Using local binary patterns as features for classification of dolphin calls

Area of Science:

• Marine Biology
• Bioacoustics
• Signal Processing

Background:

• Dolphin vocalization classification is crucial for understanding marine mammal communication.
• Traditional methods like Time-Frequency Parameters (TFP) rely on contour analysis, which can be complex.
• Image processing offers potential for more robust feature extraction.

Purpose of the Study:

• To evaluate the effectiveness of Local Binary Patterns (LBP) for dolphin vocalization classification.
• To compare the performance of LBP with the established TFP approach.
• To determine if LBP can improve classification accuracy for complex dolphin whistles.

Main Methods:

• Utilized the Local Binary Patterns (LBP) image processing technique to generate feature vectors.
• Compared LBP performance against the contour-based Time-Frequency Parameters (TFP) method.
• Conducted experimental classification of dolphin whistle types.

Main Results:

• Local Binary Patterns (LBP) demonstrated consistent classifier accuracy for dolphin whistle calls.
• LBP outperformed TFP in classifying complex whistle shapes.
• The LBP operator effectively handles vocalizations with impulsive clicks and anthropogenic harmonics.

Conclusions:

• Local Binary Patterns (LBP) is a promising technique for dolphin bioacoustics analysis.
• LBP offers a simplified and more robust alternative to traditional contour-based methods.
• This approach enhances the accuracy of classifying diverse dolphin vocalizations.