1D-CNN_1D-CNN_1DCNN_CNN_CNN-_CNN时间序列_

from keras.layers import merge, Conv1D, Input, MaxPool1D, Dense, Dropout, GlobalAveragePooling1D from keras.models import Model,load_model,optimizers from keras.callbacks import ModelCheckpoint import numpy as np model_path = r"F:\Project&Learn\肌电信号\1D-CNN\model//" # 传感器数量 num_senor = 6 # 周期采样数 time_periods = 100 # 分类个数 num_class = 4 def build_1d_cnn(): input_layer = Input(shape=(time_periods, num_senor)) # block_1 x = Conv1D(8, 3, activation='relu', name='block_1_1')(input_layer) x = Conv1D(8, 3, activation='relu', name='block_1_2')(x) # block_2 a = Conv1D(8, 1, padding='same', activation='relu', name='block_2_a')(x) b = Conv1D(8, 1, padding='same', activation='relu', name='block_2_b')(x) b = Conv1D(8, 3, padding='same', activation='relu', name='block_2_b2')(b) c = Conv1D(8, 1, padding='same', activation='relu', name='block_2_c')(x) c = Conv1D(8, 5, padding='same', activation='relu', name='block_2_c2')(c) d = MaxPool1D(3, padding='same', strides=1, name='block_2_d')(x) d = Conv1D(8, 1, padding='same', activation='relu', name='block_2_d2')(d) x = merge([a, b, c, d], mode="concat", concat_axis=2) # block_3 x = MaxPool1D(3, name='block_3_0')(x) a = Conv1D(16, 1, padding='same', activation='relu', name='block_3_a')(x) b = Conv1D(16, 1, padding='same', activation='relu', name='block_3_b')(x) b = Conv1D(16, 3, padding='same', activation='relu', name='block_3_b2')(b) c = Conv1D(16, 1, padding='same', activation='relu', name='block_3_c')(x) c = Conv1D(16, 5, padding='same', activation='relu', name='block_3_c2')(c) d = MaxPool1D(3, padding='same', strides=1, name='block_3_d')(x) d = Conv1D(16, 1, padding='same', activation='relu', name='block_3_d2')(d) x = merge([a, b, c, d], mode="concat", concat_axis=2) # output_layer x = MaxPool1D(2, padding='same', strides=2, name='op_1')(x) x = Dropout(0.5, name='op_2')(x) x = GlobalAveragePooling1D(name='op_3')(x) x = Dropout(0.5, name='op_4')(x) output_layer = Dense(num_class, activation='softmax', name='op_5')(x) model = Model(input=input_layer, outputs=output_layer) # print(model.summary()) return model def fit_with_cnn_model(x_train, y_train, my_model): # 回调函数 callbacks_list = [ ModelCheckpoint( filepath=model_path+'event_model.{epoch:02d}-{val_loss:.2f}.h5', monitor='val_loss', save_best_only=False, period=500 ) ] model.compile(loss='categorical_crossentropy', optimizer=optimizers.Adam(lr=1e-3,decay=1e-5), metrics=['accuracy']) Batch_size = 32 Epochs = 4000 history = my_model.fit( x_train, y_train, batch_size=Batch_size, epochs=Epochs, callbacks=callbacks_list, validation_split=0.2, verbose=1 ) model.save("model.h5") def get_label(index): label = np.zeros(num_class) label[index] = 1 return label def data_input(): # load label Y = [] for i in range(num_class): for j in range(500): label = get_label(i) Y.append(label) Y = np.array(Y) # load data # a = np.load("X.npy") X = np.load("X.npy") # X = np.append(a,X, axis=0) index = np.arange(2000) np.random.shuffle(index) X_train = X[index] Y_train = Y[index] return X_train, Y_train if __name__ == '__main__': model = build_1d_cnn() # model = load_model("model2.h5") X, Y = data_input() fit_with_cnn_model(X, Y, model)