Enhancing Bird Species Identification Using a Custom CNN Architecture

Bird watching at zoos can be a fun activity that frequently sparks interest in the different species. Information boards on bird cages usually provide information, however occasionally these boards are broken, lose their color, or only include a little amount of information. This may make it more difficult for a visitor to discover more about the birds. While it is a possibility, hiring a local guide might be expensive. In order to solve this problem, we suggest a method in which users only need to snap a photo of a bird and upload it to a website in order to get detailed species information. In order to do this, we used the Google Chrome plugin Fatkun to manually gather pictures of 15 different bird species, resulting in a collection of 4,447 photos from 15 different classes. With the use of this dataset, we trained a bespoke Convolutional Neural Network (CNN) model, yielding testing, validation, and training accuracies of 89.50%, 92.53%, and 89.58%, respectively.