CNN模型训练报错：conv2d_6层输入不兼容问题求助

问题描述 #

训练CNN模型对1896个文件的二分类数据集进行分类时，训练阶段出现输入维度不匹配报错。

定义的参数 #

IMAGE_SIZE_1 = 1280
IMAGE_SIZE_2 = 720
BATCH_SIZE=30
EPOCHS = 20
CHANNELS = 2

预处理与数据增强代码 #

resize_and_rescale = tf.keras.Sequential([
    layers.experimental.preprocessing.Resizing(IMAGE_SIZE_1,IMAGE_SIZE_2),
    layers.experimental.preprocessing.Rescaling(1.0/255)
])

data_argumentation = tf.keras.Sequential([
    layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),
    layers.experimental.preprocessing.RandomRotation(0.2)
])

模型构建代码 #

input_shape = ( BATCH_SIZE,IMAGE_SIZE_1, IMAGE_SIZE_2, CHANNELS)
n_classes = 2

model = models.Sequential([
    resize_and_rescale,
    data_argumentation,
    #Input layer
    layers.Conv2D(35, (3,3), activation="relu", input_shape = input_shape),
    layers.MaxPooling2D((2,2)),
    #Pooling
    layers.Conv2D(64, (3,3), activation="relu"),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64, (3,2), activation="relu"),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64, (3,3), activation="relu"),
    layers.MaxPooling2D((2,2)),
   
    #Flatenning
    layers.Flatten(),
    layers.Dense(64,activation="relu"),
    #Output LAyer
    layers.Dense(n_classes, activation="softmax")
])

编译与训练代码 #

model.compile(
optimizer="adam",
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=["accuracy"])

history = model.fit(
train_ds,
epochs=EPOCHS,
verbose=1,
batch_size = BATCH_SIZE,
validation_data=val_ds)

报错信息 #

ValueError: Exception encountered when calling layer 'sequential_3' (type Sequential).

Input 0 of layer "conv2d_6" is incompatible with the layer: expected axis -1 of input shape to have value 2, but received input with shape (None, 1280, 720, 3)

Call arguments received by layer 'sequential_3' (type Sequential):
• inputs=tf.Tensor(shape=(None, 1280, 720, 3), dtype=float32)
• training=True
• mask=None

报错原因 #

1. 通道数不匹配：你定义CHANNELS=2，但实际输入的是3通道RGB图像，导致Conv2D层期望输入通道为2，却接收到3通道数据。
2. input_shape定义错误：Keras中input_shape不需要包含批量大小（BATCH_SIZE），正确格式应为(高, 宽, 通道数)，错误加入BATCH_SIZE会导致输入维度校验失败。

修复方案 #

1. 修正input_shape定义 #

移除input_shape中的BATCH_SIZE，同时根据实际数据调整CHANNELS参数：

# 如果是RGB图像，将CHANNELS改为3；如果是自定义2通道数据，保持CHANNELS=2
CHANNELS = 3
input_shape = (IMAGE_SIZE_1, IMAGE_SIZE_2, CHANNELS)

2. 规范模型输入层写法 #

建议在模型开头显式添加Input层，替代在Conv2D层中指定input_shape，让输入逻辑更清晰：

model = models.Sequential([
    layers.Input(shape=input_shape),  # 显式指定输入形状
    resize_and_rescale,
    data_argumentation,
    layers.Conv2D(35, (3,3), activation="relu"),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64, (3,3), activation="relu"),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64, (3,2), activation="relu"),
    layers.MaxPooling2D((2,2)),
    layers.Conv2D(64, (3,3), activation="relu"),
    layers.MaxPooling2D((2,2)),
   
    layers.Flatten(),
    layers.Dense(64,activation="relu"),
    layers.Dense(n_classes, activation="softmax")
])

3. 确认数据集加载的通道数 #

检查数据集加载代码，确保加载的图像通道数与CHANNELS一致：

# 示例：加载RGB图像（3通道）
train_ds = tf.keras.utils.image_dataset_from_directory(
    train_dir,
    image_size=(IMAGE_SIZE_1, IMAGE_SIZE_2),
    batch_size=BATCH_SIZE,
    color_mode="rgb"  # 若为灰度图，改为"grayscale"对应1通道
)

内容的提问来源于stack exchange，提问作者Rafael Vieira