Multivariate preprocessing techniques towards optimising response of fused sensor from electronic nose and electronic tongue

With the advent of electronic nose and electronic tongue, instrumental quality estimation of black tea has become possible. Electronic nose detects tea samples on the basis of aroma and taste can be detected by electronic tongue. But to develop human like quality judgment system, both the sensor responses must be considered. The objective of this work is to find optimized dataset by combining the sensor responses of e-nose and e-tongue so that the improved classification of tea can be done. Like individual sensor systems the merged sensor response should be preprocessed with suitable preprocessing method before feeding the data set to the classifier. In order to find the optimized dataset from both the systems, several multivariate preprocessing techniques are explored in this study. The quantitative measurements of preprocessing techniques applied to individual sensor system as well as fused dataset are estimated first with cluster separability measure (SI) and Dunn index (DI) calculation. Then each preprocessed dataset are tested with neural network based classifier. For getting optimized merged sensor response two methods of data merging have been attempted in this work. One by directly merging the raw sensor responses and then tested with preprocessing techniques and another is to find the optimized response from individual sensor systems and then to merge those optimized responses. The results show that the second method of fusion gives much better performance than direct merging of sensor responses.