A convolutional neural network trained to distinguish between commercial aircraft types by scraping images from Airliners.net. It achieved 93% accuracy across multiple aircraft types. The experiment also revealed how training data biases creep in: the model “learned” that a 747 is usually on the ground taxiing, because that’s what most of the photos showed. Built with Python and PyTorch.

The interesting part was visualizing what the network actually learned using class activation mapping, showing it focusing on the tail, engine, and gear, which are the same features a human spotter would use.