Single Chip Implementation of Support Vector Machine Based Bi - Classifier

This paper proposes the implementation of Support Vector Machine (SVM) on a single chip (dsPIC), which makes it suitable for standalone portable applications. The usual practice is to implement SVMs using general purpose computers, since its implementation demands fairly large amount of memory [1]. SVM implementation in this work uses Sequential Minimal Optimization (SMO) [1], with necessary modifications to heed of its limitations for the single IC implementation, as the learning algorithm. Methods for a better pace in the training process of the machine are also proposed. Applications of this work include use of independent and cheap SVM based classification solution in pattern recognition[2], fault detection of motors & other systems, character recognition[3], finger print reading etc. in industries, house hold equipment and portable devices. The detection of unbalanced fault condition of an induction motor power supply serves as a test case to verify the proposed SVM implementation in dsPIC. The implemented system was tested as an online real time classifier, which classifies balanced and unbalanced voltage conditions in a 3 phase line very effectively, proving the SVM algorithm implementation.