Patch based image training and combine their probability from an image

Question

Firstly, I have implemented a simple VGG16 network for image classification.

model = keras.applications.vgg16.VGG16(include_top = False,