Research On Pattern Classification System Of Small Sample Infrasound Signal

In daily life,many natural phenomena and human activities may produce a low-frequency signal that the human ear cannot perceive.This signal is an infrasound wave,and its frequency is usually between 0.002 and 20 Hz.With the deepening of research on infrasound signals,the characteristic information exhibited by infrasound signals has also been given more and more important significance.In 1996,the Comprehensive Nuclear Test Ban Treaty(CTBT)was established,and it was determined that the infrasound signal is one of the four kinds of effective ways to detect nuclear tests,thus the infrasound research is pushed to a new height.Pattern recognition theory has received extensive attention in many scientific and technical fields.In the field of infrasound monitoring,there are a large number of natural environmental noise signals in the infrasound signals collected by the monitoring system.The main task of classification and recognition work is to extract one type of infrasound signals from strong background interference noise signals,and to distinguish from other types of infrasound events,which has always been the focus and difficulty of infrasound signal research.Different from the pattern recognition in other fields,the infrasound classification is restricted by two factors.First,the infrasound signals such as meteorite landing,missile explosion and rocket launch are less likely to occur than other acoustic signals.The sample size is scarce,and the classification method based on the large number of sample models is hindered in the infrasound signal.Secondly,due to the low frequency characteristics of the infrasound signal itself,it is susceptible to background noise such as ambient wind.These factors all increase the difficulty of high accuracy classification.In this paper,based on the above analysis,the construction and preprocessing of the data sets,feature extraction and pattern recognition of infrasound signals from the aspects of the universality and particularity of infrasound signal properties are studied.To achieve the purpose of improving the recognition accuracy of infrasound events under the premise of small sample data volume.The main work of this thesis is as follows:The paper uses the infrasound signals generated by earthquake,chemical explosion,meteorite landing and rocket launching as the data base,and preprocesses on the data to reduce the background noise to improves signal quality.Then,combining the three feature extraction techniques,which is the feature extraction technique based on information entropy,on information entropy and EMD decomposition,and on the EMD decomposition and the Hilbert transform,with the LSSVM classification method which is excellent in the small sample model,.The accuracy of classification recognition is obtained by Matlab simulation and comparative analysis is carried out.After analysis,the pattern classification process is optimized from two aspects to further improve the infrasound event.The classification accuracy rate and the experimental results show that the optimization effect is more obvious,and finally the optimal pattern recognition method for small sample events is selected.