Python Tensorflow-Conv神经网络的定值误差与报警

我正在使用Tensorflow创建一个CNN模型，该模型将尺寸为124129的图像分为8个类别

我需要帮助理解为什么会出现错误：

ValueError:“图像”必须具有3维或4维。

我也收到了警告

WARNING:tensorflow:Model was constructed with shape (1, 124, 129, 8) for input KerasTensor(type_spec=TensorSpec(shape=(1, 124, 129, 8), dtype=tf.float32, name='input_36'), name='input_36', description="created by layer 'input_36'"), but it was called on an input with incompatible shape (None, 129).

就在我尝试将模型拟合到训练集时出现错误之前

以下是模型的代码：

from tensorflow.keras import layers
from tensorflow.keras import models

for spectrogram, _ in training_spect_data.take(1):
  input_shape = spectrogram.shape

print(input_shape)
print(len(commands))

model = models.Sequential([
    layers.Input((124,129,8), batch_size= 1),
    layers.experimental.preprocessing.Resizing(32, 32), 
    layers.Conv2D(32, 3, activation='relu'),
    layers.Conv2D(64, 3, activation='relu'),
    layers.MaxPooling2D(),
    layers.Flatten(),
    layers.Dense(128, activation='relu'),
    layers.Dense(num_labels),
])

model.summary()
model.compile(
    optimizer=tf.keras.optimizers.Adam(),
    loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
    metrics=['accuracy'],
)

model.fit(
    training_spect_data, 
    validation_data=validation_spect_data,  
    epochs=10,
  callbacks=tf.keras.callbacks.EarlyStopping(verbose=1, patience=2),
)

如果有帮助，这就是training/test/val集中的一个数据点的样子：

tf.Tensor(
[[4.22809680e-04 1.20909873e-03 1.30543171e-03 ... 1.11539455e-04
  7.03251426e-05 5.72346325e-05]
 [1.37844472e-06 5.68726333e-04 1.01903011e-03 ... 1.72739034e-04
  7.02477628e-05 2.15965847e-05]
 [1.90013321e-04 5.55736362e-04 7.45545258e-04 ... 1.08729822e-04
  1.73325971e-04 1.51859131e-04]
 ...
 [1.93573331e-04 5.46126859e-04 1.61838590e-03 ... 1.15362825e-04
  1.83291835e-04 2.17455061e-04]
 [1.49126354e-04 7.04471953e-04 1.06320635e-03 ... 8.47642514e-05
  3.19860228e-05 1.25371589e-05]
 [1.29039981e-05 2.79012456e-04 5.54071739e-04 ... 3.47834612e-05
  7.82721399e-05 7.47569429e-05]], shape=(124, 129), dtype=float32) tf.Tensor(b'yes', shape=(), dtype=string)

非常感谢您为解决上述错误/警告提供的任何帮助。

您已经掌握了代码

 layers.Input((124,129,8), batch_size= 1)

我认为应该是这样

 layers.Input((124,129), batch_size= 1)

8是与输入形状无关的类数。实际上，我也会省略batch_size参数，以便使用

 layers.Input((124,129))

model.fit将默认批次大小设置为32。您还可以将其作为模型中的最后一层

layers.Dense(num_labels),

我想你最好用

layers.Dense(num_label, activation='softmax')

然后将损失从

loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)

到

如果使用SparseCategoricalCrossentropy，请确保标签是整数

loss=tf.keras.losses.SparseCategoricalCrossentropy