基于SAM实例分割的茶树叶片病害严重度估计方法

打开文本图片集

中图分类号：S571.1；TP391.4 文献标识码：A 文章编号：2095-5553（2026）04-0139-08

Abstract：Tea plants arevulnerable to diseases infection during their growth process.Byestimating the severityof tea plantdiseases，theteayieldcanbeefectivelyestimated，thus improving themanagement eficiencyof tea gardens.Inthis study，an estimation method for the severity of tea treeleaf disease，SAM—Swin，based on SAMcase segmentation was proposed.In thismethod，the afected tealeaves were accuratelycoded by SAM（Segment Anything Model）algorithm， thedisease was provided basedonthecoded information，and the single-stage calculation method ofdisease spot size was realizedbasedontheinstancesegmentationalgorithmofTransformer.Thismethodachievedthehighestclassification accuracy （Top1-Acc） of 97.80% in the dataset of diseased tea leaves. The range of determining coefficient （ ⋅R2 ）and root mean square eror（RMSE） of this method in predicting the leaf lesion size of four diseases is 0.819-0.992 and （204 0.140%-1.840% . The SAM—Swin method proposed in this study is superior to the traditional two-stage severity estimationalgorithmandthe traditional single-stageinstancesegmentationalgorithminaccuracyandspeed，and can quicklyandacuratelyestimate theseverityof tea leaf diseases，thus providing moreeficientandadvancedcultivation detection technology for tea garden management.

Keywords：tea plant diseases; severity estimation; instance segmentation; deep learning;one stage method

0 引言

化农业需求，计算机视觉技术被越来越广泛地应用于农业生产中，特别在植物病害检测与识别方面取得了显著进展[1]。（剩余14595字）