IMACA – Automated wood identification system of Colombian timber species using convolutional neural networks

Abstract

Abstract Monitoring and controlling illegal timber trafficking remains a formidable global challenge. The timber sector faces this issue without practical and on-site support systems to facilitate these tasks, and there exists a limited availability of technological and automated tools to assist control personnel in fulfilling their responsibilities. The challenge intensifies in regions where workers possess inadequate expertise in confidently identifying the forest species involved in illegal trade. This paper introduces the architectural framework and a computational model for a digital support tool designed to recognize twenty timber species that are illicitly traded in the Colombian Amazon region. A lightweight convolutional neural network was trained using the transfer learning approach and an in-house generated dataset. The resulting model was deployed on the cloud, following Software as a Service principles, and on a portable embedded system. The prototype exhibits a classification performance exceeding 93%, successfully emulating real-world conditions in the field, including challenges such as imprecise cutting techniques, low-resolution image capture devices, and images captured at varying orientations. Furthermore, the classifier model has been incorporated into a chatbot and a low-cost microcomputer, enabling rapid responses in less than ten seconds. This integration enhances versatility, reduces the subjectivity of the identification process, supports both online and offline operation, and offers potential scalability for the entire system.