使用传统交叉熵损失函数的Unet分割算法代码

以下是使用传统交叉熵损失函数的Unet分割算法的代码示例：

import tensorflow as tf
from tensorflow.keras.layers import Conv2D, MaxPooling2D, UpSampling2D, Concatenate
from tensorflow.keras import Input, Model
from tensorflow.keras.losses import SparseCategoricalCrossentropy

def unet(input_shape):
    inputs = Input(shape=input_shape)

    # Contracting path
    c1 = Conv2D(64, 3, activation='relu', padding='same')(inputs)
    c2 = Conv2D(64, 3, activation='relu', padding='same')(c1)
    p1 = MaxPooling2D()(c2)

    c3 = Conv2D(128, 3, activation='relu', padding='same')(p1)
    c4 = Conv2D(128, 3, activation='relu', padding='same')(c3)
    p2 = MaxPooling2D()(c4)

    c5 = Conv2D(256, 3, activation='relu', padding='same')(p2)
    c6 = Conv2D(256, 3, activation='relu', padding='same')(c5)
    p3 = MaxPooling2D()(c6)

    c7 = Conv2D(512, 3, activation='relu', padding='same')(p3)
    c8 = Conv2D(512, 3, activation='relu', padding='same')(c7)

    # Expansive path
    u1 = UpSampling2D()(c8)
    c9 = Conv2D(256, 3, activation='relu', padding='same')(u1)
    c10 = Conv2D(256, 3, activation='relu', padding='same')(Concatenate()([c6, c9]))

    u2 = UpSampling2D()(c10)
    c11 = Conv2D(128, 3, activation='relu', padding='same')(u2)
    c12 = Conv2D(128, 3, activation='relu', padding='same')(Concatenate()([c4, c11]))

    u3 = UpSampling2D()(c12)
    c13 = Conv2D(64, 3, activation='relu', padding='same')(u3)
    c14 = Conv2D(64, 3, activation='relu', padding='same')(Concatenate()([c2, c13]))

    # Output layer
    outputs = Conv2D(1, 1, activation='sigmoid')(c14)

    model = Model(inputs=inputs, outputs=outputs)
    model.compile(optimizer='adam', loss=SparseCategoricalCrossentropy(from_logits=True), metrics=['accuracy'])

    return model

在这个例子中，我们使用了TensorFlow 2.x构建了一个基本的Unet模型，其中包含了多个卷积和池化层的收缩路径，以及多个上采样和卷积层的扩张路径。模型的输出是一个2D的分割图像，通过使用sigmoid激活函数对像素进行二元分类（前景和背景）。此外，我们使用了传统的交叉熵损失函数，以及准确率作为评估指标。