Automatic identification of bird targets with radar via patterns produced by wing flapping

Area of Science:

• Radar ornithology
• Signal processing
• Machine learning

Background:

• Accurate bird identification using radar is crucial for migration studies, environmental impact assessments, and aviation safety.
• Radar signals contain distinct patterns related to bird wing flapping, differentiating them from insects and ground clutter.
• Expert labeling categorized radar data into BIRD, INSECT, CLUTTER, and UFO classes.

Purpose of the Study:

• To develop and evaluate an automated classification algorithm for recognizing bird targets in radar data.
• To assess the algorithm's generalization capabilities across different datasets representing various locations or times.
• To provide a foundation for a real-time bird identification radar system.

Main Methods:

• Extraction of signal intensity and wing flapping pattern variables using continuous wavelet transform.
• Development of predictive models utilizing support vector classifiers.
• Estimation of classification performance through cross-validation on four distinct datasets.

Main Results:

• Extremely to moderately high classification performance when training and testing on the same dataset.
• Sustained extremely to moderately high performance when training on one dataset and testing on others, indicating good generalization.
• Significant time savings in identifying birds within large radar signal collections.

Conclusions:

• The developed algorithm effectively automates bird identification from radar signals.
• The method demonstrates robust generalization, performing well across different datasets.
• This research marks a significant advancement towards developing a real-time radar system for bird identification.