Underwater Live Fish Recognition by Deep Learning

Abstract

Recently, underwater videos have gained great interest by marine ecologists for studying fish populations. Actually, this technique produces large amount of visual data and does not affect fish behavior. However, visual processing and analyzing of the recorded data can be subjective, time consuming and costly. We propose in this paper to use the convolutional neural network AlexNet with transfer learning for automatic fish species classification. We extract features from foreground fish images of the available underwater dataset using the pretrained AlexNet network either with or without fine-tunig. For classification, we use a linear SVM classifier. The experiment results demonstrate the effectiveness of the proposed approach on the Fish Recognition Ground-Truth dataset. We achieve an accuracy of 99.45%.