用keras实现ESA_block。要求：用def的形式实现，不要用class

以下是用 Keras 实现 ESA_block 的代码，采用了 def 的形式实现：

from keras.layers import Conv2D, BatchNormalization, Activation, Add

def ESA_block(x, filters, kernel_size=3, strides=1):
    # convolutional block
    conv1 = Conv2D(filters, kernel_size, strides=strides, padding='same')(x)
    conv1 = BatchNormalization()(conv1)
    conv1 = Activation('relu')(conv1)
    
    # spatial attention block
    conv2 = Conv2D(filters, kernel_size=1, strides=1, padding='same')(conv1)
    conv2 = BatchNormalization()(conv2)
    conv2 = Activation('sigmoid')(conv2)
    x = x * conv2
    
    # channel attention block
    avg_pool = keras.layers.GlobalAveragePooling2D()(conv1)
    max_pool = keras.layers.GlobalMaxPooling2D()(conv1)
    pool = Add()([avg_pool, max_pool])
    pool = Dense(filters // 4, activation='relu')(pool)
    pool = Dense(filters, activation='sigmoid')(pool)
    pool = keras.layers.Reshape((1, 1, filters))(pool)
    x = x * pool
    
    # residual connection
    x = Add()([x, conv1])
    
    return x

其中，ESA_block 包含两个子块：Spatial Attention Block 和 Channel Attention Block。Spatial Attention Block 用于增强空间维度上的信息交互，Channel Attention Block 用于增强通道维度上的信息交互。最后再将两个子块的结果进行残差连接。