A Survey of Underwater Acoustic Target Recognition Methods Based on Machine Learning

Journal of Marine Science and Engineering, Journal Year: 2023, Volume and Issue: 11(2), P. 384 - 384

Published: Feb. 9, 2023

Underwater acoustic target recognition (UATR) technology has been implemented widely in the fields of marine biodiversity detection, search and rescue, seabed mapping, providing an essential basis for human economic military activities. With rapid development machine-learning-based acoustics field, these methods receive wide attention display a potential impact on UATR problems. This paper reviews current based machine learning. We focus mostly, but not solely, target-radiated noise from passive sonar. First, we provide overview underwater acquisition process briefly introduce classical signal feature extraction methods. In this paper, are classified learning algorithms used as technologies using statistical methods, deep models, transfer data augmentation UATR. Finally, challenges method summarized directions future put forward.

Language: Английский

---

Reducing Cetacean Interactions With Bottom Set‐Nets and Purse Seining Using Acoustic Deterrent Devices in Southern Iberia

Aquatic Conservation Marine and Freshwater Ecosystems, Journal Year: 2025, Volume and Issue: 35(2)

Published: Feb. 1, 2025

ABSTRACT In southern Iberia (NE Atlantic), cetacean bycatch is reported in several fisheries, whereas depredation by bottlenose dolphin ( Tursiops truncatus ) commonly observed bottom set‐net fisheries. This study tested the effectiveness of acoustic deterrent devices discouraging small cetaceans from approaching set‐nets and purse seine to reduce interactions. The used were interactive for fishery common Delphinus delphis bycatch. Data collection was carried out at‐sea observers trained fishing vessel crew observers. Hauls with without compared analysed investigate differences catch per unit effort, factors affecting interaction, probability interaction habituation (in only). set‐nets, rate significantly lower reduced about 50% hauls using devices. Habituation dolphins but gradual. fishery, 100% when Overall, results are promising, different reduction efficiencies between gear types indicate that potential application should be considered on a métier‐by‐métier basis. Other mitigation measures developed, especially static gears, collaboration sector an inclusive management approach direct interactions fisheries cetaceans.

Language: Английский

---

Long-term biome biomonitoring

Elsevier eBooks, Journal Year: 2025, Volume and Issue: unknown, P. 69 - 94

Published: Jan. 1, 2025

Language: Английский

---

Multiclass CNN Approach for Automatic Classification of Dolphin Vocalizations

Sensors, Journal Year: 2025, Volume and Issue: 25(8), P. 2499 - 2499

Published: April 16, 2025

Monitoring dolphins in the open sea is essential for understanding their behavior and impact of human activities on marine ecosystems. Passive Acoustic (PAM) a non-invasive technique tracking dolphins, providing continuous data. This study presents novel approach classifying dolphin vocalizations from PAM acoustic recording using convolutional neural network (CNN). Four types common bottlenose (Tursiops truncatus) were identified underwater recordings: whistles, echolocation clicks, burst pulse sounds, feeding buzzes. To enhance classification performances, edge-detection filters applied to spectrograms, with aim removing unwanted noise components. A dataset nearly 10,000 spectrograms was used train test CNN through 10-fold cross-validation procedure. The results showed that achieved an average accuracy 95.2% F1-score 87.8%. class-specific high whistles (97.9%), followed by clicks (94.5%), buzzes (94.0%), sounds (92.3%). highest obtained exceeding 95%, while other three vocalization typologies maintained above 80%. method provides promising step toward improving passive monitoring contributing both species conservation mitigation conflicts fisheries.

Language: Английский

---

Sounds of the deep: How input representation, model choice, and dataset size influence underwater sound classification performance

The Journal of the Acoustical Society of America, Journal Year: 2025, Volume and Issue: 157(4), P. 3017 - 3032

Published: April 1, 2025

Convolutional neural networks (CNNs) have proven highly effective in automatically identifying and classifying underwater sound sources, enabling efficient analysis of marine environments. This work examines two key design choices for a CNN classifier: input representation network architecture, analyzing their importance as training data size varies effectiveness generalizing between sites. Passive acoustic from three offshore sites Western Scotland were used hierarchical classification; categorizing sounds into one four classes: delphinid tonal, clicks, vessels, ambient noise. Three different representations the signals investigated along with architectures, including pre-trained image classification tasks. Experiments show that custom-built shallow can outperform more complex ar chitectures if is chosen appropriately. For example, using Mel-spectrogram normalised per channel energy normalization (MS-PCEN) achieved 12.5% accuracy improvement over ResNet model when small amounts are available. Studying performance across demonstrates an important factor achieving robust results sites, MS-PCEN best performance. However, choice decreases dataset increases.

Language: Английский

---

Automatic detection and taxonomic identification of dolphin vocalisations using convolutional neural networks for passive acoustic monitoring

Ecological Informatics, Journal Year: 2023, Volume and Issue: 78, P. 102291 - 102291

Published: Sept. 12, 2023

A novel framework for acoustic detection and species identification is proposed to aid passive monitoring studies on the endangered Indian Ocean humpback dolphin (Sousa plumbea) in South African waters. Convolutional Neural Networks (CNNs) were used both of vocalisations tasks, performance was evaluated using custom pre-trained architectures (transfer learning). In total, 723 min data annotated presence whistles, burst pulses echolocation clicks produced by Delphinus delphis (~45.6%), Tursiops aduncus (~39%), Sousa plumbea (~14.4%), Orcinus orca (~1%). The best performing models detecting segments (spectral windows) two second lengths trained images with 70 90 dpi, respectively. model built a customised architecture achieved an accuracy 84.4% all test set, 89.5% high signal noise ratio. also correctly identified S. (96.9%), T. (100%), D. (78%) encounters testing dataset. developed designed based knowledge complex sounds it may assists finding suitable CNN hyper-parameters other or populations. Our study contributes towards development open-source tool assist long-term species, living highly diverse habitats, monitoring.

Language: Английский

---

Observational study on the non-linear response of dolphins to the presence of vessels

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 13, 2024

Abstract With the large increase in human marine activity, our seas have become populated with vessels that can be overheard from distances of even 20 km. Prior investigations showed such a dense presence impacts behaviour animals, and particular dolphins. While previous explorations were based on linear observation for changes features dolphin whistles, this work we examine non-linear responses bottlenose dolphins ( Tursiops Truncatus ) to vessels. We explored response by continuously recording acoustic data using two long-term recorders deployed near shipping lane habitat Eilat, Israel. Using deep learning methods detected number 50,000 which clustered associate whistle traces characterize their discriminate vocalizations dolphins: both structure quantities. classifier, whistles categorized into classes representing or absence nearby vessel. Although database does not show observable change obtained true positive negative rates exceeding 90% accuracy separate, left-out test sets. argue success classification serves as statistical proof

Language: Английский

---

Convolutional Neural Networks for Enhancing Detection of Dolphin Whistles in a Dense Acoustic Environment

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 127141 - 127148

Published: Jan. 1, 2024

Language: Английский

---

Building Ensemble of Resnet for Dolphin Whistle Detection

Published: June 7, 2023

To effectively preserve marine environments and manage endangered species, it is necessary to employ efficient, precise, scalable solutions for environmental monitoring. Ecoacoustics provides several benefits as enables non-intrusive, prolonged sampling of sounds, making a promising tool conducting biodiversity surveys. However, analyzing interpreting acoustic data can be time-consuming often demands substantial human supervision. This challenge addressed by harnessing contemporary methods automated audio signal analysis, which have exhibited remarkable performance due advancements in deep learning research. paper introduces research investigation into developing an automatic computerized system detect dolphin whistles. The proposed method utilizes fusion various resnet50 networks integrated with augmentation techniques. Through extensive experiments conducted on publically available benchmark, our findings demonstrate that ensemble yields significant enhancements across all evaluated metrics. MATLAB/PyTorch source code freely at: https://github.com/LorisNanni/

Language: Английский

---

Automated Audio Dataset Generation Tool for Classification Tasks in Marine Science

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: April 24, 2024

This study presents a comprehensive data processing pipeline designed to convert diverse labeled audio recordings into uniform dataset suitable for input machine learning algorithms. The primary emphasis lies in applying this marine classification, with an illustrative example from domain serving as experimental validation the proposed approach.

Language: Английский

---