FULLY EVOLVED KERNEL METHOD EMPLOYING SVM ASSESSMENT FOR FEATURE COMPUTATION FROM MULTISENSOR SIGNALS

Abstract

The design of intelligent sensor systems for pattern recognition applications requires sophisticated methods from conventional signal processing and computational intelligence. A significant part of the overall system architecture still has to be manually elaborated in a tedious and time consuming process by an experienced designer for each new application or modification. Clearly, an automated method for auto-configuration of sensor systems would be attractive. This paper expands our previous works on automatic design of multisensor systems applying optimized Gaussian kernel for feature computation. The purpose of optimizing feature computation in the design of sensor systems is to increase the classification accuracy and the computation efficiency as well as to reduce storage and computational requirements in embedded sensor systems. Here, we investigate the feature computation of Gaussian kernel assessed by support vector machine (SVM) using genetic algorithms (GA). The experiment results are compared with the previous work using k-nearest neighbor (k-NN) and LDA/k-NN. Our experiments are tested with gas-sensor benchmark data. From the experimental results, we verify that SVM can achieve better results than k-NN and LDA/k-NN methods.