A Methodological Literature Review of Acoustic Wildlife Monitoring Using Artificial Intelligence Tools and Techniques

Sustainability, Journal Year: 2023, Volume and Issue: 15(9), P. 7128 - 7128

Published: April 24, 2023

Artificial intelligence (AI) has become a significantly growing field in the environmental sector due to its ability solve problems, make decisions, and recognize patterns. The significance of AI wildlife acoustic monitoring is particularly important because vast amounts data that are available this field, which can be leveraged for computer vision interpretation. Despite increasing use ecology, future remains uncertain. To assess potential identify needs, scientific literature review was conducted on 54 works published between 2015 March 2022. results showed significant rise utilization techniques over period, with birds (N = 26) gaining most popularity, followed by mammals 12). commonly used algorithm Convolutional Neural Network, found more accurate beneficial than previous categorization methods monitoring. This highlights play crucial role advancing our understanding populations ecosystems. However, also show there still gaps Further examination previously algorithms bioacoustics research help researchers better understand patterns areas improvement autonomous In conclusion, rapidly lot potential. While progress been made recent years, much done fully realize field. needed limitations opportunities monitoring, develop new improve accuracy usefulness technology.

Language: Английский

---

Acoustic animal identification using unsupervised learning

Methods in Ecology and Evolution, Journal Year: 2023, Volume and Issue: 14(6), P. 1500 - 1514

Published: April 17, 2023

Abstract Passive acoustic monitoring is usually presented as a complementary approach to wildlife communities and assessing ecosystem conditions. Automatic species detection methods support biodiversity analysis by providing information on the presence–absence of species, which allows understanding structure. Therefore, different alternatives have been proposed identify species. However, algorithms are parameterized specific Analysing multiple would help monitor quantify biodiversity, it includes taxonomic groups present in soundscape. We an unsupervised methodology for multi‐species call recognition from ecological soundscapes. The proposal based clustering algorithm, specifically learning algorithm multivariate data (LAMDA) 3pi automatically suggests number clusters associated with sonotypes. Emphasis was made improving segmentation audio analyse whole soundscape without parameterizing according each group. To estimate performance our proposal, we used four datasets locations, years habitats. These contain sounds major that dominate terrestrial soundscapes (birds, amphibians, mammals insects) audible ultrasonic spectra. presents performances between 75% 96% recognition. Using methodology, measured compared indices (ACI, NP, SO BI). Our performs assessments similar advantage about need prior knowledge recordings.

Language: Английский

---

Individual identification in acoustic recordings

Trends in Ecology & Evolution, Journal Year: 2024, Volume and Issue: 39(10), P. 947 - 960

Published: June 12, 2024

Language: Английский

---

Stress‐Testing Monitoring Design to Lock in Conservation Success

Austral Ecology, Journal Year: 2025, Volume and Issue: 50(2)

Published: Jan. 30, 2025

ABSTRACT Effective monitoring of threatened species is key to identifying trends in populations and informing conservation management decisions. However, clearly defined questions that are informed by local circumstances traits commonly neglected. We propose a decision framework as guide prioritise what data collect methods use for population monitoring. applied our trial Gang‐gang Cockatoos ( Callocephalon fimbriatum ), threatened, iconic Southeast Australia. To meet program objectives, we trailed distance sampling surveys estimate abundance across the urban landscape Australian Capital Territory. Despite consistently high reporting rates study area, detection were too low Cockatoos. As part assessing appropriateness an approach, simulated under hypothetically inflated survey effort size. Simulations show even if field was doubled or size improbably high, detections would remain be practical approach. then revisit make new recommendations future demonstrate importance clear when evaluating how best achieve goals context methodological uncertainty. The first steps designing implementing crucial—our offers practitioners clear, reasoned approach deciding which needed address their along with contingencies plans go awry.

Language: Английский

---

Convolutional Neural Networks for the Identification of African Lions from Individual Vocalizations

Journal of Imaging, Journal Year: 2022, Volume and Issue: 8(4), P. 96 - 96

Published: April 1, 2022

The classification of vocal individuality for passive acoustic monitoring (PAM) and census animals is becoming an increasingly popular area research. Nearly all studies in this field inquiry have relied on classic audio representations classifiers, such as Support Vector Machines (SVMs) trained spectrograms or Mel-Frequency Cepstral Coefficients (MFCCs). In contrast, most current bioacoustic species exploits the power deep learners more cutting-edge representations. A significant reason avoiding learning identity tiny sample size collections labeled individual vocalizations. As well known, require large datasets to avoid overfitting. One way handle small with methods use transfer learning. work, we evaluate performance three pretrained CNNs (VGG16, ResNet50, AlexNet) a small, publicly available lion roar dataset containing approximately 150 samples taken from five male lions. Each these networks retrained eight samples: MFCCs, spectrogram, Mel along several new ones, VGGish stockwell, those based recently proposed LM spectrogram. networks, both individually ensembles, analyzed corroborated using Equal Error Rate shown surpass previous attempts dataset; best single network achieved over 95% accuracy ensembles 98% accuracy. contributions study makes include demonstrating that it valuable possible, caution, problem domain. We also make contribution bioacoustics generally by offering comparison many state-of-the-art representations, including first time spectrogram stockwell All source code GitHub.

Language: Английский

---

Cross-corpus open set bird species recognition by vocalization

Ecological Indicators, Journal Year: 2023, Volume and Issue: 154, P. 110826 - 110826

Published: Aug. 28, 2023

In the wild, bird vocalizations of same species across different populations may be (e. g., so called dialect). Besides, number is unknown in advance. These two facts make task recognition based on vocalization a challenging one. This study treats this as an open set (OSR) cross-corpus scenario. We propose Instance Frequency Normalization (IFN) to remove instance-specific differences corpora. Furthermore, x-vector feature extraction model integrated Time Delay Neural Network (TDNN) and Long Short-Term Memory (LSTM) are designed better capture sequence information. Finally, threshold-based Probabilistic Linear Discriminant Analysis (PLDA) introduced discriminate extracted features discover classes. When compared best results existing method, average ACCs for single-corpus experiments improved, implying that our method can provide potential solution improve performance condition.

Language: Английский

---

Windy events detection in big bioacoustics datasets using a pre-trained Convolutional Neural Network

The Science of The Total Environment, Journal Year: 2024, Volume and Issue: 949, P. 174868 - 174868

Published: July 20, 2024

Passive Acoustic Monitoring (PAM), which involves using autonomous record units for studying wildlife behaviour and distribution, often requires handling big acoustic datasets collected over extended periods. While these data offer invaluable insights about wildlife, their analysis can present challenges in dealing with geophonic sources. A major issue the process of detection target sounds is represented by wind-induced noise. This lead to false positive detections, i.e., energy peaks due wind gusts misclassified as biological sounds, or negative, noise masks presence sounds. dominated makes vocal activity unreliable, thus compromising and, subsequently, interpretation results. Our work introduces a straightforward approach detecting recordings affected windy events pre-trained convolutional neural network. facilitates identifying wind-compromised data. We consider this dataset pre-processing crucial ensuring reliable use PAM implemented preprocessing leveraging YAMNet, deep learning model sound classification tasks. evaluated YAMNet as-is ability detect tested its performance Transfer Learning scenario our annotated from Stony Point Penguin Colony South Africa. achieved precision 0.71, recall 0.66, those metrics strongly improved after training on dataset, reaching 0.91, 0.92, corresponding relative increment >28 %. study demonstrates promising application bioacoustics ecoacoustics fields, addressing need wind-noise-free released an open-access code that, combined efficiency peak be used standard laptops broad user base.

Language: Английский

---

Identification of dialects and individuals of globally threatened yellow cardinals using neural networks

Ecological Informatics, Journal Year: 2023, Volume and Issue: 78, P. 102372 - 102372

Published: Nov. 10, 2023

Language: Английский

---

Duets convey information about pair and individual identities in a Neotropical bird

Current Zoology, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 21, 2024

Abstract Vocal individuality is essential for social discrimination but has been poorly studied in animals that produce communal signals (duets or choruses). Song overlapping and temporal coordination make the assessment of more complex. In addition, selection may favor accurate identification pairs over individuals by receivers year-round territorial species with duetting long-term pair bonding. Here, we individual vocal signatures polyphonal duets rufous horneros Furnarius rufus, a Neotropical bird known its bonds. Hornero partners engage to deter intruders protect their partnership can discern from neighbors versus strangers. Using dataset 471 43 2 populations, measured fine-scale acoustic features across different duet levels (e.g., complete non-overlapping syllable parts) analysis (pair individual). Permuted linear discriminant function analyses classified accurately than expected chance (means: 45% 47% vs. 4 2%). Pair identity explained variance multivariate population identities. The initial frequency showed strong potential encoding identity. traits contributing most varied between sexes, which might facilitate simultaneous duetters’ identities receivers. Our study indicates exist even intricate innate elucidates mechanisms employed ability.

Language: Английский

---

Insights from Deep Learning in Feature Extraction for Non-supervised Multi-species Identification in Soundscapes

Lecture notes in computer science, Journal Year: 2022, Volume and Issue: unknown, P. 218 - 230

Published: Jan. 1, 2022

Language: Английский

---