Machine Learning with Apache Spark Quick Start Guide

Convolution layers

The next transformations that occur in the CNN are processed in convolution layers. The purpose of the convolution layers is to detect features in the image, which is achieved through the use of filters (also called kernels). Imagine taking a magnifying glass and looking at an image, starting at the top-left of the image. As we move the magnifying glass from left to right and top to bottom, we detect the different features in each of the locations that our magnifying glass moves over. At a high level, this is the job of the convolution layers, where the magnifying glass represents the filter or kernel and the size of each step that the filter takes, normally pixel by pixel, is referred to as the stride size. The output ...