Comparison of computer vision models in application to pollen classification using light scattering

Daunys, Gintautas; Šukienė, Laura; Vaitkevičius, Lukas; Valiulis, Gediminas; Sofiev, M.; Šaulienė, Ingrida

doi:10.1007/s10453-022-09769-0

Cited by 4 publications

(3 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The ResNet architecture with shortcut connections was employed due to its demonstrated superior performance in classifying pollen based on Rapid-E measurements (Matavulj et al, 2023, Daunys et al, 2022. Due to variability of input data, a variation of 18-layer ResNet model was implemented (i.e.…”

Section: Creating Classification Algorithmmentioning

confidence: 99%

Classification accuracy and compatibility across devices of a new Rapid-E+ flow cytometer

Sikoparija,

Matavulj,

Simovic

et al. 2024

Preprint

View full text Add to dashboard Cite

Abstract. The study evaluated a new model of a Plair SA air flow cytometer, Rapid-E+, and assessed its suitability for airborne pollen monitoring within operational networks. Key features of the new model are compared with the previous one, Rapid-E. A machine learning algorithm is constructed and evaluated for (i) classification of reference pollen types in laboratory conditions and (ii) monitoring in real-life field campaigns. The second goal of the study was to evaluate the device usability in forthcoming monitoring networks, which would require similarity and reproducibility of the measurement signal across devices. We employed three devices and analysed (dis-)similarities of their measurements in laboratory conditions. The lab evaluation showed similar recognition performance as that of Rapid-E, but field measurements in conditions when several pollen types are present in the air simultaneously, showed a notably lower agreement of Rapid-E+ with manual Hirst-type observations than those of the older model. An exception was the total-pollen measurements. Comparison across the Rapid-E+ devices revealed noticeable differences in fluorescence measurements between the three devices tested. As a result, application of the recognition algorithm trained on the data of one device to another one led to large errors. The study confirmed the potential of the fluorescence measurements for discrimination between different pollen classes, but each monitor needed to be trained individually to achieve acceptable skills. A large uncertainty of fluorescence measurements and their variability between different devices need to be addressed to improve the device usability.

show abstract

Section: Creating Classification Algorithmmentioning

confidence: 99%

Classification accuracy and compatibility across devices of a new Rapid-E+ flow cytometer

Sikoparija,

Matavulj,

Simovic

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Hirst samplers), multispectral imaging flow cytometry (Dunker et al, 2021), and dedicated airborne particle identifiers (e.g. Daunys et al, 2022; Matavulj et al, 2023). A first network of autonomous airborne pollen detectors, known as the e-PIN system, has recently been established in Bavaria (https://epin.lgl.bayern.de/pollenflug-aktuell).…”

Section: Introductionmentioning

confidence: 99%

A trainable object finder, selector and identifier for pollen, spores and other things: A step towards automated pollen recognition in lake sediments

Theuerkauf,

Siradze,

Gillert

2023

The Holocene

View full text Add to dashboard Cite

Pollen records are the most important proxy for reconstructing past terrestrial vegetation. While new approaches for improved quantitative interpretation of pollen data have been developed over the last decades, the availability of pollen records remains mostly limited because pollen samples are still analysed manually, which is a time-consuming task and requires extensive training. Here, we present an approach for automated recognition of pollen and spores from lake sediments using deep convolutional neural networks and machine learning. The approach includes two stages. The detector first locates pollen and spores in the sample matrix, and the classifier then classifies these objects. We have trained the approach on two pollen datasets from two lakes in north-eastern Germany. So far, our approach is able to automatically recognise 10 pollen types and Lycopodium spores with high accuracy. As soon as more training data are available, more pollen and spore types can be added. The preparation of training data, the training of the neural networks and their application are accessible via a freely available, browser-based user interface called TOFSI.

show abstract

“…Two papers then describe the further developments of the automatic identification algorithms. Daunys et al, (2022)…”

mentioning

confidence: 99%

Editorial: AutoPollen special issue

Tummon,

Clot

2024

Aerobiologia

View full text Add to dashboard Cite

Comparison of computer vision models in application to pollen classification using light scattering

Cited by 4 publications

References 40 publications

Classification accuracy and compatibility across devices of a new Rapid-E+ flow cytometer

Classification accuracy and compatibility across devices of a new Rapid-E+ flow cytometer

A trainable object finder, selector and identifier for pollen, spores and other things: A step towards automated pollen recognition in lake sediments

Editorial: AutoPollen special issue

Contact Info

Product

Resources

About