Combining the results of several neural network classifiers
Neural Networks
IEEE Transactions on Pattern Analysis and Machine Intelligence
Experimental evaluation of expert fusion strategies
Pattern Recognition Letters - Special issue on pattern recognition in practice VI
Cyclostationarity: half a century of research
Signal Processing
Hi-index | 0.01 |
Radar emitter recognition is of great importance in modern ELINT and ESM systems. The conventional methods for emitter recognition usually use one classifier. For specific emitter recognition, there are slight differences between the feature vectors from radars with the same type. So the recognition result of single classifier is unreliable and instable. In this paper we propose a new combining method of multiple SVM Classifiers based on Dempster-Shaffer theory. We use a new training scheme to increase the uncertainty of single classifiers by classes' combination of the training data. This training scheme is not only accords with the character of specific radar emitter recognition, but also exerts the function of DS theory. The simulation experiments on actual pulses of six radars with the same type verify the correctness and validity of this method, which can enhance the recognition rate and decrease the reject rate.