Automated recognition by multiple convolutional neural networks of modern, fossil, intact and damaged pollen grains

Bourel, Benjamin; Marchant, Ross; Garidel‐Thoron, Thibault de; Tetard, Martin; Barboni, Doris; Gally, Yves; Beaufort, Luc

doi:10.1016/j.cageo.2020.104498

Cited by 35 publications

(41 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although significant progress has been made towards automated pollen analysis during recent years (e.g. Bourel et al., 2020; Dunker et al., 2020; Holt & Bennett, 2014; Sevillano et al., 2020), many field studies are still constrained by laborious and costly pollen identification and quantification.…”

Section: Introductionmentioning

confidence: 99%

“…Beil et al., 2008; Persson et al., 2018; Wood et al., 2017). Methods include both bright field and dark field microscopy, and can be combined with different pollen preparation methods such as staining of fresh pollen or acetolysis (Kearns & Inouye, 1993), with the latter also possible to use on fossil pollen (Bourel et al., 2020). For some applications, manual microscopy might still be preferred, but technological developments are starting to provide alternatives, including molecular methods such as meta‐barcoding (e.g.…”

Section: Introductionmentioning

confidence: 99%

“…Since then, several new studies with promising results have been published (e.g. Bourel et al., 2020; Daood et al., 2016; Dunker et al., 2020; Sevillano & Aznarte, 2018; Sevillano et al., 2020). This illustrates the general progress made in the field of machine learning and artificial intelligence, especially using deep neural networks, which are well suited for classifying two‐dimensional images (Sevillano & Aznarte, 2018).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Efficient, automated and robust pollen analysis using deep learning

et al. 2021

View full text Add to dashboard Cite

Pollen analysis is an important tool in many fields, including pollination ecology, paleoclimatology, paleoecology, honey quality control, and even medicine and forensics. However, labour‐intensive manual pollen analysis often constrains the number of samples processed or the number of pollen analysed per sample. Thus, there is a desire to develop reliable, high‐throughput, automated systems. We present an automated method for pollen analysis, based on deep learning convolutional neural networks (CNN). We scanned microscope slides with fuchsine stained, fresh pollen and automatically extracted images of all individual pollen grains. CNN models were trained on reference samples (122,000 pollen grains, from 347 flowers of 83 species of 17 families). The models were used to classify images of different pollen grains in a series of experiments. We also propose an adjustment to reduce overestimation of sample diversity in cases where samples are likely to contain few species. Accuracy of a model for 83 species was 0.98 when all samples of each species were first pooled, and then split into a training and a validation set (splitting experiment). However, accuracy was much lower (0.41) when individual reference samples from different flowers were kept separate, and one such sample was used for validation of models trained on remaining samples of the species (leave‐one‐out experiment). We therefore combined species into 28 pollen types where a new leave‐one‐out experiment revealed an overall accuracy of 0.68, and recall rates >0.90 in most pollen types. When validating against 63,650 manually identified pollen grains from 370 bumblebee samples, we obtained an accuracy of 0.79, but our adjustment procedure increased this to 0.85. Validation through splitting experiments may overestimate robustness of CNN pollen analysis in new contexts (samples). Nevertheless, our method has the potential to allow large quantities of real pollen data to be analysed with reasonable accuracy. Although compiling pollen reference libraries is time‐consuming, this is simplified by our method, and can lead to widely accessible and shareable resources for pollen analysis.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Efficient, automated and robust pollen analysis using deep learning

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Recent technological advances in image acquisition, processing and recognition have enable automated procedures, from microscopic slide field-of-view acquisition to taxonomic identification, that can ease radiolarian studies. In the early 1980s, some authors had already proposed the automatic analysis of the size and shape of a large number of digitized images of assemblages of microfossils (Budai et al, 1980) in order to investigate the variability of their morphology and use it as a palaeoenvironmental descriptor. For more than 20 years now, the CEREGE laboratory has been a pioneer in automated image acquisition and recognition for several microfossil groups.…”

Section: Introductionmentioning

confidence: 99%

“…Several workflows inspired by SYRACO and now using CNNs have been successively developed at CEREGE and applied to microfossil taxa (e.g. Marchant et al, 2020;Bourel et al, 2020). Regarding radiolarians, previous attempts have mainly focused on the identification step.…”

Section: Introductionmentioning

confidence: 99%

Technical note: A new automated radiolarian image acquisition, stacking, processing, segmentation and identification workflow

et al. 2020

Self Cite

View full text Add to dashboard Cite

Abstract. Identification of microfossils is usually done by expert taxonomists and requires time and a significant amount of systematic knowledge developed over many years. These studies require manual identification of numerous specimens in many samples under a microscope, which is very tedious and time-consuming. Furthermore, identification may differ between operators, biasing reproducibility. Recent technological advances in image acquisition, processing and recognition now enable automated procedures for this process, from microscope image acquisition to taxonomic identification. A new workflow has been developed for automated radiolarian image acquisition, stacking, processing, segmentation and identification. The protocol includes a newly proposed methodology for preparing radiolarian microscopic slides. We mount eight samples per slide, using a recently developed 3D-printed decanter that enables the random and uniform settling of particles and minimizes the loss of material. Once ready, slides are automatically imaged using a transmitted light microscope. About 4000 specimens per slide (500 per sample) are captured in digital images that include stacking techniques to improve their focus and sharpness. Automated image processing and segmentation is then performed using a custom plug-in developed for the ImageJ software. Each individual radiolarian image is automatically classified by a convolutional neural network (CNN) trained on a Neogene to Quaternary radiolarian database (currently 21 746 images, corresponding to 132 classes) using the ParticleTrieur software. The trained CNN has an overall accuracy of about 90 %. The whole procedure, including the image acquisition, stacking, processing, segmentation and recognition, is entirely automated via a LabVIEW interface, and it takes approximately 1 h per sample. Census data count and classified radiolarian images are then automatically exported and saved. This new workflow paves the way for the analysis of long-term, radiolarian-based palaeoclimatic records from siliceous-remnant-bearing samples.

show abstract