基于注意力机制融合的果实蝇多模态检测优化研究

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）19-0049-08

Abstract： Aimingatthe problem offruit flytarget detection，a multi-modal detectionoptimization method combining self-atentiondostetonmeassispoposdirstlyult-aleeraccalattinoduleisscted， whichcanworktogetherondiffrentresolutionfeature maps，sothat themodelcancapture thesubtlefeaturesoffruitfies more accuratelyandefectivelyperceivetheirelationshipwiththesurroundingenvironment，tussignifcantlyimprovingthedetection accuracyoffruitfliesincomplexorchardenvironmentSecondlyastructuredatentiongudance mechanismisproposed，which combines theLong Short-Term Memory network to model thevisual featuresof video frames in time series，andcanaccurately track theactivitytrajectoryoffruit fliesinmult-modal video.Anonlineattentionupdate mechanismis introduced to further enhance theadaptabilityofthe model.Theexperimentalresultsshow that theoptimized modelperforms wellinmulti-modal data processing，ndthedetectionaccracyunderdiferentlightconditionshasbengreatlyimproved，providingaeffetnd reliable new method for agricultural pest monitoring.

Keywords： attention mechanism fusion; multi-modal detection; fruit fly;optimization strategy; computer vision

0 引言

在智慧农业飞速发展的背景下，果实蝇的检测对果蔬行业的虫害防治具有重要意义。（剩余12754字）