A novel classification method based on AdaBoost for electromagnetic emission

Abundant characteristics information of equipment or systems could be obtained from electromagnetic emission data. In this paper, those typical characteristics, like harmonics, damped oscillations, of electromagnetic emission are classified via the adaptive boosting (Adaboost) algorithm and they are validated through measurement results. Based on the “basic emission waveform theory”, three types of the basic fundamental elements, characteristics-harmonic, narrowband and envelope-of complex emission in frequency domain, are considered in our proposed method. By taking weights combination patterns to effectively improve the classification performance of a single classifier, quite high classification accuracy could be achieved by Adaboost algorithm in our simulations. In our study, 100% precision classification accuracy of three types of characteristics could be obtained using Adaboost with 13 decision tree weak-classifiers. Compared with other classification methods, the Adaboost algorithm with decision tree weak-classifier used to classify typical characteristics of electromagnetic emission is the most accurate. At the same time, it is very effective to process the measured data. Only through the classification of multiple emission signals can identification and positioning of electromagnetic interference sources further.