Resumen de Open tools for dendrochronology. Advances in sample digitization and deep learning methods for image segmentation
Dendrochronological techniques are paramount in forest research. The current climate change scenario and the central role of forests in biogeophysical cycles enforce the importance of novel techniques to get accurate data from trees and their relationship with the environment in faster ways. Recent technological advances and the place of open source software and hardware are making free, user-developed tools for forest research available to the research community. The aim of this Ph.D. thesis is the development of tools for image acquisition and data collection in dendrochronology based on open source software and hardware. Thus, four different tools for dendrochronological research are presented in five different chapters.
The first chapter focuses on the development of a do-it-yourself tool based on open source hardware for image acquisition and wood sample digitization at high resolution. We used open hardware equipment from Arduino and Python programming to develop CaptuRING and published the entire free open source tool as: “CaptuRING: A Do-It-Yourself tool for wood sample digitization” in Methods in Ecology and Evolution, 2022; 13:1185–1191. Furthermore, the original software was registered in the Registro General de Propiedad Intelectual (00/2022/737) of Ministerio de Cultura y Deporte (Spain).
The second chapter presents “How to build and install your own CaptuRING”. This contribution presents a series of videos with a step-by-step guide to promote the use of CaptuRING in the research community. The manuscript and related videos have been submitted for publication.
The third chapter describes ρ-MtreeRing. This free and open-source software, which is written in R, analyzes X-ray films from dendrochronological samples to get microdensity values automatically segmented through a graphical user interface. The open source tool and manuscript are published as: “ρ-MtreeRing. A graphical user interface for X-ray microdensity analysis” in Forests. 2021; 12(10):1405.
The fourth chapter describes the potential of deep learning methods to automatically segment xylem vessels. We trained three different convolutional neural networks to segment vessels in stained wood microsections using the Keras framework in Python. Our results demonstrate the potential of these techniques to automatically segment xylem vessels and overcome derived problems from image illumination, which hamper segmentation using classical image segmentation methods. The manuscript is published as “Convolutional neural networks for segmenting xylem vessels in stained cross-sectional images” in: Neural Computing & Applications, 2020; 32:17927–17939.
The fifth chapter develops an algorithm to delineate annual ring limits in stained wood microsections of a species with diffuse porous wood using convolutional neural networks. We used Python for image processing and the Keras framework for the algorithm training. The results show the ability of this techniques to obtain accurate tree ring segmentation for quantitative wood anatomy, reaching similar or even outperforming conventional manual delimitation in most of the evaluated cases. The results of this chapter will be presented in the manuscript “Deep Learning for ring bordering in stained cross-sectional images”.
This PhD Thesis presents four open source tools to get accurate information from wood features to unveil how trees respond to the environment. From digitization at macroscopic perspective, automatic data collection and the development of feature segmentation on microscopic samples. The presented four novel dendrochronological tools based on open source software facilitates forest research in the current climate change scenario.
