Research On The Detection And Magnitude Estimation Method Of Hail Acoustic Wave Signa

Hail is an extreme weather phenomenon that can cause huge losses to people’s lives and property.Therefore,hail detection has always been a focus of attention.Traditional hail detection methods mainly rely on weather radar monitoring and manual statistical observation,which not only have large errors but also lack timeliness.Therefore,this article takes hail sound signals as the research object,and uses signal processing technology and deep learning methods to study the identification of hail magnitude.The main research contents are as follows:(1)To simulate hail more accurately,this article first studies the statistical characteristics of the real hail form to understand its physical formation mechanism and the differences in the size distribution of hail particles with different forms.Based on this,this article divides the hail magnitude and calculates the empirical formula for the final velocity of real hail using kinematics.The final velocity at a fixed height is taken as a fixed parameter,and then simulated hail sound signals are collected.To achieve this,the research team set up a hail sound signal collection system,which uses a designed single-channel hail sound signal detection sensor to collect signals and establish a dataset.This article proposes an improved CEEMD hail sound signal denoising algorithm and a blind source separation algorithm for LMD adaptive singlechannel hail sound signals after optimizing the mixed signals to obtain pure hail sound signals,improve the quality and reliability of hail sound signals,and help identify hail magnitude more accurately.(2)The focus of this study is to accurately identify different hail magnitudes.Based on the acoustic signals generated by hail falling to the pickup device,an improved 1D-CNN hail magnitude recognition algorithm based on multi-domain feature fusion is proposed.Specifically,the extracted time domain features,frequency domain features and transform domain features are used as the training data of the model,and the features are further extracted by the improved 1D-CNN algorithm.Finally,the accuracy of hail magnitude recognition is99.64%.(3)Considering the versatility of the algorithm in identifying hail magnitude in different scenarios,that is,the hail magnitude recognition algorithm does not depend on the specific hail scene to identify its magnitude,this paper proposes an end-to-end deep learning method: hail magnitude recognition algorithm based on dual-channel heterogeneous deep residual shrinkage network and multi-task learning.The algorithm can identify the magnitude of hail in different scenarios.The experimental results show that the comprehensive recognition rate of hail in various scenarios is as high as 99.36%.In order to prevent the training imbalance of the two recognition tasks in this paper,an uncertain Gaussian likelihood maximization method is proposed to adaptively adjust the loss weights of the two tasks.Compared with the traditional DRSN network structure,the DCI-DRSN structure proposed in this paper has stronger ability to filter noise.When the signal-to-noise ratio of hail sound signal is-5d B,0d B,5d B and 10 d B,the comprehensive recognition rate of hail magnitude is 83.85%,87.30%,91.33% and 95.72%respectively.Compared with the traditional DRSN structure,the accuracy is improved by 7.11%,3.47%,4.13% and 4.10%,which verifies that the algorithm has better generalization and robustness.