基于迁移学习的MobileNetV2猪个体识别方法研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）24-0051-06

Abstract：Inorder to solve the problemoflarge workloadandlow eficiencyof manual identification in the intelligent bredingscenario，thisstudyusesDeepLeaingalgorithmtocayoutaimalindivdualidentificationresearch.Inteproce ofmodel training，duetothesmallsaleand insuffcient diversityofthedataset，the model tends tobeinsuficientlytrained， which affects its generalizationperformance.Therefore，the TransferLearning methodis introduced toavoid therepeated extractionof primary features byreusing thefeatureextractionabilityofthepre-trained model，soastoimprovethe training eficiencyofthemodel.Intheexperiment，theAnimals-1datasetisusedasthesource domainforpre-training，andthe targetdomain usestheimagesetof30pigs intheJingdongpublic pig breedingdataset.Theexperimentalresultsshowthat therecognition accuracy of TL-MobileNetV2 model with TransferLearning is 99.4% ，which is O.5 percentage points higher thanthatoffoundationmodelwithoutTransferLeaing，andtheeffectivenessofTransferLearning insmal-samplelivestock recognition task is verified.

Keywords： MobileNetV2; Transfer Learming; Deep Learning

0 引言

我国工厂化养猪在发展历程中逐步完成了从人工、半自动、自动化到智能化的演进。（剩余5382字）