EVALUATION OF DATASET DIVERSITY TOWARD THE PERFORMANCE OF TRANSFER LEARNING MODEL FOR BIRD IDENTIFICATION
Loading...
Files
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Malaysia Sarawak (UNIMAS)
Abstract
Description
The rapid advancement of machine learning techniques has significantly improved the accuracy of bird identification systems, particularly through the application of transfer learning. This project investigates the impact of dataset diversity, specifically defined as the presence or absence of image background elements on the performance of transfer learning models in bird identification tasks. By analyzing a single bird dataset with two variants (with background and without background), this project aims to establish a correlation between visual background and model efficacy. Convolutional neural networks (CNNs) that have already been trained, like EfficientNetB0, were used to refine models on datasets with various levels of diversity. The model’s performance will be evaluated using metrics including accuracy, precision, recall, and F1 score. This research highlights the importance of creating balanced and diverse datasets to enhance the effectiveness of transfer learning models in bird identification, thereby facilitating better wildlife monitoring and conservation initiatives.