Effect of Different Kernel Functions on Hazardous Liquid Detection Using a Patch Antenna and Support Vector Machines

Abstract

Microwave spectroscopy method has become widespread in many applications including liquid classification. In this study, a microwave spectroscopy system that can classify liquids without opening the lid of their containers is proposed. Thus, the operators are prevented from being exposed to harmful substances and wasting time. Everyday liquids such as carbonated drinks, fruit juices, shampoo, cream and alcoholic beverages and hazardous liquids were characterized remotely by the microwave spectroscopy method in which spectroscopic signatures of a total of 52 liquids were used. In order to be able to classify liquids with the highest accuracy, it is also important to determine the most suitable measurement system as well as the correct selection of the classification algorithm and algorithm parameters that show the best performance. In this study, Support Vector Machines algorithm, which is a very successful algorithm in separating binary classes, is used. In addition, the effects of the algorithm on the classification performance have been examined using different kernel functions and cross-validation technique has been used for the performance analysis. As a result of the performance analysis, it is seen that up to 100% success can be achieved when linear or polynomial kernel functions have been preferred.