348 IEEE SENSORS JOURNAL, VOL. 4, NO. 3, JUNE 2004 Fusion of Three Sensory Modalities for the Multimodal Characterization of Red Wines María Luz Rodríguez-Méndez, Alvaro A. Arrieta, Vicente Parra, Alfonso Bernal, Alberto Vegas, Sonia Villanueva, Ricardo Gutiérrez-Osuna, Member, IEEE, and José Antonio de Saja Abstract—This work represents the first attempt to develop a sensory system, specifically designed for the characterization of wines, which combines three sensory modalities: an array of gas sensors, an array of electrochemical liquid sensors, and an optical system to measure color by means of CIElab coordinates. This new analytical tool, that has been called “electronic panel,” includes not only sensors, but also hardware (injection system and electronics) and the software necessary for fusing information from the three modules. Each of the three sensory modalities (volatiles, liquids, and color) has been designed, tested, and optimized separately. The discrimination capabilities of the system have been evaluated on a database consisting of six red Spanish wines prepared using the same variety of grape (tempranillo) but differing in their geo- graphic origins and aging stages. Sensor signals from each module have been combined and analyzed using pattern recognition tech- niques. The results of this work show that the discrimination capa- bilities of the system are significantly improved when signals from each module are combined to form a multimodal feature vector. Index Terms—Electronic nose, electronic panel, electronic tongue. I. INTRODUCTION T HE LAST few years have witnessed a great deal of research in the area of artificial olfactory systems [1]–[3], with a number of efforts focusing on the discrimination of wines [4], [5]. Discrimination of alcoholic beverages has been shown to be a challenging problem due to the notorious cross selectivity of odor sensors to water and ethanol, requiring the use of preconcentration techniques to reduce these effects [6], [7]. The initial success of “electronic noses” motivated the idea of constructing an “electronic tongue” using an array of non- selective electrochemical sensors coupled with chemometric methods for the recognition and discrimination of complex liquids [8]–[10]. Early efforts in this field involved signal generation from potentiometric sensors [11]. Voltammetric sensors have also been used to classify liquid samples and Manuscript received January 22, 2003; revised November 15, 2003. This work was supported in part by the CICYT under Grant AGL2001-2104-C02-01, in part by the UE Project V Framework program CRAFT-1999-70722, and in part by NSF Career Award 9984426/0229598. The associate editor coordinating the review of this paper and approving it for publication was Prof. Thaddeus Roppel. M. L. Rodríguez-Méndez, A. A. Arrieta, and V. Parra are with the Depart- ment of Inorganic Chemistry, University of Valladolid, 47011 Valladolid, Spain (e-mail: mluz@eis.uva.es). A. Bernal, A. Vegas, S. Villanueva, and J. A. de Saja are with the Department of Condensed Matter Physics, Faculty of Sciences, University of Valladolid, 47011 Valladolid, Spain. R. Gutiérrez-Osuna is with the Department of Computer Science, Texas A&M University, College Station, TX 77843 USA. Digital Object Identifier 10.1109/JSEN.2004.824236 beverages [10], [12]–[15]. These sensors are advantageous due to their high versatility. They can be used with a range of voltages, wave forms, and electroactive materials to obtain sensing units with different selectivity. It has been demonstrated that the simultaneous utilization of electronic noses and electronic tongues can increase the amount of information extracted from the samples [4], [16]. However, the complete characterization of beverages requires taking into account their visual aspects. The objective of this work is to develop an innovative system that integrates a sensory system for the analysis of volatiles (commonly referred to as an “electronic nose”), a sensory system for the analysis of liquids (an “electronic tongue”), and a device able to measure the color of the samples (an “electronic eye”). The system has been specifically designed for the characterization and discrimination of wines. To reduce the abovementioned cross selectivity of gas sensors to water and ethanol, the proposed system combines an array of metal-oxide sensors with solid phase micro extraction (SPME). One of the novelties of the work lies in the development of a multi-channel taste sensor based on new voltammetric elec- trodes of rare-earth bisphthalocyanines and conducting poly- mers [13], [14], which produce a variety of electrochemical re- sponses when exposed to wines. In addition, their electrochemical properties can be modu- lated by introducing chemical modifications in the sensitive ma- terials [17], [18]. Finally, information about the color properties of the wine is captured in our system by measuring the trans- mittance at wavelengths representative of the psycho-physics attributes of color [19]. Finally, a common software that allows to collect the outputs of each sensor modality and to carry out the data treatment in a combined manner to form a multimodal feature vector was developed. Using this method, signals from each sensory modality were combined, and the discrimination capabilities of the integrated system were evaluated using pattern recognition software. II. EXPERIMENTAL The gas sensor array was constructed using fourteen inor- ganic sensors from Figaro and FIS. (Table I). The sensors were mounted in a stainless steel chamber with a volume of 150 ml. The analyzes were conducted with a custom-made odor-de- livery system combining the gas sensor array with a preconcen- tration system based on solid-phase microextraction (SPME). Three milliliters of the wines were placed in 10 ml vials, and 1530-437X/04$20.00 © 2004 IEEE