Research On Hail Detection Method Based On Acoustic Signal Feature Analysis

The occurrence of hail has a huge persecution on industrial and agricultural production,military and aerospace.At present,the effective prediction of hail is mainly realized by radar,but the actual hail situation and radar prediction statistics are not strong.In addition,the accuracy of hail prediction and the estimation of hail magnitude mainly rely on the traditional manual monitoring,which is not only time-consuming and labor-consuming,but also lack of intelligence and timeliness.Therefore,in view of the shortcomings of the current hail monitoring technology,this paper proposes to classify and identify hail from the perspective of "acoustics".This project focuses on the hail signal noise reduction,blind source separation,feature extraction and classification recognition technology.(1)The collected acoustic signal is preprocessed by noise reduction and blind source separation to extract the pure hail signal.By using wavelet soft threshold denoising method,the denoising effects of different wavelet basis functions and different decomposition layers are compared and analyzed,and the most appropriate wavelet basis function and wavelet decomposition layers are selected for hail signal denoising.Fast ICA algorithm is used to blind source separation of acoustic signal,and independent hail acoustic signal data is obtained.Then,the independent acoustic signal is preprocessed by endpoint detection.(2)How to correctly identify the hail acoustic signal is the key problem of this paper.After preprocessing the acoustic signal,the time domain,frequency domain and wavelet packet energy spectrum features of the independent hail acoustic signal are extracted.Then,the fuzzy clustering algorithm and the generalized regression neural network are used to recognize the hail signal.Using Mahalanobis distance instead of Euclidean distance,and introducing the method of K-means + + to select the initial center point,this paper innovatively combines the fuzzy c-means algorithm based on Mahalanobis distance(M-FCM)with K-means++ to improve the M-FCM algorithm,and the recognition rate reaches 94.26%.In addition,aiming at the problem of single time-frequency feature,combined with the characteristics of acoustic signal in time-frequency domain,the feature extraction method of combining time-domain,frequency-domain and wavelet domain is adopted,and the entropy method and generalized regression neural network(GRNN)are combined.This paper proposes a hail recognition method based on entropy method and GRNN,which improves the accuracy of hail recognition,up to 97.82%.(3)On the basis of hail signal recognition,the Mel frequency cepstrum coefficient(MFCC)and its difference parameters of hail signal of different magnitude are extracted,and the feature vector is input into GRNN training to realize the estimation of hail magnitude.The classification accuracy reaches 84%.