基于改进Swin-Transformer的果树病叶分类模型

打开文本图片集

中图分类号：TP391.4 文献标志码：A 文章编号：2096-9902（2025）14-0033-04

Abstract：Inrecentyears，climatechangeandchangesinagriculturalactivitieshaveincreasedthefrequencyandseverityof plantdiseases，havingamajorimpactonfoodproductionandqualitysafety.Therefore，toensurefoodsecurity，timelyand accuratedetectionanddiagnosisofplantdiseasesarecrucial.Thispaperdesignsatreediseaseleafclassficationmodelbased ontheimprovedSwin-Transformer，whichoptimizesfeaturesbyintegratingdual-pathatentionmechanisms.Atthefeature processinglevel，amulti-levelprocesingstructureicludinglayerandardzatiodaptivepolingndfullonecdasif isdesiged.ThiscompositearchitecturemaintainsteadvantagesofTransfomersglobalmodelingandsignificantlyimprovesthe eficiencyofcapturingfine-grainedpathologicalfeaturesthroughanatention-guidedfeatureenhancementmechanism.The proposed model achieves greater accuracy than previous convolution and visual transformer-based models.

Keywords：deeplearning;atentionmechanism;convolutional neural network；plantdiseaserecognition;smartagriculture

植物病害是导致粮食产量与质量降低的重要因素。（剩余5333字）