Stray Character Extraction And Classification Technique For Low-frequency Radiation

As a key technology of electronic confrontation, the individual identification of radiation source has attracted much attention in recent years. It principally contains two parts: feature extraction and classification recognition. Due to the hardware differences between the same types of low frequency radiation source, the transmitted signal has little signal difference and the signal feature exhibit more nonlinear and non-stationary. With the gradual increase in electronic facilities, the modern environment tends to be complicated. It is a key problem to get the access which distinguish different stray characteristics of radiation source form the signal and recognize their transmission stations through the classification effectively.This thesis principally studies the following contents: to research behavior modeling of low-frequency radiation source; to extract the stray feature from different radiation sources under the same type and the same pattern; to use classifiers to classify the extracted feature effectively.For the behavior modeling of low-frequency radiation source, this thesis mainly studies the behavior modeling of power amplifier. The deep learning theory is introduced to modeling of nonlinear system. BP-RBMs model and deep reconstruction model(DRM) are proposed by integrating deep learning theory with back-propagation(BP) neural network and Elman neural network respectively. The comparison of power amplifier modeling performance through five kind of behavioral models, Volterra-Laguerre, Kautz-Volterra, BP-RBMs, Elman neural network and deep reconstruction model.For the feature extraction of radiation source, this thesis studies the algorithm of higher order cumulant and bispectrum to extracts diagonal slice of bispectrum. A feature extraction method based on the deep learning theory is proposed. The specific algorithm of box dimension is studied by fractal theory. An algorithm is designed which is based on the diagonal slice of bispectrum and its dimension algorithm to extract the stray feature. This method can be implemented by the experiment.For the classification problems, the support vector machines(SVM) classifier and the AdaBoost algorithm are studied. A combining classifier is proposed which is based on SVM and AdaBoost algorithm. The new combining classifier is compared with SVM and AdaBoost, and gets a good recognition result.Experiments show that the presented methods are accurate and effective for feature extraction and classification of low frequency radiation source.