基于三维点云和改进PointNet++的大田烟株叶片计数方法

打开文本图片集

中图分类号：S572；S126 文献标识码：A文章编号：1007-5119（2025）03-0089-09

Field Tobacco Leaf Counting Method Based on 3D Point Clouds and Improved PointNet++

NAN Dewang1，LI Junying2*，LIANG Hong1， MA Erdeng²， ZHANG Hong²， XIAO Hengshu1 （1.Schoo ofInformationScience&Enginering，Yunnan UniversityKunming 650504，China; 2.YunnanAcademyoTobaco Agricultural Sciences， Kunming 650021， China）

Abstract：The leafcountof tobacco plants isoneof theimportant phenotypic parameters fortobacco leaf yield estimationTo address thechalengesof taditionalmanaltobaccoleafcounting，afieldtobaccoleafcountingmethodintegratingthre-dimensional point cloudsand improved PointNet++was proposed.This method employs UAVobliquephotographyto acquirefeld tobacco plant images and generate three-dimensional point clouds.An improved PoinNnet ++ algorithm is then utilized to perform leaf point cloud segmentation.The proposedalgorithmreplaces the MLPwith KANtoenhance leamingcapacityand minimize training loss.A DGSTD atention mechanism was proposed， which integrated DGST network and DBB multi-branch block to enhance accuracy. Additionallyarfocaloswasicooratdtodsthesibaace intouddstrutionacrosategores.allyte MeanShiftclustering algorithm wasemployed toclustertheleaf point clouds，from which theleafcount was derived.Theresults showed that the accuracy of point cloud segmentation was 92.55% ，and the mean Intersection over Union （mIoU） was 76.33% representing improvementsof06and2.81 percentagepoitsover teorgialmodel，respectively.Teproposedmethodachevesa leaf counting precision of 94.35% ， successfully implementing leaf counting of field tobacco plants in three-dimensional space.

Keywords： field tobacco plants; leaf counting; PointNet++; 3D point clouds; UAV oblique photography

烟草是以收获叶片为目的的特殊经济作物，叶片数是衡量烟株生长状况、预测产量的重要指标[1-2]传统人工叶片计数方法效率低、周期长，劳动强度大且成本高，无法满足现代农业高通量快速获取叶片数的需求。（剩余15888字）