Anatomical traits of Cryptomeria japonica tree rings studied by wavelet convolutional neural network

Tree ring analysis is an important field of science, and is vital in modeling the environmental response system of tree growth. In most cases, analyses have been conducted using one parameter from one tree ring, e.g., ring-width, density, or ratio of stable isotopes. The information within a ring, h...

Full description

Saved in:
Bibliographic Details
Published in:IOP conference series. Earth and environmental science 2020-01, Vol.415 (1), p.12027
Main Authors: Nakajima, T, Kobayashi, K, Sugiyama, J
Format: Article
Language:English
Subjects:
Artificial neural networks
Cryptomeria japonica
Dendrochronology
Environment models
Isotopes
Neural networks
Photomicrographs
Softwoods
Stable isotopes
Tree rings
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Tree ring analysis is an important field of science, and is vital in modeling the environmental response system of tree growth. In most cases, analyses have been conducted using one parameter from one tree ring, e.g., ring-width, density, or ratio of stable isotopes. The information within a ring, however, has been less studied, although it offers many more possibilities for investigation, such as seasonal responses over shorter time scales. Therefore, to elucidate the sub-seasonal climatic response of softwood (Cryptomeria japonica), we investigate the use of a wavelet-convolutional neural network (CNN) model, which incorporates spectral information that is normally lost in conventional CNN models. This paper highlights the usefulness of the wavelet-CNN for classifying cross-sectional optical micrographs and extracting structural information specific to a calendar year. Class activation maps indicate that the dimension and position of cells in a radial file are likely to be discriminative features for the wavelet-CNN. This study shows that wavelet-CNNs have the potential to be highly effective methods for dendrochronology.
ISSN:1755-1307
1755-1315
DOI:10.1088/1755-1315/415/1/012027