帮我用keras改写

好的，以下是使用Keras改写的代码：

from keras.layers import Layer, average, GlobalAveragePooling2D, GlobalMaxPooling2D, Reshape, Dense, multiply, Permute, Lambda
import math

class ChannelAttentionModuleConv(Layer):
    def __init__(self, gamma=2, b=1, **kwargs):
        super(ChannelAttentionModuleConv, self).__init__(**kwargs)
        self.gamma = gamma
        self.b = b

    def build(self, input_shape):
        channels = input_shape[3]
        kernel_size = int(abs((math.log(channels, 2) + self.b) / self.gamma))
        kernel_size = kernel_size if kernel_size % 2 else kernel_size + 1
        self.avg_pooling = GlobalAveragePooling2D()
        self.max_pooling = GlobalMaxPooling2D()
        self.conv = Dense(1, kernel_size, padding='same', use_bias=False, activation='sigmoid')
        super(ChannelAttentionModuleConv, self).build(input_shape)

    def call(self, inputs):
        avg_x = self.avg_pooling(inputs)
        max_x = self.max_pooling(inputs)
        avg_out = self.conv(avg_x)
        max_out = self.conv(max_x)
        out = average([avg_out, max_out])
        out = multiply([inputs, out])
        return out

    def compute_output_shape(self, input_shape):
        return input_shape

这里我们使用了Keras中的一些常用层，如GlobalAveragePooling2D、GlobalMaxPooling2D、Dense等，同时也使用了一些Keras中的实用功能，如average、multiply。注意，我们将卷积操作改为了全连接层，并将输出的激活函数设置为sigmoid。