Detection and classification of chilli leaf disease using a squeeze-and-excitation-based CNN model

Ecological Informatics, Journal Year: 2022, Volume and Issue: 69, P. 101663 - 101663

Published: May 6, 2022

Language: Английский

---

A review of automatic recognition technology for bird vocalizations in the deep learning era

Ecological Informatics, Journal Year: 2022, Volume and Issue: 73, P. 101927 - 101927

Published: Nov. 25, 2022

Language: Английский

---

Unsupervised classification to improve the quality of a bird song recording dataset

Ecological Informatics, Journal Year: 2022, Volume and Issue: 74, P. 101952 - 101952

Published: Dec. 12, 2022

Language: Английский

---

DBS-NET: A Dual-Branch Network Integrating Supervised and Contrastive Self-Supervised Learning for Birdsong Classification

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(10), P. 5418 - 5418

Published: May 12, 2025

Birdsong classification plays a crucial role in monitoring species distribution, population structure, and environmental changes. Existing methods typically use supervised learning to extract specific features for classification, but this may limit the generalization ability of model lead errors. Unsupervised feature extraction are an emerging approach that offers enhanced adaptability, particularly handling unlabeled diverse birdsong data. However, their drawback bring additional time cost downstream tasks, which impact overall efficiency. To address these challenges, we propose DBS-NET, Dual-Branch Network Model classification. DBS-NET consists two branches: branch (Res-iDAFF) unsupervised (based on contrastive approach). We introduce iterative dual-attention fusion (iDAFF) module backbone enhance contextual extraction, linear residual classifier is exploited further improve accuracy. Additionally, class imbalance dataset, weighted loss function introduced adjust cross-entropy with optimized weights. training efficiency, networks both branches share portion weights, reducing computational overhead. In experiments self-built 30-class dataset Birdsdata proposed method achieved accuracies 97.54% 97.09%, respectively, outperforming other methods.

Language: Английский

---

Automatic detection for bioacoustic research: a practical guide from and for biologists and computer scientists

Biological reviews/Biological reviews of the Cambridge Philosophical Society, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 17, 2024

ABSTRACT Recent years have seen a dramatic rise in the use of passive acoustic monitoring (PAM) for biological and ecological applications, corresponding increase volume data generated. However, sets are often becoming so sizable that analysing them manually is increasingly burdensome unrealistic. Fortunately, we also computing power capability machine learning algorithms, which offer possibility performing some analysis required PAM automatically. Nonetheless, field automatic detection events still its infancy biology ecology. In this review, examine trends bioacoustic their implications burgeoning amount needs to be analysed. We explore different methods other tools scanning, analysing, extracting automatically from large volumes recordings. then provide step‐by‐step practical guide using bioacoustics. One biggest challenges greater bioacoustics there gulf expertise between sciences computer science. Therefore, review first presents an overview requirements bioacoustics, intended familiarise those science background with community, followed by introduction key elements artificial intelligence biologist understand incorporate into research. building pipeline data, conclude discussion possible future directions field.

Language: Английский

---

Classification of birdsong spectrograms based on DR-ACGAN and dynamic convolution

Ecological Informatics, Journal Year: 2023, Volume and Issue: 77, P. 102250 - 102250

Published: Aug. 7, 2023

Birdsongs are highly valuable for bird studies as they provide insights into various aspects such species distribution, population structures, and habitat. Recognizing birdsongs plays a crucial role in conservation efforts. However, manually collecting large number of from the natural environment is expensive time-consuming. Moreover, using limited birdsong data often results low classification accuracy models. To better identification birdsongs, we utilize wavelet transform(WT) to convert them spectrograms, which contain abundant energy frequency information. Effectively extracting these features vital improve model. address this problem, proposed an improved ACGAN model based on residual structure attention mechanism named DR-ACGAN, achieved stable training high-quality generated spectrograms. The dynamic convolution kernel then fused with MobileNetV2, ResNet18, VGG16 models trained different datasets, used ways mixing original experimental show that after augmentation improves by 6.66%, 4.35%, 2.29% compared dataset three base classifiers. After adding convolutional structure, further 1.68%, 0.67%, 0.38% average achieves highest 97.60%.

Language: Английский

---

Evaluating community-wide temporal sampling in passive acoustic monitoring: A comprehensive study of avian vocal patterns in subtropical montane forests

F1000Research, Journal Year: 2024, Volume and Issue: 12, P. 1299 - 1299

Published: Jan. 23, 2024

Background From passive acoustic monitoring (PAM) recordings, the vocal activity rate (VAR), vocalizations per unit of time, can be calculated and is essential for assessing bird population abundance. However, VAR subject to influences from a range factors, including species environmental conditions. Identifying optimal sampling design obtain representative data estimation crucial research objectives. PAM commonly uses temporal strategies decrease volume recordings resources needed audio management. Yet, comprehensive impact this approach on remains insufficiently explored. Methods In study, we used extracted 12 species, taken at 14 stations situated in subtropical montane forests over four-month period, assess across three distinct scales: short-term periodic, diel, hourly. For periodic analysis, employed hierarchical clustering analysis (HCA) coefficient variation (CV). Generalized additive models (GAMs) were utilized diel determined average difference values minute hourly analysis. Results We identified significant day species-specific fluctuations. The survey season was divided into five segments; earliest two showed high variability are best avoided surveys. Data days with heavy rain strong winds reduced should excluded Continuous spanning least seven days, extending minimizing variance. Morning chorus effectively capture majority vocalizations, frequent, shorter intervals aligns closely continuous recording outcomes. Conclusions While our findings context-specific, they highlight significance strategic avian monitoring, optimizing resource utilization enhancing breadth efforts.

Language: Английский

---

An acoustic detection dataset of birds (Aves) in montane forests using a deep learning approach

Biodiversity Data Journal, Journal Year: 2023, Volume and Issue: 11

Published: Feb. 24, 2023

Long-term monitoring is needed to understand the statuses and trends of wildlife communities in montane forests, such as those Yushan National Park (YSNP), Taiwan. Integrating passive acoustic (PAM) with an automated sound identifier, a long-term biodiversity project containing six PAM stations, was launched YSNP January 2020 currently ongoing. SILIC, identification model, used extract sounds species information from recordings collected. Animal vocal activity can reflect their breeding status, behaviour, population, movement distribution, which may be affected by factors, habitat loss, climate change human activity. This massive amount vocalisation dataset provide essential for Park's headquarters on resource management decision-making. It also valuable studying effects animal distribution behaviour at regional or global scale. To our best knowledge, this first open-access occurrence data extracted soundscape artificial intelligence. We obtained seven bird release, more other taxa, mammals frogs, updated annually. Raw over 1.7 million one-minute collected between years 2021 were analysed SILIC identified 6,243,820 vocalisations 439,275 recordings. The automatic detection had precision 0.95 recall ranged 0.48 0.80. In terms balance recall, we prioritised increasing order minimise false positive detections. dataset, summarised count detected per class recording resulted 802,670 records. Unlike traditional observation methods, number observations Darwin Core "organismQuantity" column refers specific cannot directly linked individuals. expect will able help fill gaps fine-scale avian temporal patterns forests contribute studies concerning impacts forest ecosystems scales.

Language: Английский

---

One-step progressive representation transfer learning for bird sound classification

Applied Acoustics, Journal Year: 2023, Volume and Issue: 212, P. 109614 - 109614

Published: Sept. 1, 2023

Language: Английский

---

Automatic vocalisation detection delivers reliable, multi-faceted, and global avian biodiversity monitoring

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: Sept. 17, 2023

Abstract Tracking biodiversity and its dynamics at scale is essential if we are to solve global environmental challenges. Detecting animal vocalisations in passively recorded audio data offers a highly automatable, inexpensive, taxonomically broad way monitor biodiversity. However, uptake slow due the expertise labour required label new fine-tune algorithms for each deployment. In this study, applied an off-the-shelf bird vocalisation detection model, BirdNET, 152,376 hours of comprising datasets from Norway, Taiwan, Costa Rica, Brazil. We manually listened subset detections species dataset found precisions over 80% 89 139 (100% 57 species). Whilst some were reliably detected across multiple datasets, performance others was specific. By filtering out unreliable detections, could extract community level insight on diel (Brazil) seasonal (Taiwan) temporal scales, as well landscape (Costa Rica) national (Norway) spatial scales. Our findings demonstrate that, with relatively fast validation step, single model can deliver multi-faceted diverse datasets; unlocking which acoustic monitoring immediate impact.

Language: Английский

---