layers.GlobalAveragePooling1D()的作用：接入Dense层之前，一定要把数据打平！

self.avgpool = layers.GlobalAveragePooling2D()可以用来打平！
如果你的数据是3D的，要用 layers.GlobalAveragePooling1D()！效果：
TensorShape([8, 200, 256]) ===> TensorShape([8, 256])

   # self.final_bn  = layers.BatchNormalization()
    # 接入Dense层之前，先打平！！！
    self.avgpool = layers.GlobalAveragePooling2D()
    self.fc1 = layers.Dense(num_classes*100)
    self.fc2 = layers.Dense(1, activation=tf.nn.sigmoid)

也可以用layers.Flatten()