Theses and Dissertations

ORCID

https://orcid.org/0000-0002-9610-9646

Advisor

Costa, Adriana

Committee Member

Owens, Frank

Committee Member

Wiedenhoeft, Alex

Committee Member

Shmulsky, Rubin

Date of Degree

5-15-2026

Original embargo terms

Embargo 2 years

Document Type

Dissertation - Open Access

Major

Forest Resrouces (Sustainable Bioproducts)

Degree Name

Doctor of Philosophy (Ph.D.)

College

College of Forest Resources

Department

Department of Sustainable Bioproducts

Abstract

Wood identification is critical to disciplines ranging from archaeology and forestry to forensics. However, traditional anatomical methods often do not provide the finer taxonomic resolution required for species-level identification, and the declining number of trained anatomists further limits their applicability. To address these limitations, various techniques have been developed, including DNA analysis, Direct Analysis in Real Time Time-of-Flight Mass Spectrometry (DART-TOFMS), Near-Infrared Spectroscopy (NIRS), and computer vision approaches. This dissertation investigates two independent approaches: a molecular diagnostic workflow and a computer vision framework. On the molecular side, species-level discrimination within wood-anatomically indistinct groups, such as southern yellow pines, is often hindered by the loss of morphological diagnostic traits after harvest. To overcome this limitation, Chapter II details the development of a multiplex SYBR Green qPCR assay. By targeting short chloroplast regions and leveraging melt-curve analysis, this workflow demonstrates high specificity and reproducibility under the Minimum Information for Publication of Quantitative Real-Time PCR Experiment (MIQE) guidelines, providing a rapid, gel-free proof-of-concept on leaf and seed DNA that may be extensible to wood DNA after further validation. On the computer vision side, Chapters III and IV address validity and reliability in computer vision wood identification (CVWID). In Chapter III, a systematic review of 82 publications and 13 macroscopic wood image datasets reveals widespread issues, including specimen preparation and image acquisition malpractice, train-test image leakage, and inconsistent metadata. To mitigate these issues, the Wood Identification: Preparation, Acquisition, Cleaning, and Exclusivity (WI-PACE) checklist was proposed to standardize dataset creation. Building on these principles, Chapter IV introduces XyloNet, a benchmark dataset designed for a reproducible CVWID evaluation under broad anatomical variability. XyloNet incorporates standardized imaging, quality control, and data leakage prevention protocols during partitioning, supporting a deployment-oriented analysis through anatomically informed class design and specimen-level evaluation. By emphasizing quality standards and operational readiness, this dissertation provides a framework for strengthening scientific rigor in the field of wood identification.

Sponsorship (Optional)

USDA Forest Products Laboratory under Agreement No 23-JV-11111134-037

Download

Available for download on Saturday, June 10, 2028

Share

COinS