Fossil charcoal particle identification and classification by two convolutional neural networks

Rehn, E., Rehn, A., and Possemiers, A. (2019) Fossil charcoal particle identification and classification by two convolutional neural networks. Quaternary Science Reviews, 226. 106038.

[img] PDF (Published Version) - Published Version
Restricted to Repository staff only

View at Publisher Website: https://doi.org/10.1016/j.quascirev.2019...
 
2


Abstract

Fire is a significant natural and cultural phenomenon, affecting spatial scales from local to global, and is represented in most palaeoenvironmental records by fossil charcoal. Analysis is resource-intensive and requires high-level expert knowledge. This study is a preliminary investigation of the application of artificial neural networks to fossil charcoal particle analysis, utilizing a U-Net variant for charcoal particle identification and VGG for particle classification by morphology. Both neural networks performed well, reaching ∼96% accuracy for particle identification and ∼75% accuracy for classification. Future work will include expansion of the training dataset, including total number of particles and number of sites. The development and application of this automated system will increase the efficiency of fossil charcoal analysis.

Item ID: 61046
Item Type: Article (Research - C1)
ISSN: 1873-457X
Keywords: Charcoal, Palaeolimnology, Palaeofire, Artificial neural networks, Quaternary
Copyright Information: © 2019 Elsevier Ltd. All rights reserved.
Funders: Australian Research Council Centre of Excellence for Australian Biodiversity and Heritage, Australian Institute of Nuclear Science and Engineering (AINSE)
Research Data: http://doi.org/10.25903/5d006c1494cf9
Date Deposited: 26 Nov 2019 23:09
FoR Codes: 04 EARTH SCIENCES > 0406 Physical Geography and Environmental Geoscience > 040606 Quaternary Environments @ 40%
05 ENVIRONMENTAL SCIENCES > 0503 Soil Sciences > 050399 Soil Sciences not elsewhere classified @ 10%
08 INFORMATION AND COMPUTING SCIENCES > 0801 Artificial Intelligence and Image Processing > 080106 Image Processing @ 50%
SEO Codes: 96 ENVIRONMENT > 9614 Soils > 961499 Soils not elsewhere classified @ 50%
97 EXPANDING KNOWLEDGE > 970104 Expanding Knowledge in the Earth Sciences @ 50%
Downloads: Total: 2
More Statistics

Actions (Repository Staff Only)

Item Control Page Item Control Page