Welcome to Acta Armamentarii ! Today is Share:

Acta Armamentarii ›› 2022, Vol. 43 ›› Issue (12): 3132-3141.doi: 10.12382/bgxb.2021.0714

• Paper • Previous Articles     Next Articles

Specific Emitter Identification of Twin Radiation Sources Based on Parallel Classifier Ensemble Learning Using Ant ColonyFeature Selection

XU Yuxin1,2, GU Chumei1,2, CAO Jianjun2, XU Jinyong2, WEI Zhihu2   

  1. (1.College of Computer Science and Technology, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu, China; 2.The 63rd Research Institute, National University of Defense Technology, Nanjing 210007, Jiangsu, China)
  • Online:2022-05-19

Abstract: To improve the accuracy and reliability of specific emitter identification, the twin specific emitter identification problem is defined and studied for the first time, and a twin specific emitter identification method based on parallel classification ensemble learning using ant colony feature selection is proposed. The difference between the distribution matrixes of the output results of different classifiers is determined using the Pearson correlation coefficient method. The design model is established for the highest classification accuracy, greatest difference between classifiers, and smallest size of input feature subsets of the parallel classifier based on ant colony feature selection. Meanwhile, combined with the characteristics of the parallel classifier, an ant colony algorithm is designed to solve the model. The weight of each sub-classifier is decided based on the degree of difference and reliability. The more different and reliable the sub-classifier is, the greater the weight will be, The final decision is derived from the weighted sum of classifiers with different weights. To verify the superiority of this method, the parallel classifier based on ant colony feature selection, single classifier, and Adaboost are compared using three groups of data, namely original radio station signals, data with 10 dB white Gaussian noise added, and data with and 5 dB white Gaussian noise added. The experimental results show that the classification accuracy of the proposed model is 88.70%, 76.70% and 64.80% respectively, all outperforming the traditional single classifier, Adaboost algorithm and Random Forest.

Key words: featureselection, supportvectormachine, ensemblelearning, antcolonyalgorithm, binaryclassification

CLC Number: