基于深度学习的设施蔬菜精准施药装置设计与试验

打开文本图片集

中图分类号：S491 文献标识码：A 文章编号：2095-5553（2026）04-0102-06

Abstract：China isamajor countryin termsof productionandconsumptionoffacilityvegetables.However，thecurrent pesticideaplication practices mainly involve manual continuous spraying，which faces problemssuch as poor equipment adaptability and high rates of pesticide waste.To addressthese issues，a precision pesticide application device based on deeplearning was designed for facilityvegetableenvironments.Firstly，field parametersurveyswereconducted to establishamodelforthepesticideaplication frame.Throughsimulationanalysis，thereliabilityand stabilityof the pesticideaplicationdevice structure wereverified.Furthermore，tomeetthereal-timerequirementsof precisionpesticide application and the complex background of field environments，improvements were made to the YOLOv5 target detection algorithm.TheoriginalC3module wasreplacedwith theFasternetmodule，reducing theamountoffloating-point operations and the numberof memory accss.The Conv module in the head network was replaced with GSConv，further reducingcomputational resourceconsumption.TheUpsample moduleinthe head network wasreplaced with the ultra-lightweightup-samplingmodule DySample，ensuring theeffective informationtransfer.Afterablationand comparativeexperiments，theeficiencyof thedetectionmodelwasvalidated.Aditionally，inthedesignof the spraying system，theoverallsystem design was completed，and thepixel coordinatesobtainedfromthetarget detection algorithm wereconverted into three-dimensionalcoordinates.Finall，orthogonal experiments were designed tooptimize the pesticide application parameters，the coefficient of variation in droplet deposition was minimized to 12.5% when the spraying height was set at 419mm ，the duty cycle of the solenoid valve was at 91% ，and the traveling speed was at

0.32m/s .Moreover，the pesticide saving rate under precision application mode was evaluated，the results showed that the larger the crop row spacing，the higher the pesticide saving rate with precision application.

Keywords：greenhouse vegetable；spray device;deep learning；spray system；precision pesticide application

0 引言

在全球范围内，我国设施蔬菜面积位居首位，达到2 567khm2 ，设施蔬菜产业总产值突破1万亿元，通过占总耕地 3% 的精耕细作，创造了超过 15% 的种植业总产值[1]。（剩余9409字）