MultiWoodID: A multispecies wood identification system based on wood cross-section images
Year: 2025
Type: article
Abstract: Summary With the expansion of the global wood trade, customs inspectors increasingly face the challenge of verifying wood species on site. To equip inspectors with a rapid and accurate identification tool, an intelligent wood identification system, MultiWoodID, was developed to capture wood cross-section images and determine wood species. The system comprises a wood polisher, a portable image acquisition device, a wood identification model, a mobile application, and a web platform. The wood polisher is designed to reveal both the macroscopic and microscopic structure of wood. A macro lens attached to a smartphone magnifies the wood cross-section image to expose finer details. Owing to the self-similarity inherent in wood cross-section images, a slicing strategy was applied to partition the original image into multiple sub-images. To address the issue of a clear central region accompanied by blurred surroundings caused by the macro lens, the softmax layer was modified by integrating weighted contributions from 25 sub-images during the identification task. To accurately identify 145 wood species and reduce errors associated with varying image resolutions, a TEFPN architecture based on ResNeSt was introduced. This modification enhances the extraction of both macroscopic and local features from wood cross-section images, thereby facilitating the discrimination of similar wood species. The model achieved a precision of 99.12%, a recall of 98.58%, and an score of 98.84%. The complete identification process, from wood polishing to result display, can now be accomplished in approximately 10 minutes, representing a significant reduction in time compared to conventional wood identification methods.
Authors: Zhuoxuan Zeng, Zelin Feng, Jingjing Wang, Jun Lyu, Jiabao Jiao, Mingsheng Wang, Jun Lu, Jun Zhu, Qing Yao, Zhiping Ding
Language: English
Topic: Wood and Agarwood Research
Subfield: Organic Chemistry
Field: Chemistry
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411254747