ISAJ

1
def CNN(image_shape, soil_pm_shape, is_in=0):
2
    img_input = tf.keras.Input(shape=image_shape)
3
    soil_pm_input = tf.keras.Input(shape=soil_pm_shape)
4

5
    x = tf.keras.layers.Conv2D(filters=16, kernel_size=3, strides=2)(img_input)
6
    x = tf.keras.layers.BatchNormalization()(x)
7
    x = tf.keras.layers.Activation('relu')(x)
8

9
    x = tf.keras.layers.MaxPooling2D(pool_size=2, strides=2)(x)
10
    x = tf.keras.layers.BatchNormalization()(x)
11
    x = tf.keras.layers.Activation('relu')(x)
12

13
    x = tf.keras.layers.Conv2D(filters=32, kernel_size=3, strides=2)(x)
14
    x = tf.keras.layers.BatchNormalization()(x)
15
    x = tf.keras.layers.Activation('relu')(x)
16

17
    x = tf.keras.layers.MaxPooling2D(pool_size=2, strides=2)(x)
18
    x = tf.keras.layers.BatchNormalization()(x)
19
    x = tf.keras.layers.Activation('relu')(x)
20

21
    x = tf.keras.layers.Flatten()(x)
22

23
    new_x = tf.keras.layers.concatenate([x, soil_pm_input], axis=1)
24

25
    x_1 = tf.keras.layers.Dense(256)(new_x) if is_in!=1 else tf.keras.layers.Dense(64)(new_x)
26
    x_1 = tf.keras.layers.BatchNormalization()(x_1)
27
    x_1 = tf.keras.layers.Activation('relu')(x_1)
28

29
    x_1 = tf.keras.layers.Dense(186)(x_1) if is_in!=1 else tf.keras.layers.Dense(2)(x_1)
30
    x_1 = tf.keras.layers.BatchNormalization()(x_1)
31
    out_put = tf.keras.layers.Activation('sigmoid')(x_1)
32

33
    model = tf.keras.Model(inputs=[img_input, soil_pm_input], outputs=out_put)
34
    return model

1
model.compile(
2
        optimizer=tf.keras.optimizers.Adam(1e-3),
3
        loss='mean_squared_error'
4
    )
5
    model.fit(
6
        x=train_pm,
7
        y=train_bg,
8
        validation_data=[confirm_pm, confirm_bg],
9
        batch_size=batch_size,
10
        epochs=epochs,
11
        verbose=1
12
    )