基于无人机RGB图像和GU一Net模型的玉米长势监测研究

打开文本图片集

中图分类号：S126 文献标识码：A 文章编号：2095-5553（2025）12-0063-07

Abstract：Thegrowthofmaizeinthefieldiscloselyrelated totheyield potential，which isanimportanttraitforvariety improvement.However，thetraditional methodsof growth monitoringrelyon manualobservationandare susceptible to subjective factors.Inorder tosolvethis problem，fieldexperiments werecarredout infive suitableecological sites in the main maizeproducingareasofJilinProvince.Theoriginal imagedata werecolected byusing theunmannedaerialvehicle （UAV）flight system，and asegmentation model basedon GAN network，GU—NET，was proposed to extract maize canopy coverage from the field images.The results show thatthe acuracy rate of the proposed GU—Netmodel onthe test set is 93.94% ，the recall rate is 90.86% ，the F1 score is 92.73% ，the average joint intersection is 87.36% ，and the Matthews correlation coefficient is 91.74% ，which is better than the traditional models such as U—Net and FCN.A validation sample setisconstructedfromcorncultivationexperimentsunderdiferentenvironmentsandnitrogenaplicationgradients，andthe canopy response curve is drawn，and the area covered by the curve is calculated to quantify the corn growth.

keywords：maize growth monitoring；UAV；coverage；deep learning；image segmentatior

0 引言

玉米是重要的粮食作物和饲料作物，生产潜力大、经济效益高，在保障粮食安全方面具有重要战略地位[1]。（剩余11813字）