Research On Extreme Learning Machine Algorithm For Bird Sound Recognition

Nature makes sounds all the time,and sound is the basic attribute of the natural environment.Natural environment sound recognition mainly realizes the perception of the environment by analyzing and identifying the sound data collected in the environment.The study of ambient sound is of great significance in the fields of public place safety monitoring,environmental and public facility management,sound source localization and classification,and insomnia treatment research.There are thousands of sounds in the natural environment.Birds are an indispensable existence in the ecosystem.Using sound monitoring technology to analyze and identify bird sounds is of great significance for protecting and evaluating the quality of the ecological environment.However,there are problems such as large environmental noise interference and unbalanced song data when collecting bird sound data in the natural environment.In order to explore and improve the method of bird sound detection,this paper uses a recognition model based on extreme learning machine algorithm to study the characteristic data of bird sounds.The main research contents are as follows:1.This paper investigates the recognition of bird sounds in environmental sounds.A total of 50 song sample datasets of common and rare birds around Shanghai were recorded from Shanghai gulf national forest park,Shanghai zoo and Shanghai wild animal park,and three bird audio datasets were made.The sound data of various birds include 66 to 563 16 bit mono channel wav formats,with the duration of one song cycle for each bird The sampling frequency is 44100 Hz.2.Bird sound data is not simply a single bird song signal,but also contains wind,rain,insects,car sounds and other noises,so it is necessary to pre-process the data set when classifying bird song data.Based on the characteristics of bird song,the bird song signal is intercepted in signal segments,noise reduction is performed,and then the parameter features that effectively represent the bird song signal are extracted.3.In bird sound recognition,a recognition model based on the extreme learning machine algorithm is used to classify the feature parameters,and three algorithms are used to build the recognition model,namely,the standard extreme learning machine bird sound recognition model,the kernel extreme learning machine bird sound recognition model with improved kernel function,and the deep learning machine composed of a cascade of three extreme learning machine autoencoders,to recognize bird sounds and compare their respective recognition results.4.To further improve the recognition accuracy,it is proposed to use particle swarm algorithm,bat algorithm and lion swarm algorithm to optimize the bird sound recognition model based on the extreme learning machine,find the optimal parameters of the model,and compare the classification performance of the three optimization models.It is also compared with the support vector machine,BP neural network and random forest classification models vertically to analyze the recognition effects of different classification models.5.Through the above steps,experiments on 50 kinds of bird sound signals from three datasets show that the recognition accuracy of the standard limit learning machine,the kernel limit learning machine and the depth limit learning machine can reach 71.10%,75.25% and79.58%,respectively.The recognition accuracy of the classification model optimized by particle swarm algorithm,bat algorithm and lion swarm algorithm are improved by 2%,3% and4%,respectively,has the best recognition effect by vertical comparison.It can be seen that the effectiveness and feasibility of bird sound recognition model based on extreme learning machine algorithm in this paper for bird identification.