Arctic mammal classification

Deep learning model for acoustic detection and classification of Bowhead whales. This repo contains code for the binary classification model, two classes are: Bowhead whales (BH), Other/background.

Following are some statistics about the dataset and model:

• CB-50 and CB-300 dataset for training and testing
• CB-50 contains 11138 BH annotations, CB-300 contains 5983 BH annotations
• Average duration of BH annotations: 1.78 sec
• Average low and high frequency of BH: 90.38 - 530.80 Hz
• Spectrogram window size: 3 sec
• DenseNet model

Training data distributation

Overall workflow diagram

Results

Training & validation performance

Following image shows the training and validation loss curves

Test performance

Following graph shows the performance metrices on the test annotations