Deep Learning Classification of Lake Zooplankton

Kyathanahally, Sreenath Pruthviraj; Hardeman, Thomas; Merz, Ewa; Bulas, Thea; Reyes, Marta; Isles, Peter D. F.; Pomati, Francesco; Baity‐Jesi, Marco

doi:10.3389/fmicb.2021.746297

Cited by 21 publications

(35 citation statements)

References 61 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We train each of the CNNs in Table 1 four times (as described in Ref. 22 ), with different realisations of the initial conditions, and show their arithmetic average ensemble and geometric average ensemble ("Ensemble learning" section) in the last two columns. We also show the performance of the ensemble model developed in Ref.…”

Section: Ensemble Comparisonmentioning

confidence: 99%

“…We also show the performance of the ensemble model developed in Ref. 22 , which ensembles over the six shown CNN architectures. We compare those with the ensembled DeiT-Base model, obtained through arithmetic average ensemble and geometric average ensemble over three different initial conditions of the model weights.…”

Section: Ensemble Comparisonmentioning

confidence: 99%

“…For CNNs, we take the best EfficientNet-B7, MobileNet and Dense121 models from Ref. 22 (each had the best validation performance from 4 independent www.nature.com/scientificreports/ runs). For DeiTs, we train a DeiT-Base model three times (with different initial weight configurations) and ensemble over those three.…”

Section: Ensemble Comparisonmentioning

confidence: 99%

“…We now show that the better performance of EDeiTs is not a property of the single models, but that it rather stems from the ensembling step. To do this, we focus on the ZooLake dataset where the previous state of the art is an ensemble of CNN models 22 that consisted of EfficientNet, MobileNet and DenseNet architectures. In Table 1, we show the single-model performances of these architectures, and those of the DeiT-Base model ("Implementation" section), which is the one we used for the results in Fig.…”

mentioning

confidence: 99%

See 3 more Smart Citations

Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Kyathanahally

Hardeman

Reyes

et al. 2022

Sci Rep

View full text Add to dashboard Cite

Monitoring biodiversity is paramount to manage and protect natural resources. Collecting images of organisms over large temporal or spatial scales is a promising practice to monitor the biodiversity of natural ecosystems, providing large amounts of data with minimal interference with the environment. Deep learning models are currently used to automate classification of organisms into taxonomic units. However, imprecision in these classifiers introduces a measurement noise that is difficult to control and can significantly hinder the analysis and interpretation of data. We overcome this limitation through ensembles of Data-efficient image Transformers (DeiTs), which not only are easy to train and implement, but also significantly outperform the previous state of the art (SOTA). We validate our results on ten ecological imaging datasets of diverse origin, ranging from plankton to birds. On all the datasets, we achieve a new SOTA, with a reduction of the error with respect to the previous SOTA ranging from 29.35% to 100.00%, and often achieving performances very close to perfect classification. Ensembles of DeiTs perform better not because of superior single-model performances but rather due to smaller overlaps in the predictions by independent models and lower top-1 probabilities. This increases the benefit of ensembling, especially when using geometric averages to combine individual learners. While we only test our approach on biodiversity image datasets, our approach is generic and can be applied to any kind of images.

show abstract

Section: Ensemble Comparisonmentioning

confidence: 99%

Section: Ensemble Comparisonmentioning

confidence: 99%

Section: Ensemble Comparisonmentioning

confidence: 99%

mentioning

confidence: 99%

See 2 more Smart Citations

Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Kyathanahally

Hardeman

Reyes

et al. 2022

Sci Rep

View full text Add to dashboard Cite

show abstract

“…ZooLake This dataset consists of 17943 images of lake plankton from 35 classes acquired using Dualmagnification Scripps Plankton Camera (DSPC) in Lake Greifensee (Switzerland) between 2018 and 2020 [13,33]. The images are colored, with a black background and an uneven class distribution.…”

Section: A Datamentioning

confidence: 99%

Ensembles of Vision Transformers as a New Paradigm for Automated Classification in Ecology

Kyathanahally¹,

Hardeman²,

Reyes³

et al. 2022

Preprint

View full text Add to dashboard Cite

Monitoring biodiversity is paramount to manage and protect natural resources, particularly in times of global change. Collecting images of organisms over large temporal or spatial scales is a promising practice to monitor and study biodiversity change of natural ecosystems, providing large amounts of data with minimal interference with the environment. Deep learning models are currently used to automate classification of organisms into taxonomic units. However, imprecision in these classifiers introduce a measurement noise that is difficult to control and can significantly hinder the analysis and interpretation of data. In our study, we show that this limitation can be overcome by ensembles of Data-efficient image Transformers (DeiTs), which significantly outperform the previous state of the art (SOTA). We validate our results on a large number of ecological imaging datasets of diverse origin, and organisms of study ranging from plankton to insects, birds, dog breeds, animals in the wild, and corals. On all the data sets we test, we achieve a new SOTA, with a reduction of the error with respect to the previous SOTA ranging from 18.48% to 87.50%, depending on the data set, and often achieving performances very close to perfect classification. The main reason why ensembles of DeiTs perform better is not due to the single-model performance of DeiTs, but rather to the fact that predictions by independent models have a smaller overlap, and this maximizes the profit gained by ensembling. This positions DeiT ensembles as the best candidate for image classification in biodiversity monitoring.

show abstract

The dependence of forecasts on sampling frequency as a guide to optimizing monitoring in community ecology

Daugaard,

Merkli,

Merz

et al. 2024

Ecosphere

Self Cite

View full text Add to dashboard Cite

Facing climate change and biodiversity loss, it is critical that ecology advances so that processes, such as species interactions and dynamics, can be correctly estimated and skillfully forecasted. As different processes occur on different time scales, the sampling frequency used to record them should intuitively match these scales. Yet, the effect of data sampling frequency on ecological forecasting accuracy is understudied. Using a simple simulated dataset as a baseline and a more complex high‐frequency plankton dataset, we tested how different sampling frequencies impacted abundance forecasts of different plankton classes and the estimation of their interactions. We then investigated whether plankton growth rates and body sizes could be used to select the most appropriate sampling frequency. The simple simulated dataset showed that the optimal sampling frequency scaled positively with growth rate. This finding was not repeated in the analyses of the plankton time series, however. There, we found that a reduction in sampling frequency worsened forecasts and led us to both over‐ and underestimate plankton interactions. This suggests that forecasting can be used to determine the ideal sampling frequency in scientific and monitoring programs. A better study design will improve theoretical understanding of ecology and advance policy measures dealing with current global challenges.

show abstract

Deep Learning Classification of Lake Zooplankton

Cited by 21 publications

References 61 publications

Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Ensembles of data-efficient vision transformers as a new paradigm for automated classification in ecology

Ensembles of Vision Transformers as a New Paradigm for Automated Classification in Ecology

The dependence of forecasts on sampling frequency as a guide to optimizing monitoring in community ecology

Contact Info

Product

Resources

About