你可以通过创建一个自定义图层来实现这一点：

class L1NormalizationLayer(tf.keras.layers.Layer):
    def __init__(self, **kwargs):
        super(L1NormalizationLayer, self).__init__(**kwargs)

    def build(self, input_shape):
        # No trainable weights for this layer
        super(L1NormalizationLayer, self).build(input_shape)

    def call(self, inputs):
        # Calculate L1 norm along the last axis (assuming channels_last data format)
        l1_norm = tf.reduce_sum(tf.abs(inputs), axis=-1, keepdims=True)

        # Normalize each patch by dividing by the L1 norm
        normalized_inputs = inputs / l1_norm

        return normalized_inputs

    def compute_output_shape(self, input_shape):
        return input_shape

字符串