This Convolutional Neural Network model attempts to identify if there is a bird in the provided image. This model was trained using CIFAR-10, a free 32x32 image dataset that can be found here.
The idea and code for this model is provided by the textbook Machine Learning is Fun! 2nd Edition by Adam Geitgey.
This model works by first separating images into a bunch of tiny, overlapping chunks. The pixel data from these small chunks is used to try and identify small parts of a potentially larger pattern.
Throughout this step and the following steps, the resulting data from each step of the neural network is "normalized" to always remain between 0 and 1.
Once the neural networks of the model figure out which segments of the image are likely to be part of a bird, the resulting list of numbers is then downsampled (in this case, down to 1/4 of its original size) by only keeping the most interesting of each 2x2 grid of chunks.
The model can then use the results of the previous neural layers to see if their overall results resemble what the model knows as a bird. Their closeness to known patterns determines the model's predicted probability that the provided image is a bird.