Using Pre-trained VGG16 for 128X128X1 images

Question

I have a grayscale dataset of image dimensions (128,128,1). I am trying to use the VGGG16 pre-trained model for classification. VGG16 input is (224,224,3). I am trying to add a 2Dconv layer before VGG16 layer (128,128,3). But I am unable to do that. I want to do the fine-tuning of weights during the training phase. The code is given below. Would appreciate the help. Thanks.

vgg_base = keras.applications.VGG16(weights="imagenet", include_top=True,
    input_tensor=Input(shape=(128, 128,3)))
vgg_base.trainable = False
input_tensor = Input(shape=(128,128,1) )
x = keras.layers.Conv2D(3,(3,3),padding='same')(input_tensor)
print(x.shape)
x = vgg_base(x)
x = keras.layers.GlobalAveragePooling2D()(x)
outputs = keras.layers.Dense(23)(x)
model = keras.Model(input_tensor, outputs)

The error that I am facing is given by:

ValueError                                Traceback (most recent call last)
<ipython-input-12-a7b1b7273d95> in <module>()
      1 
      2 vgg_base = keras.applications.VGG16(weights="imagenet", include_top=True,
----> 3     input_tensor=Input(shape=(128, 128,3)))
      4 vgg_base.trainable = False
      5 input_tensor = Input(shape=(128,128,1) )

2 frames
/usr/local/lib/python3.7/dist-packages/keras/backend.py in batch_set_value(tuples)
   4017   if tf.compat.v1.executing_eagerly_outside_functions():
   4018     for x, value in tuples:
-> 4019       x.assign(np.asarray(value, dtype=dtype_numpy(x)))
   4020   else:
   4021     with get_graph().as_default():

ValueError: Cannot assign value to variable ' fc1/kernel:0': Shape mismatch.The variable shape (8192, 4096), and the assigned value shape (25088, 4096) are incompatible.