IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 53, NO. 1, FEBRUARY2006 313 A Fuzzy-Similarity-Based Self-Organized Network Inspired by Immune Algorithm for Three-Mixture-Fragrance Recognition Muhammad Rahmat Widyanto, Student Member, IEEE, Benyamin Kusumoputro, Hajime Nobuhara, Member,IEEE, Kazuhiko Kawamoto, and Kaoru Hirota, Member,IEEE Abstract—A fuzzy-similarity-based self-organized network in- spired by immune algorithm (F-SONIA) is proposed in order to develop an artificial odor discrimination system for three- mixture-fragrance recognition. It can deal with an uncertainty in frequency measurements, which is inherent in odor acquisition devices, by employing a fuzzy similarity. Mathematical analysis shows that the use of the fuzzy similarity results on a higher dissimilarity between fragrance classes, therefore, the recognition accuracy is improved and the learning time is reduced. Exper- iments show that F-SONIA improves recognition accuracy of SONIA by 3%–9% and the previously developed artificial odor discrimination system by 14%–25%. In addition, the learning time of F-SONIA is three times faster than that of SONIA. Index Terms—Artificial odor discrimination, fuzzy similar- ity, immune algorithm, self-organized network, three-mixture- fragrance problem. I. I NTRODUCTION O DOR discrimination is required to control the qualities in a variety of industrial fields, e.g., food and beverage industries, cosmetics and perfume industries. Conventionally, odors are discriminated by trained persons based on their human sensory system. The human sensory is, however, un- avoidably affected by the state of the health and the mood of the inspector, resulting in discrepancies among them. Accordingly, an artificial odor discrimination system [1] has been developed to replace the human sensory system. The artificial odor discrimination system shows a high- recognition accuracy to classify pure fragrances as well as two-mixture fragrances [1], but the recognition accuracy for three-mixture fragrances is a subject to improve due to a higher problem complexity. To improve the recognition accuracy for three-mixture fragrances, the backpropagation (BP)-based self- organized network inspired by immune algorithm (SONIA) [2], [3] is used as pattern classifier. SONIA cannot, however, deal with an uncertainty in frequency measurements during odor data acquisition. This makes the learning convergence Manuscript received October 30, 2003; revised December 30, 2004. Abstract published on the Internet November 25, 2005. M. R. Widyanto, H. Nobuhara, K. Kawamoto, and K. Hirota are with the Department of Computational Intelligence and Systems Science, Tokyo Institute of Technology, Yokohama 226-8502, Japan (e-mail: widyanto@ hrt.dis.titech.ac.jp). B. Kusumoputro is with the Faculty of Computer Science, University of Indonesia, Depok 16424, Indonesia. Digital Object Identifier 10.1109/TIE.2005.862212 of SONIA is slow, and it cannot reach the optimum recog- nition accuracy. To further improve the capability of SONIA, the use of a fuzzy similarity [4] instead of the Euclidean distance is proposed. The proposed method is called fuzzy-similarity- based self-organized network inspired by immune algorithm (F-SONIA). The minimum, the mean, and the maximum val- ues of fragrance data acquisition are used to form triangular fuzzy numbers. Then, the fuzzy similarity measure is used to define the relationship between fragrance inputs and connection strengths of hidden units. The fuzzy similarity is defined as the maximum value of the intersection region between triangular fuzzy sets of input vectors and the connection strengths of hidden units. The use of the fuzzy similarity results on a higher dissimilarity between fragrance classes, therefore, the recogni- tion accuracy is improved and the learning time is reduced. Ex- periments on three data sets of three-mixture vegetal fragrances show that F-SONIA improves the recognition accuracy of SONIA by 3%–9%. In addition, the learning time of F-SONIA is three times faster than that of SONIA. Compared to the previously developed artificial odor discrimination system that used fuzzy learning vector quantization (FLVQ) [5] as pattern classifier, the recognition accuracy is increased by 14%–25%. In Section II, the scheme of the artificial odor discrimination system is mentioned. The F-SONIA is proposed in Section III. Experimental results on the three-mixture-fragrance problem are shown in Section IV. A mathematical analysis of F-SONIA and its comparison with SONIA are summarized in Section V. II. ARTIFICIAL ODOR DISCRIMINATION SYSTEM The artificial odor discrimination system consists of three subsystems, i.e., a sensory system, a frequency counter system, and a neural network as a pattern classifier system. The sensory system and the frequency counter system are used to measure frequency changes during data acquisition, and the pattern classifier system is used to discriminate odor characteristics obtained by the other systems. Fig. 1 shows a diagram of the artificial odor discrimination system. The sensory system used is quartz-resonator crystals that are constructed by sensitive thin chemical membranes. When odorant molecules are absorbed onto the membranes, the reso- nance frequency of the crystals will decrease significantly and return to the normal resonance frequency after the deabsorption process. The change of the frequency is proportional to the 0278-0046/$20.00 © 2006 IEEE