A small underwater object detection model with enhanced feature extraction and fusion DOI Creative Commons
Tao Li,

Yijin Gang,

Sumin Li

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Jan. 18, 2025

In the underwater domain, small object detection plays a crucial role in protection, management, and monitoring of environment marine life. Advancements deep learning have led to development many efficient techniques. However, complexity environment, limited information available from objects, constrained computational resources make challenging. To tackle these challenges, this paper presents an convolutional network model. First, CSP for lightweight (CSPSL) module is introduced enhance feature retention preserve essential details. Next, variable kernel convolution (VKConv) proposed dynamically adjust size, enabling better multi-scale extraction. Finally, spatial pyramid pooling (SPPFMS) method presented features objects more effectively. Ablation experiments on UDD dataset demonstrate effectiveness methods. Comparative DUO datasets that model delivers best performance terms cost accuracy, outperforming state-of-the-art methods real-time tasks.

Language: Английский

Machine learning in marine ecology: an overview of techniques and applications DOI Creative Commons
Peter Rubbens, Stephanie Brodie, Tristan Cordier

et al.

ICES Journal of Marine Science, Journal Year: 2023, Volume and Issue: 80(7), P. 1829 - 1853

Published: Aug. 3, 2023

Abstract Machine learning covers a large set of algorithms that can be trained to identify patterns in data. Thanks the increase amount data and computing power available, it has become pervasive across scientific disciplines. We first highlight why machine is needed marine ecology. Then we provide quick primer on techniques vocabulary. built database ∼1000 publications implement such analyse ecology For various types (images, optical spectra, acoustics, omics, geolocations, biogeochemical profiles, satellite imagery), present historical perspective applications proved influential, serve as templates for new work, or represent diversity approaches. Then, illustrate how used better understand ecological systems, by combining sources Through this coverage literature, demonstrate an proportion studies use learning, pervasiveness images source, dominance classification-type problems, shift towards deep all types. This overview meant guide researchers who wish apply methods their datasets.

Language: Английский

Citations

53
Computer vision and deep learning for fish classification in underwater habitats: A survey DOI
Alzayat Saleh, Marcus Sheaves, Mostafa Rahimi Azghadi

et al.

Fish and Fisheries, Journal Year: 2022, Volume and Issue: 23(4), P. 977 - 999

Published: April 15, 2022

Abstract Marine scientists use remote underwater image and video recording to survey fish species in their natural habitats. This helps them get a step closer towards understanding predicting how respond climate change, habitat degradation fishing pressure. information is essential for developing sustainable fisheries human consumption, preserving the environment. However, enormous volume of collected videos makes extracting useful daunting time‐consuming task being. A promising method address this problem cutting‐edge deep learning (DL) technology. DL can help marine parse large volumes promptly efficiently, unlocking niche that cannot be obtained using conventional manual monitoring methods. In paper, we first provide computer visions (CVs) studies conducted between 2003 2021 on classification We then give an overview key concepts DL, while analysing synthesizing studies. also discuss main challenges faced when processing propose approaches them. Finally, insights into research domain shed light what future may hold. paper aims inform who would like gain high‐level state‐of‐the‐art DL‐based habitat.

Language: Английский

Citations

68
A Review on the Use of Computer Vision and Artificial Intelligence for Fish Recognition, Monitoring, and Management DOI Creative Commons
Jayme Garcia Arnal Barbedo

Fishes, Journal Year: 2022, Volume and Issue: 7(6), P. 335 - 335

Published: Nov. 17, 2022

Computer vision has been applied to fish recognition for at least three decades. With the inception of deep learning techniques in early 2010s, use digital images grew strongly, and this trend is likely continue. As number articles published grows, it becomes harder keep track current state art determine best course action new studies. In context, article characterizes by identifying main studies on subject briefly describing their approach. contrast with most previous reviews related technology recognition, monitoring, management, rather than providing a detailed overview being proposed, work focuses heavily challenges research gaps that still remain. Emphasis given prevalent weaknesses prevent more widespread type practical operations under real-world conditions. Some possible solutions potential directions future are suggested, as an effort bring developed academy closer meeting requirements found practice.

Language: Английский

Citations

46
Applications of deep learning in fish habitat monitoring: A tutorial and survey DOI Creative Commons
Alzayat Saleh, Marcus Sheaves, Dean R. Jerry

et al.

Expert Systems with Applications, Journal Year: 2023, Volume and Issue: 238, P. 121841 - 121841

Published: Oct. 1, 2023

Marine ecosystems and their fish habitats are becoming increasingly important due to integral role in providing a valuable food source conservation outcomes. Due remote difficult access nature, marine environments often monitored using underwater cameras record videos images for understanding life ecology, as well preserve the environment. There currently many permanent camera systems deployed at different places around globe. In addition, there exists numerous studies that use temporary survey habitats. These generate massive volume of digital data, which cannot be efficiently analysed by current manual processing methods, involve human observer. Deep Learning (DL) is cutting-edge Artificial Intelligence (AI) technology has demonstrated unprecedented performance analysing visual data. Despite its application myriad domains, habitat monitoring remains under explored. this paper, we provide tutorial covers key concepts DL, help reader grasp high-level how DL works. The also explains step-by-step procedure on algorithms should developed challenging applications such monitoring. comprehensive deep learning techniques including classification, counting, localisation, segmentation. Furthermore, publicly available datasets, compare various domains. We discuss some challenges opportunities emerging field processing. This paper written serve scientists who would like develop it following our tutorial, see evolving facilitate research efforts. At same time, suitable computer state-of-the-art DL-based methodologies

Language: Английский

Citations

35
Survey of automatic plankton image recognition: challenges, existing solutions and future perspectives DOI Creative Commons
Tuomas Eerola, Daniel Batrakhanov,

Nastaran Vatankhah Barazandeh

et al.

Artificial Intelligence Review, Journal Year: 2024, Volume and Issue: 57(5)

Published: April 12, 2024

Abstract Planktonic organisms including phyto-, zoo-, and mixoplankton are key components of aquatic ecosystems respond quickly to changes in the environment, therefore their monitoring is vital follow understand these changes. Advances imaging technology have enabled novel possibilities study plankton populations, but manual classification images time consuming expert-based, making such an approach unsuitable for large-scale application urging automatic solutions analysis, especially recognizing species from images. Despite extensive research done on recognition, latest cutting-edge methods not been widely adopted operational use. In this paper, a comprehensive survey existing recognition presented. First, we identify most notable challenges that make development systems difficult restrict deployment Then, provide detailed description found literature. Finally, propose workflow specific new datasets recommended approaches address them. Many important remain unsolved following: (1) domain shift between hindering instrument independent system, (2) difficulty process previously unseen classes non-plankton particles, (3) uncertainty expert annotations affects training machine learning models. To build harmonized location agnostic purposes should be addressed future research.

Language: Английский

Citations

9
ConvFishNet: An efficient backbone for fish classification from composited underwater images DOI

Huishan Qu,

Gai‐Ge Wang, Li Yun

et al.

Information Sciences, Journal Year: 2024, Volume and Issue: 679, P. 121078 - 121078

Published: June 19, 2024

Language: Английский

Citations

9
Surveying the deep: A review of computer vision in the benthos DOI Creative Commons
Cameron Trotter, Huw J. Griffiths, Rowan J. Whittle

et al.

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 102989 - 102989

Published: Jan. 1, 2025

Language: Английский

Citations

1
Ecological inferences about marine mammals from passive acoustic data DOI Creative Commons
Erica Fleishman, Danielle Cholewiak, Douglas Gillespie

et al.

Biological reviews/Biological reviews of the Cambridge Philosophical Society, Journal Year: 2023, Volume and Issue: 98(5), P. 1633 - 1647

Published: May 4, 2023

ABSTRACT Monitoring on the basis of sound recordings, or passive acoustic monitoring, can complement serve as an alternative to real‐time visual aural monitoring marine mammals and other animals by human observers. Passive data support estimation common, individual‐level ecological metrics, such presence, detection‐weighted occupancy, abundance density, population viability structure, behaviour. also some community‐level species richness composition. The feasibility certainty estimates is highly context dependent, understanding factors that affect reliability measurements useful for those considering whether use data. Here, we review basic concepts methods sampling in systems often are applicable mammal research conservation. Our ultimate aim facilitate collaboration among ecologists, bioacousticians, analysts. Ecological applications acoustics require one make decisions about design, which turn requires consideration propagation, signals, storage. One must signal detection classification evaluation performance algorithms these tasks. Investment development automate classification, including machine learning, increasing. more reliable presence than species‐level metrics. Use distinguish individual remains difficult. However, information probability, vocalisation cue rate, relations between vocalisations number behaviour increases estimating density. Most sensor deployments fixed space sporadic, making temporal turnover composition tractable estimate spatial turnover. Collaborations acousticians ecologists most likely be successful rewarding when all partners critically examine share a fundamental target variables, process, analytical methods.

Language: Английский

Citations

18
A critical review of machine-learning for “multi-omics” marine metabolite datasets DOI

Janani Manochkumar,

Aswani Kumar Cherukuri, Raju Suresh Kumar

et al.

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 165, P. 107425 - 107425

Published: Aug. 29, 2023

Language: Английский

Citations

14
Towards global traceability for sustainable cephalopod seafood DOI
Ian G. Gleadall, Hassan Moustahfid, WHH Sauer

et al.

Marine Biology, Journal Year: 2023, Volume and Issue: 171(2)

Published: Dec. 30, 2023

Language: Английский

Citations

14