| Infrasound is a kind of low frequency signal that human can’t hear, its frequency range from 0.01 ~ 20 Hz. Many events will produce infrasound signals, such as tsunami, aurora, volcanic eruptions, earthquakes in nature, and the nuclear blast, rocket, artillery shooting in human activities. Because of the different mechanism of various events in producing infrasound, the energy of signals also distribute in different frequency. Then we can analyze the characteristics of detected infrasound signals to complete the classification. Thus we can achieve the purpose of infrasound signal classification.The research of characteristic extraction and classification has been a hot spot in the field of infrasound signal processing. As we can see from the previous researches, infrasound classification consists of two parts: feature extraction and classification recognition. The key to the classification of infrasonic events is how to extract effective feature vectors from a signal. Effective feature extraction techniques are the foundation of the classification of infrasound events. On the hand of feature extraction, the research focuses on mining the characteristics that can distinguish the signal category and how to extract these characteristics. On the other hand, the main task of classification is to study the algorithm and structure to design accurate and efficient classifiers.This paper will study the infrasound signal event for the purpose of natural disasters classification. Research is focused on technical of feature extraction methods and pattern recognition classification algorithm. The main contents of this paper are as follows:Firstly, a detailed presentation was made for the background and significance of classification and recognition for infrasound signal of the natural disasters. By summarizing the research status of infrasound signals, the theory of infrasound signal classification and recognition were elaborated in the paper, illustrating the various components of the respective roles of the recognition system. The research focuses on the comparative analysis of the advantages and disadvantages of various sub-acoustic signal feature extraction algorithm. Then, three technical methods were used to extract the feature vectors of infrasound signals. The article also studied two classification algorithms used in this article: support vector machines and artificial neural network. There is a detailed analysis of the theoretical basis of these two algorithms, implementation, advantages and disadvantages in the article. The two methods were used on infrasound signal classification.The data used in this research come from CTBTO National Data Center and the Chinese Academy of Dongchuan Debris Flow Observation and Research Station. The effectiveness of several feature extraction methods were discussed in the study. Finally, future research directions are summarized in the paper. |
