Integration of Wood Anatomy and Artificial Intelligence: A Technological Framework Based on the UTN Xylotheque for Forensic Identification and Forest Governance in Ecuador

Traditional wood anatomy provides the gold standard for timber identification, yet its reliance on centralized laboratory infrastructure severely limits its efficacy during real-time field inspections. This study addresses a critical research question: How can physical xylotheque resources, national timber extraction registries, and edge-computing computer vision be integrated into a cohesive framework to enable robust, forensic-level wood identification at field control stations? To resolve this, we implemented a three-tier methodology: first, we audited historical records from Ecuador’s Forest Administration System (SAF) encompassing 129 commercial timber species; second, we conducted a gap analysis using the Wood Anatomy Laboratory and Xylotheque (LAMX) repository (510 cataloged samples, 2267 histological preparations) to secure botanically validated references; and third, we leveraged a curated database of high-resolution digital cross-section captures (4900 images) to evaluate CNN architectures via k-fold cross-validation and a standard 70/15/15% training/validation/testing split. Benchmarking demonstrated that the lightweight MobileNetV2 architecture achieved a global accuracy of 94.04% and an F1-score of 0.976. External field validation conducted across commercial timber yards in Ibarra confirmed an offline inference latency of just 145 ms on mid-range Android devices, proving the framework’s operational transparency and low-cost scalability. Furthermore, Explainable AI analysis using Class Activation Maps (Grad-CAM) provided visual evidence indicating that the neural network targeted diagnostic xylotomic features (vessel distribution and axial parenchyma), minimizing reliance on external environmental noise. In conclusion, this study demonstrates that hybridizing physical taxonomic reference collections with targeted edge AI models provides a scalable, transparent, and low-cost solution that successfully bridges academic research and active forest law enforcement in tropical regions.