Research On Bird Song Recognition Based On Improved Genetic Algorithm To Optimize BP Neural Network

In a world with scarce resources and a growing population,the management and protection of wildlife is becoming an increasingly important issue,and many wild populations are under pressure on human activities.Birds have always played the role of sentinels and are one of the indicators of the health of the entire biodiversity.However,China has a vast territory,with more than 1,400 bird species.There are huge difficulties in monitoring species diversity and richness.Traditional survey methods are usually costly,have space-time limitations,and have observer bias,and it is difficult to digitize permanent storage,which cannot meet the current management and protection needs.Based on the improved genetic algorithm BP artificial neural network,this study designed a species recognition model suitable for bird song recognition to achieve accurate and rapid bird recognition.The specific research contents and results are as follows:(1)By collecting sample data onto-the world's largest bird song database-xeno-canto(https://www.xeno-canto.org/),bird net(http://niaolei.org.cn/sounds)and bird sound net(http://www.1380898.com/).A series of pre-processing operations such as normalization,Pre-emphasis,framing and window,and Voice Activity Detection(VAD)were performed on the bird song fragments in the field with a complex environmental background.The five feature quantities of MFCC,LPCC,Short-term energy,Zero-crossing rate,and Formant were determined through experiments.The average recognition rate of each feature parameter reached75.242%,63.264%,53.394%,47.76%and 38.338%respectively.And these five feature quantities are used as the combined feature parameters of the bird song signal.(2)Establish a bird species recognition model based on improved genetic algorithm to optimize the BP neural network.In this paper,genetic algorithm and BP neural network algorithm are selected among many methods for research.Genetic algorithms have great advantages in adjusting operator parameters and seeking optimal solutions,and have broad application prospects in the field of neural network optimization.Through in-depth analysis and research of the existing algorithms,combined with the characteristics of the data,and improved genetic algorithm with adaptive cross probability and mutation probability is proposed to optimize the weights and thresholds of the BP neural network.In the algorithm,the improved fusion GA?BP network structure is determined to be 43-16-6,with a total of 784weights(including the weight of the input layer and hidden layer neurons is 688,and the number of weights of neurons in the hidden layer and the output layer is 96);the neuron has 22 thresholds(the hidden layer neuron threshold is 16,and the output layer neuron threshold is 6);the learning rate is 0.1.In the genetic algorithm,the individual chromosome cod length is 806,the population size is 50,and the number of iterations is 250,P_c=0.75,P_m=0.05.(3)Select windows 7.0 as the operating system for identification and detection,and implement data preprocessing and GA?BP fusion algorithm programming based on MATLAB(R2017b),to realize the identification of bird song fragments.Six species of birds,Including White wagtail,Brownish-flanked bush-warbler,Great tit,Oriental magpie,Chinese blackbird,Streak-breasted scimitar babbler,were selected as the experimental objects.The experimental results showed that the recognition rates of BP,SGA?BP and AGA?BP were 77.65%,86.17%,and 88.25%respectively,while the recognition rates of IAGA?BP were as high as 93.16%.The fitness was IAGA>AGA>SGA.The iteration times of the three networks are IAGA?BP(1310times)<AGA?BP(1632 times)<SGA?BP(1727 times).In this paper,the proposed adaptive strategy based on the individual can significantly improve the search speed,solution accuracy,and robustness,and is of the great significance of in-depth research on bird acoustic monitoring,species identification,community ecological protection,etc.,Can help accurately and efficiently analyze large audio data sets to monitor wild bird communities.