PhotoId-Whale: Blue whale dorsal fin classification for mobile devices

Area of Science:

• Marine biology
• Computer vision
• Ecological monitoring

Background:

• Manual photo-identification (photo-id) in field studies introduces subjectivity and potential inaccuracies.
• Automated, portable, and unbiased photo-id systems are crucial for efficient animal monitoring.
• Current methods face challenges with data accuracy due to human criteria in classification.

Purpose of the Study:

• To develop an autonomous and portable photo-identification system for blue whale dorsal fins.
• To optimize image classification algorithms for high statistical dependence and offline processing on mobile platforms.
• To create a bias-free, user-friendly mechanism for ecological risk prediction and prevention.

Main Methods:

• Utilized Scale Invariant Feature Transform (SIFT) for image feature extraction.
• Employed statistical discriminators (variance, standard deviation) to refine extracted data.
• Focused on classifying dorsal fin images of blue whales using offline processing on a mobile platform (Jetson Tegra TK1).

Main Results:

• Qualitatively and quantitatively validated the photo-id method's sensitivity, specificity, and accuracy.
• Optimized SIFT algorithm to balance results with computational cost.
• Demonstrated effective elimination of external factors affecting image quality.

Conclusions:

• The developed system offers an economical and portable solution for blue whale monitoring.
• This technology enhances the reliability of field study data for scientists and the public.
• The system facilitates improved ecological risk assessment through accurate, automated animal identification.