Residual networks based classification of right whales in the ocean

Automatic classification and recognition of various species of whales is useful in the close monitoring of various species of whales for scientific purposes and for population count as well. Marine animals including whales play an integral part in balancing the ecosystem and if left unsupervised whaling and other fishing activities (legal or illegal) can lead certain species to extinction, thus causing an imbalance in the ecosystem. In this work, a modified Residual Network (ResNet) has been proposed for the classification of right whales amongst 50 different classes. ResNets with their skip-connections feature bypasses the problem of vanishing gradient which results from the use of CNN with many convolutional layers. The deep ResNet is composed of 72 layers and involves from 5 up to a maximum of 30 iterations for the purpose of classifying the right whales. The use of the proposed ResNet achieved a recognition rate of 92.15%, showing a very high accuracy of correct whale classification. The accuracy achieved in this paper is better than those reported in previous literature.