This is a project developed for the curricular unit of Soft Computing.
Classify the bird species that appear on an image based on the use of a Convolutional Neural Network model (CNN).
- Preparation and analysis of dataset;
- Training and validation of learning models, specifically Convolutional Neural Networks (CNN);
- Use of Genetic Algorithms (GA) for learning model hyper parameter optimization, structure optimization and loss function optimization.
The proposed dataset has the following features:
- Bird Species: 250
- Training Images: 35215 (not balanced, however has at least 100 training image files per species);
- Validation Images: 1250 (5 per species);
- Test Images: 1250 (5 per species);
- Images Size: 224 x 224 x 3 color channels in jpg format;
- Species gender: 80% of total images are of male while the remaining 20% are of female - the classifier may not perform as well on female specie images.