Understanding the convolutional neural network layer

Now, let’s focus on the foundations of the convolutional layer, starting with Figure 3.1, which shows the operational process of a convolutional filter. A filter is a small matrix of weights that’s used to extract features or patterns from an input array. A convolutional filter is a type of filter that slides over an image, performing convolution operations to extract features by calculating dot products:

Figure 3.1 – Operation of a convolutional filter on an image of a t-shirt from the Fashion MNIST dataset

Convolutional layers are made out of multiple convolutional filters of the same sizes. Convolutional filters are the main pattern detectors in a CNN, where each filter will learn to identify multidimensional patterns that exist in an image. The patterns can range from low-level patterns such as lines and edges to mid-level patterns such as circles or squares and finally to high...