First release of the Pelagic Size Structure database: global datasets of marine size spectra obtained from plankton imaging devices

Earth system science data, Год журнала: 2024, Номер 16(6), С. 2971 - 2999

Опубликована: Июнь 26, 2024

Abstract. In marine ecosystems, most physiological, ecological, or physical processes are size dependent. These include metabolic rates, the uptake of carbon and other nutrients, swimming sinking velocities, trophic interactions, which eventually determine stocks commercial species, as well biogeochemical cycles sequestration. As such, broad-scale observations plankton distribution important indicators general functioning state pelagic ecosystems under anthropogenic pressures. Here, we present first global datasets Pelagic Size Structure database (PSSdb), generated from imaging devices. This release includes bulk particle normalized biovolume spectrum (NBSS) (PSD), along with their related parameters (slope, intercept, R2) measured within epipelagic layer (0–200 m) by three sensors: Imaging FlowCytobot (IFCB), Underwater Vision Profiler (UVP), benchtop scanners. Collectively, these instruments effectively image organisms detrital material in 7–10 000 µm range. A total 92 472 IFCB samples, 3068 UVP profiles, 2411 scans passed our quality control were standardized to produce consistent instrument-specific spectra averaged 1° × latitude longitude year month. Our span major ocean basins, except for have ingested, exclusively collected northern latitudes, cover decadal time periods (2013–2022 IFCB, 2008–2021 UVP, 1996–2022 scanners), allowing a further assessment space time. The that constitute PSSdb's available at https://doi.org/10.5281/zenodo.11050013 (Dugenne et al., 2024b). addition, future updates data products can be accessed https://doi.org/10.5281/zenodo.7998799.

Язык: Английский

---

Machine learning for non-experts: A more accessible and simpler approach to automatic benthic habitat classification

Ecological Informatics, Год журнала: 2024, Номер 81, С. 102619 - 102619

Опубликована: Май 10, 2024

Automating identification of benthic habitats from imagery, with Machine Learning (ML), is necessary to contribute efficiently and effectively marine spatial planning. A promising method adapt pre-trained general convolutional neural networks (CNNs) a new classification task (transfer learning). However, this often inaccessible non-specialist, requiring large investments in computational resources time (for user comprehension model training). In paper, we demonstrate simpler transfer learning framework for classifying broad deep-sea habitats. Specifically, take an 'off-the-shelf' CNN (VGG16) use it extract features (pixel patterns) images (without further The default outputs VGG16 are then fed Support Vector (SVM), classical than deep networks. For comparison, also train the remaining layers using stochastic gradient descent. discriminative power these approaches demonstrated on three datasets (574–8353 images) Norwegian waters; each unique imaging platform. Benthic broadly classified as Soft Substrate (sands, muds), Hard (gravels, cobbles boulders) Reef (Desmophyllum pertusum). We found that relatively simplicity SVM classifier did not compromise performance. Results were competitive consistently high, test accuracy ranging 0.87 0.95 (average = 0.9 (±0.04)) across datasets, somewhat increasing dataset size. Impressively, results achieved 2.4–5× faster training had significantly less dependency high-specification hardware. Our suggested approach maximises conceptual practical simplicity, representing realistic baseline novice users when approaching habitat classification. This has wide potential. It allows automated image grouping aid annotation or selection, well screening old-datasets. especially suited offshore scenarios can provide quick, albeit crude, insights into presence, allowing adaptation sampling protocols near real-time.

Язык: Английский

---

DAPlankton: Benchmark Dataset For Multi-Instrument Plankton Recognition Via Fine-Grained Domain Adaptation

2022 IEEE International Conference on Image Processing (ICIP), Год журнала: 2024, Номер unknown, С. 158 - 164

Опубликована: Сен. 27, 2024

Язык: Английский

---

Integrating Machine Learning with Flow-Imaging Microscopy for Automated Monitoring of Algal Blooms

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Ноя. 15, 2024

Abstract Real-time monitoring of phytoplankton in freshwater systems is critical for early detection harmful algal blooms so as to enable efficient response by water management agencies. This paper presents an image processing pipeline developed adapt ARTiMiS, a low-cost automated flow-imaging device, real-time specifically and environmental systems. addresses several challenges associated with autonomous imaging aquatic samples such artifacts (i.e., out-of-focus background objects), well specific open identification novel objects). The leverages Random Forest model identify out- of-focus particles accuracy 89% custom particle algorithm remove that erroneously appear consecutive images >97±2.8% accuracy. Furthermore, convolutional neural network (CNN), trained classify distinct classes comprising both taxonomical morphological categories, achieved 94% closed dataset. Nonetheless, the supervised closed-set classifiers struggled accurate classification objects when challenged debris which are common complex environments; this limits applications requiring extensive manual oversight. To mitigate this, three methods incorporating rejection were tested improve precision excluding irrelevant or unknown classes. Combined, these advances present fully integrated, end-to-end solution HAB thus enhancing scalability dynamic environments. Highlights more generalizable than Convolutional Neural Networks particles. A two-stage clustering effective at removing flow microscopy. Closed-set CNN classifier performance deteriorates Classification improves samples.

Язык: Английский

---

Producing plankton classifiers that are robust to dataset shift

Limnology and Oceanography Methods, Год журнала: 2024, Номер unknown

Опубликована: Ноя. 27, 2024

Abstract Modern plankton high‐throughput monitoring relies on deep learning classifiers for species recognition in water ecosystems. Despite satisfactory nominal performances, a significant challenge arises from dataset shift, which causes performances to drop during deployment. In our study, we integrate the ZooLake dataset, consists of dark‐field images lake (Kyathanahally et al. 2021a), with manually annotated 10 independent days deployment, serving as test cells benchmark out‐of‐dataset (OOD) performances. Our analysis reveals instances where classifiers, initially performing well in‐dataset conditions, encounter notable failures practical scenarios. For example, MobileNet 92% accuracy shows 77% OOD accuracy. We systematically investigate conditions leading performance drops and propose preemptive assessment method identify potential pitfalls when classifying new data, pinpoint features that adversely impact classification. present three‐step pipeline: (i) identifying degradation compared performance, (ii) conducting diagnostic causes, (iii) providing solutions. find ensembles BEiT vision transformers, targeted augmentations addressing robustness, geometric ensembling, rotation‐based test‐time augmentation, constitute most robust model, call BEsT . It achieves an 83% accuracy, errors concentrated container classes. Moreover, it exhibits lower sensitivity reproduces abundances. proposed pipeline is applicable generic contingent availability suitable cells. By critical shortcomings offering procedures fortify models against study contributes development more reliable classification technologies.

Язык: Английский

---

“UDE DIATOMS in the Wild 2024”: a new image dataset of freshwater diatoms for training deep learning models

GigaScience, Год журнала: 2024, Номер 13

Опубликована: Янв. 1, 2024

Abstract Background Diatoms are microalgae with finely ornamented microscopic silica shells. Their taxonomic identification by light microscopy is routinely used as part of community ecological research well status assessment aquatic ecosystems, and a need for digitalization these methods has long been recognized. Alongside their high morphological diversity, several other factors make diatoms highly challenging deep learning–based using images. These include (i) an unusually intraclass variability combined small between-class differences, (ii) rather different visual appearance specimens depending on orientation the microscope slide, (iii) limited availability diatom experts accurate annotation. Findings We present largest image dataset thus far, aimed at facilitating application benchmarking innovative learning to problem realistic data, “UDE DIATOMS in Wild 2024.” The contains 83,570 images 611 taxa, 101 which represented least 100 examples 144 50 each. showcase this 2 analyses that address individual aspects above challenges subclustering deal visually heterogeneous classes, out-of-distribution sample detection, semi-supervised learning. Conclusions image-based both important environmental from machine perspective. By making available so far dataset, accompanied analyses, contribution will facilitate addressing points scientific community.

Язык: Английский

---

Computer Vision Techniques for Morphological Analysis and Identification of Two Pseudo-nitzschia Species

Water, Год журнала: 2024, Номер 16(15), С. 2160 - 2160

Опубликована: Июль 31, 2024

The diversity of phytoplankton influences the structure and processes that occur in marine ecosystems, with size other morphological traits being crucial for nutrient uptake retention euphotic zone. Our research introduces a machine learning method can facilitate analysis functional from image data. We use computer vision to identify quantify species estimate size-related based on cell morphology. study uses transfer learning, where generic, pre-trained YOLOv8 models are fine-tuned microscope data Adriatic Sea. shows that, this task, it is possible effectively fine-tune trained out-of-domain images small training dataset. results show high accuracy detecting segmenting cells microscopic two selected taxa. For detection, model achieves AP scores 88.1% Pseudo-nitzschia cf. delicatissima 90.9% calliantha, while segmentation, 88.4% 91.2% calliantha. Compared manual analysis, developed automatic significantly increases number samples be processed.

Язык: Английский

---

Morphotype-Resolved Characterization of Microalgal Communities in a Nutrient Recovery Process with ARTiMiS Flow Imaging Microscopy

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Дек. 16, 2024

Abstract Microalgae-driven nutrient recovery represents a promising technology to reduce effluent phosphorus while simultaneously generating biomass that can be valorized offset treatment costs. As full-scale processes come online, system parameters including composition must carefully monitored optimize performance and prevent culture crashes. In this study, flow imaging microscopy (FIM) was leveraged characterize microalgal community in near real-time at municipal wastewater plant (WWTP) Wisconsin, USA, population morphotype dynamics were examined identify relationships between water chemistry, composition, performance. Two FIM technologies, FlowCam ARTiMiS, evaluated as monitoring tools. ARTiMiS provided more accurate estimate of total biomass, estimates derived from particle area proxy for biovolume yielded better approximations than counts. Deep learning classification models trained on annotated image libraries demonstrated equivalent convolutional neural network (CNN) classifiers proved significantly when compared feature table-based deep (DNN) models. Across two-year study period, Scenedesmus spp. appeared most important removal, which negatively associated with elevated temperatures nitrite/nitrate concentrations. Chlorella Monoraphidium also played an role For both , smaller morphological types often high performance, whereas larger morphotypes implied stress response correlating poor rates. These results demonstrate the potential critical high-resolution characterization industrial processes. Graphical

Язык: Английский

---

Unsupervised Learning Approaches for Zooplankton Classification: Recent Trends and Advances

Опубликована: Дек. 13, 2023

Zooplankton are key components of the aquatic food web and present a lot taxonomic diversity. Over years, various Machine Learning techniques have been employed for classification zooplankton. Supervised has widely utilised in zooplankton classification, presenting commendable performance. However, it requires substantial amount manually labelled images, volume collected images is extensive. Consequently, Unsupervised proven exceptionally valuable popular clustering unlabelled data. Our study compiles elucidates methods applied to while also comparing their performance popularity over based on observations from respective experimental studies.

Язык: Английский

---