concept CNN architecture in category deep learning

This is an excerpt from Manning's book Deep Learning for Vision Systems.
Just like how regular neural networks contain multiple layer which allow each layer to find successively complex features, CNNs work the same way. The first layer of convolutions learns some basic features (edges and lines), then the next layer will learn features that are a little more complex (circles, squares, etc), and then the following layer will find even more complex features (like parts of the face, car wheel, dog whiskers, etc.) and so on. You will see this demonstrated real soon. For now, know that the CNN architecture will follow the same pattern we started with neural networks: we will stack neurons in hidden layers on top of each other, we will have weights that will be randomly initiated and learned during the network training, we will apply activation functions, calculate error (y - ŷ) then backpropagate the error to update the weights. All the same. The difference is that we will use convolutional layers instead of regular fully connected layers for the feature learning piece.
Before we look into the CNN architecture, let’s back up for a moment to see the big picture. Remember the image classification pipeline we discussed in chapter 1?