Odor of Tropical Timber 2

Published: 08-12-2020| Version 2 | DOI: 10.17632/gvwr8hmv2t.2
Alexander Sepúlveda-Sepúlveda


The data is used for timber species classification and verification from their aromas. Timber samples were collected from wood deposits, making data collection closer to practical situations. The measurements of the volatile compounds were made by using an array of 16 chemical sensors, whose curves are the inputs to a pattern recognition based system. Support vector machine is used for Classification; and, Gaussian mixture modeling with Universal Background Model is used for timber verification from the odor. Sensors: SENSOR(j) BRAND REF 1 HANWEI MQ-2 2 HANWEI MQ-3 3 HANWEI MQ-4 4 HANWEI MQ-6 5 HANWEI MQ-7 6 HANWEI MQ-8 7 HANWEI MQ-135 8 HANWEI MQ-9 9 FIGARO TGS-832 10 HANWEI MQ-6 11 FIGARO TGS-823 12 FIGARO TGS-816 13 FIGARO TGS-822 14 FIGARO TGS-813 15 FIGARO TGS-826 16 HANWEI MQ-3 These are de features: * Ginicialj (G0) : initial conductance value, mean of the first 100 samples of the total response. * Gfinalj (Gf) : final conductance value, mean of the last 50 samples of the phase 2 of the total response. * Gmaxj (Gmax) : maximum conductance value. * Gminj (Gmin) : minimum conductance value. * Poloj (A) : pole location, corresponding to an adjusted first-order auto-regressive model [1]: * Gananciaj(B) : gain coefficient of the same adjusted first-order auto-regressive model [1]; [1] Mantilla-Ramirez, Naren A.; Ortega-Boada, H.; Paja-Sarria, M. and Sepúlveda-Sepúlveda, A. (2020). A Low Cost Electronic Nose with a GMM-UBM Approach for Wood Species Verification.In Proceedings of the 9th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM, ISBN 978-989-758-397-1, ISSN 2184-4313, pages 333-341. DOI: 10.5220/0008978003330341


Steps to reproduce

Procedure for data collecting: A total of 309 samples (woodblocks) of different wood species were taken from wood timber stocks of different cities in the Gran Santander region in Colombia. Before taking samples and perform all the measurement experiment, it was necessary to develop two previous tasks: e-nose preparation and sample preparation. First, the e-nose is turned on for one hour so that the sensors reach their steady-state operation in the corresponding environment. Later, each sample (woodblock) is prepared by brushing it 20 times with a wooden brush, and the resulting material is discarded to eliminate possible contamination by contact with another sample, or interference with other elements. Then perform the following procedure as described below for each timber sample: • Brush the sample 20 more times and take approximately 1 cm3 of the resulting wood chip. • Sense (sniff) the sample with the e-nose. Asresult, 16 response curves, corresponding to the conductance variations of each sensor in the matrix array, are obtained. This group of curves is known as the smell-print of the wood sample. • Let the sensors rest during 5 minutes, allowing the entry of airflow generated by a fan, in order to avoid previous trials interfere with the current trial. Each response curve was taken at a sampling period of 270 ms, following three steps. First, the sensors react to air for 100 samples; then, the corresponding wood chips are placed inside sensors chamber during 300 samples; finally, the wood chips are removed and the sensors are exposed to the air again with residual air for additional 100 samples.