Abstract

MANTILLA RAMIREZ, Naren Arley; PORRAS GOMEZ, Iván Darío  and  SEPULVEDA-SEPULVEDA, Alexander. Wood species detection from chemical array sensors by using L1 regularization and Gaussian Mixture Models. Rev. logos cienc. tecnol. [online]. 2023, vol.15, n.1, pp.8-18.  Epub Dec 11, 2022. ISSN 2145-594X.  https://doi.org/10.22335/rlct.v15i1.1642.

Species identification from timber helps combat ilegal wood trafficking. In the present work, it is shown a method that aims to detect wood species from previously cut and stored pieces through the interaction of the volatile compounds that they emanate with an array of chemical sensors (electronic nose). The device processes the response of the chemical sensor array using linear regularization and probabilistic machine learning methods trying to resemble current biometric systems. In particular, this method includes a heuristic parameter estimation stage on the signals provided by the sensor array, followed by a variable selection stage through L1 regularization, to finally use Gaussian mixture models [GMMs] for the process of probabilistic modeling. As a result, a performance (measured by EER) of 17.5 % is obtained for the detection of four wood species; and, for the particular case of Sapán, an EER of 12 % is achieved. In conclusion, this biometric approach gives good results compared to previous works, taking into account that in the present work, the experiments are carried out in conditions that are closer to reality.

Keywords : statistical learning; dimensionality reduction; mixture models; electronic noses; wood industry.

        · abstract in Spanish | Portuguese     · text in Spanish     · Spanish ( pdf )