Aircraft Type Recognition In Short-wave Speech Communication

Aircraft type recognition in Short-wave speech communication is a new task in the fieldof non-cooperative communication, it applies widely both in civil use and military use, also itis meaningful for the national security. By the use of acoustic signal in the aircraft cockpit inShort-wave speech communication, we can get bigger promotion than the other recognitionmethods. First, the method is passive method to monitoring, and is difficult to be found;second, Short-wave communication spreads distant, the distance can be counted as thousandkilometers; at last, the short-wave monitoring device costs low, can be constructed andpopularized easily.First, this paper analysis non speech segment acoustic signal in the aircraft cockpit,feature extraction method of thef₀Hz skewness kurtosis wavelet packet energy entropy canbe proposed. Compared with the existing background noise feature extraction methods, themethod proposed in this paper has many advantages: in the aircraft cockpit, the characteristicsof acoustic signal which the average frequency isf₀Hz within a certain range is clear, with ahigh robustness; Wavelet packet energy entropy can present not only how much theinformation of acoustic signal in the aircraft cockpit, but also the uncertainty degree of systemstate; Skewness and kurtosis can present the Gaussian features of acoustic signal in theaircraft cockpit.Second, we analysis speech segment acoustic signal in the aircraft cockpit. There is nonon-speech segment acoustic signal in some types of aircrafts, but speech segment acousticsignal exits. And we recognize the aircraft type by no non-speech segment acoustic signal, soit is very important to remove the speech in the speech segment. Taking account of what hassaid, we present a method of removing speech by wavelet threshold which based on Empiricalmode decomposition. The method which is presented as before can remove speech in thespeech segment well, which is just left non-speech segment acoustic signal. Then we extractfeature from the non-speech segment acoustic signal that we have gotten by the method off₀Hz skewness kurtosis wavelet packet energy entropy.At last, we set up the experiment platform, by using this platform we can get thecharacters of both non-speech segment acoustic signal and the non-speech segment acoustic signal that is gotten by removing the speech in the speech segment. Then we can recognizethe aircraft type by using SVM classifiers and Naive Bayes classifiers on the base of thesedata. Comparing the results of the two experiments, we can know that Bayes classifiers’recognition result is better, and the time of learning and predicting is less. At the same time,the character of non-speech segment acoustic signal can get higher aircraft type recognitionrate than the character of non-speech segment acoustic signal that is gotten by removing thespeech in the speech segment.