Research On Qiantang River Tidal Bore Detection Based On Spectral Features And Nonlinear Theory

As a result of Qiantang River its unique geographical position and the influence of the celestial force,formed the Qiantang River tidal bore which is remarkable natural landscape.On the one hand the tidal bore has brought a high ornamental value,but on the other hand we can’t ignore the damage it may bring,because the QiantangRiver tidal bore leads to the loss of personnel and property were reported every year.If we can detect the tidal bore rapidly and accurately and release warning information timely,we can reduce the loss of life and property effectively.In this paper,a tidal bore detection method based on acoustic recognition was proposed.The tidal bore detection is realized by combining the characteristics of the tidal bore and the classifier.The main works are as follows:(1)A large number of acoustic signals are collected as a positive sample by using a specific sound level meter and a large number of non-tidal sound signals are collected as negative samples to enhance the robustness of the recognition system.The characteristic parameters of the sound signal are extracted,The corresponding tidal bore identification system is constructed by using the relevant nonlinear theory.(2)Researching time domain and frequency domain characteristics of the tidal sound signal,and extract the frequency domain characteristic parameters of the sound signal.Based on the LPC transform coefficients(LPCC)feature extraction method and the Mel frequency spectrum coefficient(MFCC)feature extraction method,the dynamic characteristic parameters of the two above algorithms are obtained based on the first order difference and the order difference.The feature extraction of acoustic signals and the training of different features are conducted to compare the effect of different characteristic value extraction methods on recognition accuracy.(3)In this paper we study two nonlinear theories BP neural network and support vector machine.The advantages and disadvantages of the two algorithms are introduced in detail.Through the training of the characteristic parameters of the acoustic signals,the corresponding models are obtained,and the sound signals are classified and identified.A large number of experiments were carried out to compare the recognition accuracy of the two classifiers,and compared the12-dimension static features,the 12-dimension static features and the 12-dimension dynamic features,and the 12-dimension static features and the 24-dimension dynamic features of the two classifiers.(4)In order to verify the feasibility of the algorithm and improve the practicability of thesystem,we also establish a new tidal bore detection system based on C#.The MFCC feature extraction method is implemented in the new system,and the open source SVM code is utilized.Compared with Matlab the new system is more intuitive and more practical.