shows-the-standard-locations-for-measuring-EEG-as-per-10-20-International_fig2_358644174

方法实现

数据读取

raw_eeg_data = pd.read_csv('../data/features_raw.csv')
raw_eeg_data.head()

# plot the F8 column
plt.figure(figsize=(20, 5))
plt.plot(raw_eeg_data['F8'])
plt.title('F8 Electrode Data')
plt.ylabel('Voltage (uV)')
plt.xlabel('Time')
plt.show()


# plot the F7 column
plt.figure(figsize=(20, 5))
plt.plot(raw_eeg_data['F7'])
plt.title('F7 Electrode Data')
plt.ylabel('Voltage (uV)')
plt.xlabel('Time')
plt.show()

数据预处理

X = eeg_emotions_data.drop(['label'], axis=1)
y = eeg_emotions_data['label']

# Encoding categorical data
from sklearn.preprocessing import LabelEncoder, OneHotEncoder

labelencoder_emotions = LabelEncoder()
y = labelencoder_emotions.fit_transform(y)


# Standardizing the features in the dataset
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()

X = scaler.fit_transform(X)

传统机器学习模型(逻辑回归，支持向量机，随机森林)

from sklearn.linear_model import LogisticRegression
import pickle

# Create a logistic regression classifier
model = LogisticRegression(random_state=2003, multi_class='multinomial', max_iter=1000)

# Train the model
model.fit(X_train, y_train)

# Evaluate the model
evaluate_model(y_test, model.predict(X_test))

from sklearn.svm import SVC

# Create a model: a support vector classifier
model = SVC(kernel='rbf', gamma='auto', C=1.0, random_state=2003)

# Train the model
model.fit(X_train, y_train)

# Evaluate the model
evaluate_model(y_test, model.predict(X_test))

from sklearn.ensemble import RandomForestClassifier

# Create a random forest Classifier.
model = RandomForestClassifier(n_estimators=100, random_state=2003)

# Train the model
model.fit(X_train, y_train)

# Evaluate the model
evaluate_model(y_test, model.predict(X_test))

在传统机器模型，我们可以发现随机森林的性能表现最好；

多层感知机模型

class EEGClassifier(nn.Module):
    def __init__(self, input_dim, num_classes, hidden_dim=256):
        super(EEGClassifier, self).__init__()

        self.fc1 = nn.Linear(input_dim, hidden_dim)
        self.relu = nn.ReLU()
        self.fc2 = nn.Linear(hidden_dim, num_classes)

    def forward(self, x):
        x = self.fc1(x)
        x = self.relu(x)
        x = self.fc2(x)
        return x


input_dim = 2548  # Number of features in EEG signal
num_classes = 3   # Number of classes for classification
model = EEGClassifier(input_dim, num_classes)
loss = nn.CrossEntropyLoss()

CNN+transfomer模型

class EEGConformer(nn.Module):
    def __init__(self, input_dim, num_classes):
        super(EEGConformer, self).__init__()

        # CNN
        self.conv1 = nn.Conv2d(1, 40, kernel_size=(1, 25), stride=(1, 1))
        self.conv2 = nn.Conv2d(40, 40, kernel_size=(1, input_dim), stride=(1, 1))
        self.batchnorm = nn.BatchNorm2d(40)

        # Transformer
        self.layernorm1 = nn.LayerNorm(40)
        self.multiheadattention = nn.MultiheadAttention(40, 1)
        self.layernorm2 = nn.LayerNorm(40)

        self.feedworward_block = nn.Sequential(
            nn.Linear(40, 32),
            nn.GELU(),
            nn.Dropout(p=0.1),
            nn.Linear(32, 40)
        )

        # MLP
        self.fc1 = nn.Linear(40, 32)
        self.fc2 = nn.Linear(32, 32)
        self.fc3 = nn.Linear(32, num_classes)

    def forward(self, x):
        # CNN
        x = x.unsqueeze(1).unsqueeze(1)
        x = self.conv1(x)
        x = self.conv2(x)
        x = self.batchnorm(x)

        # Transformer
        x = x.squeeze()
        x = self.layernorm1(x)
        attn_out = self.multiheadattention(x, x, x)
        x = x + nn.Dropout(0.1)(attn_out[0])
        x = self.layernorm2(x)
        x = self.feedworward_block(x)
        x = nn.Dropout(p=0.1)(x)

        # MLP
        x = self.fc1(x)
        x = F.elu(x)
        x = nn.Dropout(p=0.5)(x)
        x = self.fc2(x)
        x = F.elu(x)
        x = nn.Dropout(p=0.3)(x)
        x = self.fc3(x)
        
        return x

input_dim = 2524  # Number of features in EEG signal
num_classes = 3   # Number of classes for classification
model = EEGConformer(input_dim, num_classes)
loss = nn.CrossEntropyLoss()