Convolutional layers are part of the name of this neural network and form a very important part of the neural network architecture. It can be broadly explained as scanning across the image to find certain features. We create a small filter, which we then slide across the entire image according to our desired stride.

So, for example, the first cell of the output would be calculated by finding the Dot Product of our 3 x 3 filter with the top-left corner of our Image, as shown in the following diagram:

And if your stride was one, it would shift one column right and continue, as shown here:

This would then continue until we had our entire output.