Designing Ocean Vision AI: An Investigation of Community Needs for Imaging-based Ocean Conservation

Published: April 19, 2023

Ocean scientists studying diverse organisms and phenomena increasingly rely on imaging devices for their research. These have many tools to collect data, but few resources automated analysis. In this paper, we report discussions with stakeholders identify community needs develop a set of functional requirements the ongoing development ocean science-specific analysis tools. We conducted 36 in-depth interviews individuals working in Blue Economy space, revealing four central issues inhibiting effective monitoring marine science. also identified twelve user archetypes that will engage these services. Additionally, held workshop 246 participants from 35 countries centered around FathomNet, web-based open-source annotated image database Findings are being used define feature interface design Vision AI, suite services advance observational capabilities life ocean.

Language: Английский

---

Characterizing Håkon Mosby Mud Volcano (Barents Sea) cold seep systems by combining ROV-based acoustic data and underwater photogrammetry

Frontiers in Marine Science, Journal Year: 2023, Volume and Issue: 10

Published: Oct. 31, 2023

Cold-seeps have a unique geo-ecological significance in the deep-sea environment. They impact variability of present-day submarine sedimentary environments, affecting evolution landscape over time and creating variety landforms, one which is Mud Volcanoes (MVs). MVs form due to extrusion mud, fluids, gas, mainly methane, from deeper layers. These natural gas seepage systems could significantly affect climate change global carbon cycle. We present comprehensive method that combines ROV-based multibeam mapping underwater photogrammetry enhance understanding physical relationships between geomorphic units characterizing Håkon Mosby Volcano (HMMV) distribution associated habitats. HMMV indeed characterized by high thermal geochemical gradients its center margins resulting clear zonation chemosynthetic communities. Our approach integrates multi-resolutions multi-sources data acquired using work-class ROV. The microbathymetry helped identify different types fine-scale landforms central part HMMV. This revealed three distinct units, with hummocky region being most complex. To further study this area, ROV images were analyzed defined Structure Motion workflow producing millimetric resolution 2D 3D models. Object-Based Image Analysis (OBIA), applied on orthomosaics, allowed us obtain fine classification main benthic communities covering total area 940m 2 , including active rim. Four major substrate identified these regions: uncovered bacterial mats high-density, low-density, sediments tubeworms . Their relationship terrain morphology activity investigated at spatial scales, contributing ecological functioning cold seep ecosystems MVs. proposed as an innovative processing technique for future studies cold-seep systems. Geomorphic processes extreme environments are inherently linked marked patterns typifying habitats environments. poorly previous studies, leaving substantial gap geomorphological drivers responsible habitat extent

Language: Английский

---

Future directions for deep ocean climate science and evidence-based decision making

Frontiers in Climate, Journal Year: 2024, Volume and Issue: 6

Published: Oct. 31, 2024

Introduction A defining aspect of the Intergovernmental Panel on Climate Change (IPCC) assessment reports (AR) is a formal uncertainty language framework that emphasizes higher certainty issues across reports, especially in executive summaries and short for policymakers. As result, potentially significant risks involving understudied components climate system are shielded from view. Methods Here we seek to address this latest, sixth report (AR6) one such component—the deep ocean—by summarizing major uncertainties (based discussions low confidence or gaps) regarding its role our changing system. The goal identify key research priorities improve IPCC levels ocean systems facilitate dissemination results high impact processes decision-makers. This will accelerate improvement global projections aid informing efforts mitigate change impacts. An analysis 3,000 pages six selected AR6 revealed 219 science gaps related ocean. These were categorized by stressor nature Results Half these biological gaps, primarily surrounding understanding changes ecosystems, fisheries, primary productivity. remaining physical (32%) biogeochemical (15%) states processes. Model deficiencies leading cited cause ice states, whereas causes most often attributed limited studies observations conflicting results. Discussion Key areas coordinated effort within observing modeling community have emerged, which state ongoing next report. list “known unknowns” includes meridional overturning circulation, deoxygenation acidification, production, food supply carbon cycle, impacts ecosystems ocean-based interventions. From findings, offer recommendations AR7 avoid omitting confidence-high risk

Language: Английский

---

LM-CycleGAN: Improving Underwater Image Quality Through Learned Perceptual Image Patch Similarity and Multi-Scale Adaptive Fusion Attention

Sensors, Journal Year: 2024, Volume and Issue: 24(23), P. 7425 - 7425

Published: Nov. 21, 2024

The underwater imaging process is often hindered by high noise levels, blurring, and color distortion due to light scattering, absorption, suspended particles in the water. To address challenges of image enhancement complex environments, this paper proposes an correction detail model based on improved Cycle-consistent Generative Adversarial Network (CycleGAN), named LPIPS-MAFA CycleGAN (LM-CycleGAN). integrates a Multi-scale Adaptive Fusion Attention (MAFA) mechanism into generator architecture enhance its ability perceive details. At same time, Learned Perceptual Image Patch Similarity (LPIPS) introduced loss function make training more focused structural information image. Experiments conducted public datasets UIEB EUVP demonstrate that LM-CycleGAN achieves significant improvements Structural Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), Average Gradient (AG), Underwater Color Quality Evaluation (UCIQE), Measure (UIQM). Moreover, excels fidelity, successfully avoiding issues such as red checkerboard artifacts blurred edge details commonly observed reconstructed images generated traditional approaches.

Language: Английский

---

Laser-induced breakdown spectroscopy instrument and spectral analysis for deep-ocean Fe-Mn crusts

Frontiers in Marine Science, Journal Year: 2023, Volume and Issue: 10

Published: March 10, 2023

Introduction Marine ferromanganese crusts are potentially important metal resources.The deep-ocean research and survey ships often need to carry out rapid chemical element component analysis of mineral resources, so as plan for the geological resource exploration mission. Methods The laser-induced breakdown spectroscopy can obtain spectrum elements by plasma excited high-energy laser irradiation on surface sample. A induced optical system detection deepocean is designed built, which meet requirements near-insitu resources ocean-going ships. Results Hyperspectral data Fe-Mn carried Laser-induced (LIBS) during a deep-sea mission at depth 2,490 m in South China Sea. experimental parameters energy spectral acquisition delay optimized improve measurement accuracy. Based calibration-free method, significant features Fe Mn were obtained through proper alignment with National Institute Standards Technology (NIST) library. Discussion LIBS instrument be placed board long-range vessels future provide fast, convenient, accurate, economical method exploration.

Language: Английский

---

Maka Niu: A low-cost, modular imaging and sensor platform to increase observation capabilities of the deep ocean

Frontiers in Marine Science, Journal Year: 2022, Volume and Issue: 9

Published: Nov. 17, 2022

The deep sea (>200 m) is vast, covering 92.6% of the seafloor and largely unexplored. Imaging sensor platforms capable surviving immense pressures at these depths are expensive often engineered by individuals institutions in affluent countries as unique, monolithic vehicles that require significant expertise investment to build, operate, maintain. Maka Niu was co-designed with a global community deep-sea researchers. It low-cost, modular imaging platform leverages off-the-shelf commodity hardware along efficiencies mass production decrease price per unit allow more communities explore previously unseen regions ocean. combines Raspberry Pi single-board computer, Camera Module V2, novel pressure housing viewport combination withstanding 1,500 m water depth. Other modules, including high-lumen LEDs, can be use same battery charging control system form factor, allowing for an ever-increasing number capabilities added system. After deployment, imagery data wirelessly uploaded Tator, integrated media management machine learning backend automated analysis classification. Niu’s mobile mission programming systems designed user-friendly. Here, described detail recorded from deployments around world.

Language: Английский

---

A robust and adhesive anti-swelling hydrophobic ionogel with prolonged stability for strain and temperature sensors

Journal of Materials Chemistry A, Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 1, 2024

With the growing attention given to wearable electronic devices, flexible sensors, as a fundamental component, are emerging focal point.

Language: Английский

---

Testing and analysis of suitable compounds to design an air-tight dome to withstand high pressure for underwater applications

International Journal on Interactive Design and Manufacturing (IJIDeM), Journal Year: 2024, Volume and Issue: unknown

Published: Nov. 25, 2024

Language: Английский

---

Prioritising future evidence needs for marine and freshwater mammal conservation action

Published: March 13, 2024

Marine and freshwater mammals are increasingly threatened due to human activity. To improve conservation practice, decisions should be informed by the available evidence on effectiveness of actions. Using a systematically collated database studies that test actions conserve marine mammals, we investigated gaps biases in scientific base. Whilst there is growing base covering address key threats (e.g. fisheries bycatch) mammal populations, identified large geographic taxonomic biases. There was no relationship between number species per ecoregion found towards coastal areas Global North, with many regions having little or available. The did not correlate i) threat level, ii) evolutionary distinctiveness, iii) public ‘popularity’ study species. We also mismatch tested suggested as needed International Union for Conservation Nature (IUCN) Red List. Several these likely reflect feasibility researching populations; can difficult access, limited baseline information populations threats, testing require costly long-term monitoring. Prioritising most cost-effective strategies will comprehensive effects Continuing build necessary data, focussing future research funding priority this important deliver target.

Language: Английский

---

Generative adversarial networks with multi-scale and attention mechanisms for underwater image enhancement

Frontiers in Marine Science, Journal Year: 2023, Volume and Issue: 10

Published: Oct. 6, 2023

The images captured underwater are usually degraded due to the effects of light absorption and scattering. Degraded exhibit color distortion, low contrast, blurred details, which in turn reduce accuracy marine biological monitoring object detection. To address this issue, a generative adversarial network with multi-scale an attention mechanism is proposed improve quality images. extract more effective features within network, several modules introduced: dilated convolution module, novel residual module. These utilized design U-shaped structure. module designed at multiple scales expand receptive field capture global information. directs network’s focus towards important features, thereby reducing interference from redundant feature discriminative power discriminator designed. It has two output maps different scales. Additionally, improved loss function for proposed. This improvement involves incorporating total variation into traditional function. performance methods enhancing evaluated using EUVP dataset UIEB dataset. experimental results demonstrate that enhanced better visual compared other methods.

Language: Английский

---