基于CNN和BiGRU的八分类特征融合模型

import numpy as np
import pandas as pd
import tensorflow as tf
from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Embedding, Conv1D, MaxPooling1D, GlobalMaxPooling1D, Dense, Concatenate
from tensorflow.keras.layers import Bidirectional, GRU, Dropout, Flatten
from tensorflow.keras.callbacks import EarlyStopping

# 读取数据集
data = pd.read_csv('data.txt', header=None)
features = data.iloc[:, :-1].values
labels = data.iloc[:, -1].values

# 将特征值转化为图片格式
image_features = features[:, :23].reshape(-1, 23, 1)

# 划分训练集、验证集和测试集
train_ratio = 0.7
val_ratio = 0.15
test_ratio = 0.15

train_size = int(len(data) * train_ratio)
val_size = int(len(data) * val_ratio)
test_size = int(len(data) * test_ratio)

train_features = image_features[:train_size]
val_features = image_features[train_size:train_size+val_size]
test_features = image_features[train_size+val_size:]

train_labels = labels[:train_size]
val_labels = labels[train_size:train_size+val_size]
test_labels = labels[train_size+val_size:]

# 定义CNN模型
cnn_input = Input(shape=(23, 1))
cnn_layer = Conv1D(64, 3, activation='relu')(cnn_input)
cnn_layer = MaxPooling1D(2)(cnn_layer)
cnn_layer = GlobalMaxPooling1D()(cnn_layer)
cnn_model = Model(inputs=cnn_input, outputs=cnn_layer)

# 定义BiGRU模型
rnn_input = Input(shape=(23,))
rnn_layer = Embedding(input_dim=256, output_dim=64)(rnn_input)
rnn_layer = Bidirectional(GRU(64, return_sequences=True))(rnn_layer)
rnn_layer = Dropout(0.2)(rnn_layer)
rnn_layer = Flatten()(rnn_layer)
rnn_model = Model(inputs=rnn_input, outputs=rnn_layer)

# 进行特征融合
combined_input = Concatenate()([cnn_model.output, rnn_model.output])
combined_output = Dense(8, activation='softmax')(combined_input)
model = Model(inputs=[cnn_model.input, rnn_model.input], outputs=combined_output)

# 编译模型
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# 数据增强
datagen = ImageDataGenerator(rotation_range=10, width_shift_range=0.1, height_shift_range=0.1)

# 训练模型
early_stopping = EarlyStopping(patience=3, monitor='val_loss', restore_best_weights=True)
history = model.fit(datagen.flow(train_features, train_labels, batch_size=32),
                    steps_per_epoch=len(train_features) / 32, epochs=10,
                    validation_data=(val_features, val_labels), callbacks=[early_stopping])

# 在测试集上评估模型
test_loss, test_acc = model.evaluate([test_features, test_features], test_labels)
print('Test Loss:', test_loss)
print('Test Accuracy:', test_acc)

在上述代码中，使用了TensorFlow和Keras库来构建和训练模型。首先读取数据集并划分为训练集、验证集和测试集。然后定义了CNN和BiGRU模型，并将两者的输出进行特征融合。接着编译模型并使用数据增强技术进行训练。最后，在测试集上评估模型的性能。