Java Deep Learning Cookbook

Constructing hidden layers for a CNN

The input layers of a CNN produce abstract images and pass them to hidden layers. The abstract image features are passed from input layers to the hidden layers. If there are multiple hidden layers in your CNN, then each of them will have unique responsibilities for the prediction. For example, one of them can detect lights and dark in the image, and the following layer can detect edges/shapes with the help of the preceding hidden layer. The next layer can then discern more complex objects from the edges/recipes from the preceding hidden layer, and so on.

In this recipe, we will design hidden layers for our image classification problem.