def MEAN_Spot(opt): # channel 1 inputs1 = layers.Input(shape=(42,42,1)) conv1 = layers.Conv2D(3, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs1) bn1 = layers.BatchNormalization()(conv1) pool1 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(bn1) do1 = layers.Dropout(0.3)(pool1) # channel 2 inputs2 = layers.Input(shape=(42,42,1)) conv2 = layers.Conv2D(3, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs2) bn2 = layers.BatchNormalization()(conv2) pool2 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(bn2) do2 = layers.Dropout(0.3)(pool2) # channel 3 inputs3 = layers.Input(shape=(42,42,1)) conv3 = layers.Conv2D(8, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs3) bn3 = layers.BatchNormalization()(conv3) pool3 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(bn3) do3 = layers.Dropout(0.3)(pool3) # merge 1 merged = layers.Concatenate()([do1, do2, do3]) # interpretation 1 merged_conv = layers.Conv2D(8, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.1))(merged) merged_pool = layers.MaxPooling2D(pool_size=(2, 2), padding='same', strides=(2,2))(merged_conv) flat = layers.Flatten()(merged_pool) flat_do = layers.Dropout(0.2)(flat) # outputs outputs = layers.Dense(1, activation='linear', name='spot')(flat_do) #Takes input u, v, os model = keras.models.Model(inputs=[inputs1, inputs2, inputs3], outputs=[outputs]) model.compile( loss={'spot':'mse'}, optimizer=opt, metrics={'spot':tf.keras.metrics.MeanAbsoluteError()}, ) return model 更改模型加入CBAM模块

可以尝试在每个卷积层后加入CBAM模块，CBAM模块可以用来增强模型的注意力机制。

以下是修改后的代码：

from keras.layers import GlobalAveragePooling2D, Reshape, Dense, multiply, Permute

def channel_attention(input_feature, ratio=8):
    channel_axis = 1 if K.image_data_format() == "channels_first" else -1
    channel = input_feature._keras_shape[channel_axis]
    shared_layer_one = Dense(channel//ratio,
                             activation='relu',
                             kernel_initializer='he_normal',
                             use_bias=True,
                             bias_initializer='zeros')
    shared_layer_two = Dense(channel,
                             kernel_initializer='he_normal',
                             use_bias=True,
                             bias_initializer='zeros')
    avg_pool = GlobalAveragePooling2D()(input_feature)
    avg_pool = Reshape((1,1,channel))(avg_pool)
    assert avg_pool._keras_shape[1:] == (1,1,channel)
    avg_pool = shared_layer_one(avg_pool)
    assert avg_pool._keras_shape[1:] == (1,1,channel//ratio)
    avg_pool = shared_layer_two(avg_pool)
    assert avg_pool._keras_shape[1:] == (1,1,channel)
    max_pool = GlobalMaxPooling2D()(input_feature)
    max_pool = Reshape((1,1,channel))(max_pool)
    assert max_pool._keras_shape[1:] == (1,1,channel)
    max_pool = shared_layer_one(max_pool)
    assert max_pool._keras_shape[1:] == (1,1,channel//ratio)
    max_pool = shared_layer_two(max_pool)
    assert max_pool._keras_shape[1:] == (1,1,channel)
    cbam_feature = Add()([avg_pool,max_pool])
    cbam_feature = Activation('sigmoid')(cbam_feature)
    if K.image_data_format() == "channels_first":
        cbam_feature = Permute((3, 1, 2))(cbam_feature)
    return multiply([input_feature, cbam_feature])


def spatial_attention(input_feature):
    kernel_size = 7

    if K.image_data_format() == "channels_first":
        channel = input_feature._keras_shape[1]
        cbam_feature = Permute((2,3,1))(input_feature)
    else:
        channel = input_feature._keras_shape[-1]
        cbam_feature = input_feature

    avg_pool = Lambda(lambda x: K.mean(x, axis=3, keepdims=True))(cbam_feature)
    assert avg_pool._keras_shape[-1] == 1
    max_pool = Lambda(lambda x: K.max(x, axis=3, keepdims=True))(cbam_feature)
    assert max_pool._keras_shape[-1] == 1
    concat = Concatenate(axis=3)([avg_pool, max_pool])
    assert concat._keras_shape[-1] == 2
    cbam_feature = Conv2D(filters = 1,
                  kernel_size=kernel_size,
                  strides=1,
                  padding='same',
                  activation='sigmoid',
                  kernel_initializer='he_normal',
                  use_bias=False)(concat)
    assert cbam_feature._keras_shape[-1] == 1

    if K.image_data_format() == "channels_first":
        cbam_feature = Permute((3, 1, 2))(cbam_feature)

    return multiply([input_feature, cbam_feature])


def CBAM(input_feature):
    cbam_feature = channel_attention(input_feature)
    cbam_feature = spatial_attention(cbam_feature)
    return cbam_feature

def MEAN_Spot_CBAM(opt):     # channel 1
    inputs1 = layers.Input(shape=(42,42,1))
    conv1 = layers.Conv2D(3, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs1)
    bn1 = layers.BatchNormalization()(conv1)
    cbam1 = CBAM(bn1)
    pool1 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(cbam1)
    do1 = layers.Dropout(0.3)(pool1)
    
    # channel 2
    inputs2 = layers.Input(shape=(42,42,1))
    conv2 = layers.Conv2D(3, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs2)
    bn2 = layers.BatchNormalization()(conv2)
    cbam2 = CBAM(bn2)
    pool2 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(cbam2)
    do2 = layers.Dropout(0.3)(pool2)
    
    # channel 3
    inputs3 = layers.Input(shape=(42,42,1))
    conv3 = layers.Conv2D(8, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.001))(inputs3)
    bn3 = layers.BatchNormalization()(conv3)
    cbam3 = CBAM(bn3)
    pool3 = layers.MaxPooling2D(pool_size=(3, 3), padding='same', strides=(3,3))(cbam3)
    do3 = layers.Dropout(0.3)(pool3)
    
    # merge 1
    merged = layers.Concatenate()([do1, do2, do3])
    
    # interpretation 1
    merged_conv = layers.Conv2D(8, (5,5), padding='same', activation='relu', kernel_regularizer=l2(0.1))(merged)
    merged_cbam = CBAM(merged_conv)
    merged_pool = layers.MaxPooling2D(pool_size=(2, 2), padding='same', strides=(2,2))(merged_cbam)
    flat = layers.Flatten()(merged_pool)
    flat_do = layers.Dropout(0.2)(flat)
    
    # outputs
    outputs = layers.Dense(1, activation='linear', name='spot')(flat_do)
    
    #Takes input u, v, os
    model = keras.models.Model(inputs=[inputs1, inputs2, inputs3], outputs=[outputs])
    model.compile(
        loss={'spot':'mse'},
        optimizer=opt,
        metrics={'spot':tf.keras.metrics.MeanAbsoluteError()},
    )
    return model