This project leverages the power of deep learning to address critical challenges in wildlife conservation. By automating the classification of camera trap images, this work aims to support ecologists in monitoring species and preserving biodiversity more efficiently. This research was recognized for its innovative approach and accepted at the ECML-PKDD 2023: Joint European Conference on Machine Learning and Knowledge Discovery in Databases.
Key Highlights:
Real-World Application: Developed a solution for classifying images from camera traps deployed in the Hajar Mountains, UAE, to assist in tracking and studying wildlife populations.
Advanced Deep Learning Models: Evaluated six pre-trained models, including DenseNet-121 and Xception, achieving a weighted accuracy of 72.72% and an F1 score of 0.957 with the best-performing model.
Overcoming Data Challenges: Tackled significant class imbalances in the dataset through innovative techniques, ensuring accurate classification across diverse species, including underrepresented ones.
Robust Validation: Used Adjusted Rand Index (ARI) scores and t-SNE visualizations to demonstrate the models’ generalizability and ability to classify unseen data effectively.
User-Friendly Application: Created a graphical user interface (GUI) to make this technology accessible to ecologists and non-technical users, streamlining conservation workflows with rapid image classification.
Impactful Results: Enabled the automatic identification of species, significantly reducing the time and effort required for manual labeling, which often takes up 80% of research efforts.
Why It Matters:
This project bridges cutting-edge AI with ecological preservation, providing a scalable and efficient tool to aid wildlife conservation efforts. By empowering ecologists with technology that saves time and enhances accuracy, this work contributes to safeguarding biodiversity and addressing global environmental challenges.