Hammerhead Shark Species Monitoring with Deep Learning

In this paper, we propose a new automated method based on deep convolutional neural networks to detect and track critically endangered hammerhead sharks in video sequences. The proposed method improved the standard YOLOv3 deep architecture by adding 18 more layers (16 convolutional and 2 Yolo layers), which increased the model performance in detecting the species under analysis at different scales. According to the frame analysis based validation, the proposed method outperformed the standard YOLOv3 model and was similar to the mask R-CNN model in terms of accuracy scores for the majority of inspected frames. Also, the mean of precision and recall on an experimental frames dataset formed using the 10-fold cross-validation method highlighted that the proposed method outperformed the remaining architectures, reaching scores of 0.99 and 0.93, respectively. Furthermore, the methods were able to avoid introducing false positive detection. However, they were unable to handle the problem of species occlusion. Our results indicate that the proposed method is a feasible alternative tool that could help to monitor relative abundance of hammerhead sharks in the wild.