Identify Aircraft Based On Limited Number Of Photos
This is an attempt to train a deep neural network to identify different models of commercial aircraft based on very few photos - without relying on any additional information (i.e., no EXIF data, no tags, etc.). The underlying technique can be used for other image classification tasks where we have a small training dataset.
As of date, I have curated a dataset of 5920 photos of 39 different models of commercial aircraft (TSV, Web). They were obtained from the Yahoo Flickr Creative Commons (YFCC) dataset, Wikipedia and other sources.
- In part 1, I train an image classifier (a convolutional neural network, aka CNN) to distinguish between photos of 4 different models of commercial aircraft. Validation accuracy 0.84-0.86, after 20 epochs. I use a VirtualBox VM with Keras and Theano installed - created using a Vagrant script (see repository).