基于双模态MobileViTv2的饲料剩余量非接触式估算方法

打开文本图片集

中图分类号：TP391 文献标识码：A

文章编号：0439-8114（2026）02-0202-07

DOI：10.14088/j.cnki.issn0439-8114.2026.02.030 开放科学（资源服务）标识码（OSID）：

A non-contact estimation method for feed residue based on dual-modal MobileViTv2

CAI Xiao-jin 1 ，BAI Tao 1，2，3 ，LI Xiang1，QIAO Rui-qiang1

（1.CollgfdanaliU; Intellnte China）

Abstract：Aimingattheproblemsoftraditionalfeedresiduedetectionmethodsrelyingoncontactsensors，highcost，andtheedto modify feeding troughs，a lightweight convolutional fusion regression model （dual-modal MobileViTv2 + CMFIM + SE）based on dual-modalMobileViTv2wasproposedtoachievenon-contactandhigh-precisionautomaticestimationofeedresidue.TakingRGBimagesanddepthimagesasinput，themodelextractedmulti-scalefeaturesrespectivelythroughthedual-modalMobileViTv2andintro ducedacrossmodalmulti-scalefeature interactionmodule（CMFIM）atfourlevels toachievespatial-channeldualinteractionbetwenRGBanddepthfeatures.AnSEmodulewasemployedtoadaptivelycalibratechannelweightsandenhancehigh-levelsemantic representationcapabilityThepredictionresultswereoutputthroughamultlayerperceptronregresionhead.Ontheself-builtdataset，the mean absolute error （ MAE ）and root mean square error（RMSE）of the dual-modal MobileViTv2 + CMFIM + SE model were 98.24g and 140.21g ，respectively，which represented reductions of 21.65% and 16.73% compared to the dual-modal MobileViTv2 modelwithouttheCMFIMandSEmodules，andtheparametersizeofthe modelwasonly9.9×106.Themodelcombinedtheadvantages ofhighaccuracy，trongobustnesandlightweightdesign，providingafeasibletechnicalpathwayforprecisionfeedinginintellgnt livestock farming.

Key words：dual-modal MobileViTv2；feed residue;non-contact estimation;RGB images;depth images用的关键手段。（剩余11823字）