Automatic detection and classification of baleen whale social calls using convolutional neural networks

Abstract

Passive acoustic monitoring has proven to be an indispensable tool for many aspects of baleen whale research. Manual detection of whale calls on these large data sets demands extensive manual labor. Automated whale call detectors offer a more efficient approach and have been developed for many species and call types. However, calls with a large level of variability such as fin whale (Balaenoptera physalus) 40 Hz call and blue whale (B. musculus) D call have been challenging to detect automatically and hence no practical automated detector exists for these two call types. Using a modular approach consisting of faster region-based convolutional neural network followed by a convolutional neural network, we have created automated detectors for 40 Hz calls and D calls. Both detectors were tested on recordings with high- and low density of calls and, when selecting for detections with high classification scores, they were shown to have precision ranging from 54% to 57% with recall ranging from 72% to 78% for 40 Hz and precision ranging from 62% to 64% with recall ranging from 70 to 73% for D calls. As these two call types are produced by both sexes, using them in long-term studies would remove sex-bias in estimates of temporal presence and movement patterns.

Deep learning algorithm outperforms experienced human observer at detection of blue whale D‐calls: a double‐observer analysis

“…Our results suggest that machine learning algorithms for detection of blue whale D-calls appear now to have advanced to the point where they can perform better, cheaper and more reliably than a human analyst-at least under the circumstances we tested. While other studies, such as (Rasmussen & Sirovi c, 2021) have hinted at this, our double-observer approach not only allowed us to unambiguously determine that the automated detector was superior, but also to better understand and quantify the reasons behind its advantage: superior performance at low and moderate SNR. The high performance and expedient operation will provide tangible improvements to the analysis of large Antarctic datasets such as those from the SOHN (Miller, Milnes, et al, 2021;Van Opzeeland et al, 2013).…”

“…Initially, it appeared that our DenseNet-automated detector, tested on an independent test dataset, may have had worse performance than other recent D-call detectors (e.g. Rasmussen & Sirovi c, 2021;Socheleau & Samaran, 2017). However, after adjudication and application of mark-recapture methods, it appeared that the performance of our detector was similar to, if not slightly better than, these previous methods.…”

“…In recent years, machine learning algorithms, such as deep‐learning, have shown great promise for fast and accurate analysis of PAM data (Bergler et al., 2019; Madhusudhana et al., 2021; Rasmussen & Širović, 2021; Shiu et al., 2020; Zhong et al., 2021). In other domains, these methods have developed to the point that they exceed the performance of human observers (e.g.…”

More than a whistle: Automated detection of marine sound sources with a convolutional neural network

“…We use EfficientNet B0, currently the smallest architecture available for off-the-shelf transfer learning (Tan and Le, 2019), to develop a computationally low-cost CNN model for multi-sound source detection,. Other architectures such as ResNet (He et al, 2016), VGG (Simonyan and Zisserman, 2014) and AlexNet (Krizhevsky et al, 2012) are common choices for transfer learning within marine mammal species detection and classification studies (Bergler et al, 2019;Rasmussen & S ̌irovic., 2021;Allen et al, 2021;Lu et al, 2021). These architectures possess more trainable parameters making them computationally more expensive, and studies have shown that larger networks do not always obtain higher accuracies (Bergler et al, 2019).…”

Applying Artificial Intelligence Methods to Detect and Classify Fish Calls from the Northern Gulf of Mexico

“…For example, Padovese et al [79] used image augmentation to generate synthetic calls to increase training data size resulting in increased classifier recall and precision for labeling North Atlantic right whale (Eubalaena glacialis) upcalls. Rasmussen and Širović [80] used scaling and translation augmentation to prevent their classifier from overfitting during the training process. Image augmentation was beneficial in this study because the number of images for training and testing was relatively small (<400 images) for each sound type; in fact, two of the six fish calls had <100 images each for training and testing (Figure 3).…”