基于双向长短记忆网络和注意力机制的鸡白痢病音频检测

打开文本图片集

中图分类号：S831.4；TP391.4 文献标识码：A 文章编号：2095-5553（2025）08-0066-10

Abstract：Chicken dysentery，causedby Salmonella pulorum，isacontagious disease that poses asignificant threattothe poultryindustry.Given the importance of early monitoring in preventing the spreadof the disease，this study proposes a deeplearning-basedaudiorecognitionmodelnamedFlockVigil—Net，aimedatachievingeficientdetectionofchicken dysentery.This paper first employs the Smooth—HF denoising method，which combines a smoothing mechanism spectral subtractionandahigh-passfiltertoextractclearchickencrowing fromcomplexsounddatathat includeschicken vocalizations，behavioral noises，and environmental noises.Subsequently，an endpoint detectionalgorithm based on time-limitedconditionssegments thechickenvocal segments，and calculates three featuresofthe chickencrowing：the spectrogram，FBANK，andMel-frequencycepstral coeficients（MFCC），further forming a feature merge graph， providingrichacousticinformationfordiseaseidentification.TheFlockVigil—Netmodel integratesconvolutionalneural networks，bidirectional long short-term memory networks （BILSTM），andmulti-head attention mechanisms，achieving high-accuracyidentificationofchickensinfectedwithchickendysentery.Experimentalresultsshowthatfromthesecondto the eighth day after confirmation of chicken dysentery，the model's recognition rate increased from 86.53% to 90.26% ： Compared with other speech recognition models，FlockVigil—Net demonstrates superiorperformance.Thisstudy provides an eficient and accurate audio detection method for the early diagnosis of chicken dysentery.

Keywords：chicken disease detection；bidirectional longshort-term memory network；atentionmechanism；spectral subtraction

0 引言

1材料和预处理

伴随着养鸡业向规模化、集约化饲养方向发展，养殖密度不断增加，传染病问题日益严重，给农民造成巨大损失。（剩余14047字）