KERAS: CNN模型的预训练，并将其用于CNN-LSTM模型

from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, LSTM, Dense

cnn_model = Sequential()
cnn_model.add(Conv2D(filters=32, kernel_size=(3, 3), activation='relu', input_shape=(image_width, image_height, channels)))
cnn_model.add(MaxPooling2D(pool_size=(2, 2)))
cnn_model.add(Flatten())

cnn_model.load_weights('pretrained_cnn_model.h5')

cnn_lstm_model = Sequential()
cnn_lstm_model.add(TimeDistributed(cnn_model, input_shape=(time_steps, image_width, image_height, channels)))
cnn_lstm_model.add(LSTM(units=64))
cnn_lstm_model.add(Dense(units=num_classes, activation='softmax'))

cnn_lstm_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])
cnn_lstm_model.fit(x_train, y_train, batch_size=batch_size, epochs=epochs)