Wildlife surveillance using deep learning methods.

Area of Science:

• Artificial Intelligence
• Wildlife Biology
• Conservation Technology

Background:

• Monitoring wild animal behavior is crucial for conservation and managing human-wildlife conflicts.
• Traditional methods of screening camera surveillance data are time-consuming and costly.
• Automated species identification can significantly improve the efficiency of wildlife monitoring.

Purpose of the Study:

• To develop and evaluate a deep learning approach for automatic species identification from images and video data.
• To create a tool for distinguishing between different animal species, with a focus on badgers for disease management.
• To enable cost-effective and efficient screening of large visual datasets in wildlife research.

Main Methods:

• Utilized a dataset of 8,368 images of wild and domestic animals.
• Developed deep learning models for both binary classification (badgers vs. others) and multiclassification (six species).
• Implemented a detection process for identifying animals in video footage.

Main Results:

• Achieved high accuracy rates: 98.05% for binary classification and 90.32% for multiclassification.
• Successfully developed a deep learning framework for automated species identification in images.
• Created the first known video detection process for identifying specific animals of interest.

Conclusions:

• Deep learning offers a powerful and accurate solution for automated wildlife species identification.
• The developed algorithms have broad applications in wildlife monitoring and conservation efforts.
• This technology can significantly reduce the labor and cost associated with analyzing visual data.