New particle formation event detection with convolutional neural networks

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F67985858%3A_____%2F24%3A00586964" target="_blank" >RIV/67985858:_____/24:00586964 - isvavai.cz</a>

Result on the web

<a href="https://www.sciencedirect.com/science/article/pii/S1352231024001626?via%3Dihub" target="_blank" >https://www.sciencedirect.com/science/article/pii/S1352231024001626?via%3Dihub</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.atmosenv.2024.120487" target="_blank" >10.1016/j.atmosenv.2024.120487</a>

Result language

angličtina

Original language name

New particle formation event detection with convolutional neural networks

Original language description

New aerosol particle formation (NPF) events play a significant role in altering aerosol concentrations and dispersion within the atmosphere, making them vital for both climate and air quality research. The primary objective of investigating NPF events is to precisely determine their occurrence dates. In this study, we introduced the ConvNeXt model for the first time to identify NPF events, and compared its performance with two other deep learning models, EfficientNet and Swin Transformer. Our main aim was to automate an objective identification and classification of NPF events accurately. All three models employed transfer learning to effectively capture critical features associated with NPF. Our results demonstrated that the ConvNeXt model significantly outperformed the other models, achieving an impressive accuracy rate of 95.3% on event days, surpassing EfficientNet (92.8%) and Swin Transformer (94.9%). Furthermore, we performed tests using different ConvNeXt variants (ConvNeXt-T/S/B/L/XL) and different pre-training weights, revealing that different configurations of ConvNeXt models exhibited improved NPF event recognition capabilities. Finally, we conducted generalizability experiments using the ConvNeXt-XL model, achieving the highest accuracy of 96.4% on event days. This study not only underscores the recognition prowess of ConvNeXt models but also highlights their practical utility in accurately detecting NPF events in real-world scenarios. This contribution aids in advancing Atmospheric our comprehension of aerosol dynamics in atmospheric environments, providing valuable insights for climate and air quality research.

Czech name

—

Czech description

—

Type

J_imp - Article in a specialist periodical, which is included in the Web of Science database

CEP classification

—

OECD FORD branch

10509 - Meteorology and atmospheric sciences

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2024

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

Atmospheric Environment

ISSN

1352-2310

e-ISSN

1873-2844

Volume of the periodical

327

Issue of the periodical within the volume

15 June

Country of publishing house

GB - UNITED KINGDOM

Number of pages

9

Pages from-to

120487

UT code for WoS article

001224901200001

EID of the result in the Scopus database

2-s2.0-85189433032