This is the main type of layer; the use of one or more of these layers in a CNN is essential. The parameters of a convolutional layer, in practice, relate to a set of workable filters. Each filter is spatially small, along the width and height dimensions, but it extends over the entire depth of the input volume to which it is applied.
Unlike normal neural networks, convolutional layers have neurons organized in three dimensions: width, height, and depth. They are shown in the following figure:
During forward propagation, each filter is translated—or more precisely, convolved—along the width and height of the input volume, ...
