Classifying Invasive Alien Species in the Philippines Using Image Processing - Policy-Based Data Augmentation With Convolutional Neural Networks
Date of Award
12-1-2022
Document Type
Thesis
Degree Name
Doctor of Philosophy in Computer Science
First Advisor
Proceso L. Fernandez, Jr., PhD
Abstract
The proliferation of Invasive Alien Species (IAS) in the is a major threat to its biodiversity. Towards reducing such threat, image processing technology can be applied to collect taxonomic information which may then assist in strategies and plans to fight IAS. This interdisciplinary study first developed two image datasets of IAS in the Philippines (PH). The First dataset had twenty-four IAS PH classes. The second dataset included the known five invasive frog species in the Philippines. A sixth class of endemic amphibians was added to the second dataset to distinguish endemic frogs from invasive frog species. Four convolutional neutral network (CNN) techniques namely ResNet18, MobileNetv2, GoogLenet and DenseNet were then applied on the two datasets. The four models generated were established as the baseline to improve upon. Precision, recall, fl-score accuracy metrics were taken. The study then used a policy-based data augmentation implementation of ImageNet's augmentation policies learned by Google's Auto Augment to increase the images with the aim of improving accuracy. The four CNN models were re-trained using the augmented training set and tested on the same testing set as the baseline. Five-fold cross validation was used to evaluate the performance of the models both before and after augmentation. Results showed improvement across models. A paired t-test on the accuracy metrics of the baseline and the augmented models, however, showed significant improvement using policy-based data augmentation only on the DendeNet model. A prototype web application was developed to deploy the best performing model.
Recommended Citation
Aliño, Rey R., (2022). Classifying Invasive Alien Species in the Philippines Using Image Processing - Policy-Based Data Augmentation With Convolutional Neural Networks. Archīum.ATENEO.
https://archium.ateneo.edu/theses-dissertations/968
