1. Consider 1 convolutional layer and 1 pooling layer as 1 layer (because the pooling layer does not have parameters w and b)
Alternatively, multiple convolutional layers correspond to one pooling layer
2. The full link layer is the ordinary neural network layer
Finally, add a classifier (such as Softmax) to get the output
3. How to choose hyperparameters
Do not rely on your own settings, refer to the literature
4. The activation size should decrease slowly, not too fast