基于轻量化YOLOv8n的玉米苗与杂草识别

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：1006-060X（2025）11-0101-06

Abstract：AlightweightmodelbasedonYOLOv8nisdevelopedtoachievetheaccuratedistinguishingbetweenmaizeseedlings and weed inthefeld.Onthe basis ofYOLOv8n，theCARAFEmodule is introduced toreplace the originalup-sampling module to enhanceteabilityofreconstructingthedetailandpositioninformationofthedetectedtargets，therebyimprovingtherecognition accuracy ofsmalltargetsand edgetargets.The GhostConv module is incorporated intothe C2f moduleof YOLOv8ntoreduce the computationand parametersof themodel.Finall，this paper replaces thestandardstaticconvolution inYOLOv8n withdyamic convolutiontocompensatefortheacuracydegradationduetolightweightingwhilerealizing modellightweighting.Theresultsshow that the algorithm developed in this paper demonstrates the mAP @0.5 of 83.4% ，the computation of 6.0 GFLOPs，and the parameters of 2.3M ，which represent a1.3 percentage points increase，a 25.9% decrease， and a 23.3% decrease，respectively， compared with thoseof YOLOv8n.The improvedlightweight YOLOv8n model has lower resource consumption and stronger potential for mobile deployment while maintaining high accuracy.

ey words： YOLOv8n; lightweighting; recognition of maize seedlings;recognition of weed

在玉米的整个生长周期中，田间杂草管理是保障高产稳产的重要环节。（剩余7648字）