基于改进 U²-Net 的小样本病虫害分割算法研究

打开文本图片集

中图分类号：S64.1 文献标志码：A 文章编号：2096-9902（2025）21-0035-05

Abstract：Asaneficienteconomiccrop，theyieldandqualityofcherrytomatoesareoftenseverelythreatenedbypestsand diseases.Accuratesegmentationofpestanddiseaseareastolocatetheextentofthediseaseisthecoreprerequisiteforearly interventionandprecisepreventionandcontrolofpestsanddiseases.Tosolvetheproblemsofmissddetectionofsmalldisease areas，bluredboundariesandlowsegmentationfiencyintraditionalcherryblossmdiseaseandpestsegmentationthispaper proposesanAtt-Netsegmentationmodelthatintegratesatentionmechanismtoachieveaccuratesegmentationandrecognitionof diseasesandpests.Using the Tomotodisease datasetfortesting，alightweight SimAM wasembeddedinthedecodinglayerof the U2 -NTP model，andtheMSAA mechanism wasusedtooptimizethefeature fusionefect.Finally，aCBMAatentionmodulewas added to the output side of each layer to construct a segmentation model AtU -2 -Net forplant diseases.Finally，it will be comparedquantiativelyandqualitativelywithothermainstreamsegmentationmodelssuchasDeepLabv3，mU-net，etc.Ithas beenverified that AttU 2⋅ -Netcan accurately segment the diseased areas of leaves and fruits，meeting the requirementsfor healthy crop cultivation in the process of planting cherry tomatoes in agriculture.

Keywords： Saint Mary's fruit diseaseand pest infestation; Semantic segmentation; Small sample size; U2. -Net;attention mechanism

在温室或大田作物种植场景中，病虫害的肆虐依然构成了一项极为严峻的挑战[。（剩余5476字）