In 2018, global trade in forest products represented a value chain of more than 550 billion USD¹ and was at the highest volume since record-keeping began in 1947 (Food and Agriculture Organization, 2018). This value chain includes logs, timbers, dressed lumber, veneers, finished products, comminuted wood products, pulp and pulp-derived products, wood fuel, and charcoal, among others. Illegal logging accounts for 15–30% of the global timber supply chain (Nellemann, 2012), resulting in lost revenue for source countries, governmental corruption, and unregulated degradation of forest lands. Of the illegal trade in timber, it is estimated that 80% is controlled by transnational criminal enterprises (Nellemann, 2012), making illegal logging the fourth most lucrative form of transnational crime after counterfeiting, drug trafficking, and human trafficking, and the most profitable form of transnational natural resource crime (May, 2017).

In part as a result of the global scale of illegal logging and its ties to transnational organized crime, industrial compliance with and governmental enforcement of laws and regulations governing trade in wood and wood-derived products have remained an international priority. These include the Convention on the International Trade in Endangered Species (CITES, 27 U.S.T. § 1087), the Lacey Act (18 U.S.C. § 42–43; 16 U.S.C. § 3371–3378), the European Union Timber Regulation (EUTR, No. 995/2010), Australia’s Illegal Logging Protection Act (2012), and Illegal Logging Protection Regulation (2014). There is also growing interest in “greening” the charcoal value chain (van Dam, 2017), which directly impacts the energy needs and livelihoods of one-third of the world’s population (Food and Agriculture Organization, 2019). Research and technology development in support of law enforcement and industrial compliance have emphasized predominantly laboratory-based approaches [as reviewed in (Dormontt et al., 2015; International Consortium on Combating Wildlife Crime, 2016; Schmitz et al., 2019)], but the first (and in some jurisdictions the only) step in the enforcement of provisions against illegal logging is identification or screening of products in the field, at ports, border crossings, or other points of control.

The current state-of-the-art for routine field screening of wood across the world is an entirely human enterprise using naked eye and hand lens observation of wood anatomical features (Miller et al., 2002; Koch et al., 2011; Wiedenhoeft, 2011; Yin et al., 2016;² Ruffinatto and Crivellaro, 2019). Field screening of wood is severely limited by the dearth of human expertise in forensic wood analysis (Wiedenhoeft et al., 2019), and there is even less field expertise for charcoal. Affordable and scalable technologies that can either dramatically extend or obviate the need for human expertise clearly have value in solving the global field-screening limitation and effective evidence-based policy development for compliance or enforcement will require context-dependent modifications to the adopted technology.

To move away from dependence on, or to complement, human expertise, various authors have advocated for computer vision based approaches to wood and charcoal identification. Several proof-of-concept systems have been reported, relying either on laboratory-acquired images (Khalid et al., 2008; Wang et al., 2013; Filho et al., 2014; Muniz et al., 2016; Barmpoutis et al., 2018; Andrade et al., 2019), or field-acquired cell phone images (Tang et al., 2018) that are relatively variable in terms of chromatic control, total magnification, spherical aberration, and other data-quality factors reviewed in Hermanson and Wiedenhoeft (2011), with two notable forays into controlling these factors for field imaging (Hermanson et al., 2019; Andrade et al., 2020). Computer vision based wood (Ravindran et al., 2018) and charcoal identification is appealing because it is affordable (Ravindran and Wiedenhoeft, 2020) and therefore scalable, operates on an accepted source of variability in wood, its anatomy, and for wood has demonstrated potential for real-world field deployment (Ravindran et al., 2019). Realization of practical, context-specific, field deployment strategies requires a high quality computer vision platform that is affordable, flexible, and open source.

Here we present the XyloTron, the world’s first complete, open-source, do-it-yourself platform for imaging, identification, and metrology of materials exhibiting useful macroscopic variability, such as wood and charcoal (Figure 1). The XyloTron provides controlled visible light and UV illumination capability, continuously-adjustable illumination positioning, and software to control the device, capture images, and deploy trained classification models for field screening. Compared to laboratory-based methods the XyloTron exists at an affordable and scalable price point—less than 1,500 USD per unit—such that it can be deployed in the developing world (Ravindran et al., 2019), in research institutions with modest budgets, and even in classrooms. We demonstrate the capabilities of the XyloTron for two applications, namely wood identification and charcoal identification—to the best of our knowledge the first time a unified field deployable system has been used for both applications. In order to foster an ecosystem around the technology we open-source the hardware design and software applications along with the trained models at https://github.com/fpl-xylotron so the platform can be adapted to meet specific contextual needs as it is adopted.

The wood identification model performance vastly exceeds the performance of trained field personnel and indeed approaches or exceeds expected field performance of forensic wood anatomy experts. If human XyloTron users of this model would employ UV illumination, performance would be further enhanced. In Figure 5 (left), misclassification of only four specimens are not solved by such a human-hybrid approach employing UV illumination: Swietenia-Detarium, Dialium-Ulmus americana, Nauclea-Tectona grandis, and Tectona grandis-Morus. In other words, the effective accuracy of a human-hybrid version of our model incorporating UV illumination increases from 97.7 to 99.1%. Field accuracy at this level distinctly exceeds even the best-performing experts in the United States when performance was evaluated at the genus level (Wiedenhoeft et al., 2019).

To the best of our knowledge, there is no program anywhere in the world to train field personnel to inspect or identify charcoal despite the fact that globally the charcoal sector generates income for more than 40 million people and caters to the energy needs of more than one-third of the world’s population, predominantly in the developing world (Food and Agriculture Organization, 2019). In the absence of an existing field identification program for charcoal we cannot directly compare our model accuracy to field inspectors. By providing a highly accurate (98.7%), field-deployable, proof-of-concept model for six classes of lump charcoal that only confuses anatomically similar charcoals, we deliver the ability to inspect and verify materials that heretofore could only be assessed reliably in the laboratory or a limited group of expert wood anatomists. Ongoing work³ in our laboratory is addressing the breadth of charcoal taxa currently identifiable with the XyloTron, which is expected to more adequately sample the charcoals that represent the remaining 26% of the FSC-certified lump charcoal in the EU market.

While our prior and current work with the XyloTron platform has emphasized solid wood and charcoal, we are also developing applications for wood veneers, various types of synthetic products, and field metrology. As demonstrated, the XyloTron is well-suited to capture macroscopic images of any suitable substrate with interesting macroscopic variation (Figure 6), and we invite researchers in a range of disciplines to adapt and enhance the platform for their applications by adding additional functionalities (e.g. multi-flash imaging, multi-color illumination, and alternative machine vision models).

By providing a field-deployable system able to image and identify solid wood and charcoal, we take a step toward providing law enforcement and industrial compliance officers with the tools needed to verify wood and charcoal supply chains. Further, the open-source XyloTron platform can be readily adopted for other materials with macroscopic variability, or adapted and modified for new use-cases.

The software apps for image dataset collection and trained model deployment along with the weights of the trained model will be made available at https://github.com/fpl-xylotron. The code for model training will be available from the corresponding author on reasonable request.

The raw data supporting the conclusions of this article will be made available by the authors on reasonable request.

PR developed the machine learning analyses. AW and BT adapted/designed/developed the XyloTron movable lighting array, PCBs, and electronics. AW and RS conducted forensic analysis of charcoal and established the scope of the identification models. PR and AW conducted data analysis, synthesis, and wrote the paper.

This work was supported in part by a grant from the US Department of State via Interagency Agreement number 19318814Y0010 to AW and in part by research funding from the Forest Stewardship Council to AW. PR was partially supported by the Wisconsin Idea Baldwin Grant.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.