基于Lite-YOLOv8的田间杂草检测方法研究

打开文本图片集

Study onAFieldWeedDetectionMethod BasedonLite-YOLOv8

DU ZhiyongWU DiLONG Yanjiang XIA HaowenWANG Ruihao ZHANG Shibo ZHANG Genmao （Henan Institute of Technology，School of Intelligent Engineering，Xinxiang ， China）

Abstract： [Purposes] Weed detection is of great significance in mechanized agricultural weeding.To address the issues of the large size of the YOLOv8 model and its diffculty in detecting occluded targets，an optimized improvement of the YOLOv8 model is proposed.[Methods] First，the backbone network was replaced with ShuffleNetV2 to reduce the computational cost and parameter size of the model，thereby enhancing its applicability on edge devices.Second，the C2f module in the detection head was replaced with an improved C2f_Faster module to improve feature fusion efficiency and detection accuracy.[FindingslExperimental validation on multiple weed datasets showed that compared with the original YOLOv8 model， the improved model reduces the number of parameters by 67.85% while maintaining the advantage of lightweight， and increases the mean average precision （ mAP ）by 1.56%.[Conclusions] The effectivenessof the proposed improvements is successfully verified，providingan eficient and reliable solution for weed detection in complex agricultural scenarios.

Keywords：object detection;machine learning;YOLOv8;lightweight model

0 引言

在我国农业生产中，杂草问题每年造成约900亿元的经济损失，危害面积超过7300万 km2 ，严重制约农作物的生长与产量。（剩余6164字）