基于改进YOLOv8的轻量化杂草识别算法研究  

Research on lightweight weed recognition algorithm based on improved YOLOv8

作　　者：张超 刘宾[1] 李坤[1] Zhang Chao;Liu Bin;Li Kun(College of Information and Communication Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学信息与通信工程学院,山西太原030051

出　　处：《电子技术应用》2025年第1期80-85,共6页Application of Electronic Technique

基　　金：山西省基础研究计划项目(202303021222095)。

摘　　要：针对目前田间杂草识别模型精度低,以及参数多难以满足在计算资源有限的移动设备和嵌入式设备中部署的问题,提出一种基于YOLOv8的轻量化田间杂草识别模型。该模型使用改进后的PP-LCNet替代原有主干网络,保证精度的前提下减少模型的计算量;其次引入Effcient-RepGFPN来作为颈部网络,并将上采样前的两个CSPStage模块使用RFAConv来替代,利用不同尺度的特征来提高目标检测的性能;最后,更换MPDIoU损失函数,增强了模型的收敛性和稳定性。实验结果表明,改进模型与原模型相比准确率提升了2.1%,召回率提升了2.8%,mAP值提升了0.2%,同时模型的大小与计算量分别减少为原始模型的68.2%和62.6%,体现了改进算法的有效性。Aiming at the problems of low accuracy of current field weed identification models and the difficulty of deploying multiple parameters in mobile devices and embedded devices with limited computing resources,a lightweight field weed identifi‐cation model based on YOLOv8 is proposed in this paper.The model uses improved PP-LCNet to replace the original backbone network,and reduces the calculation amount of the model on the premise of ensuring the accuracy.Then,Effcient-RepGFPN is in‐troduced as the neck network,and RFAConv is used to replace the two CSPStage modules before up-sampling.Different scale features are used to improve the performance of target detection.Finally,the MPDIoU loss function is replaced to enhance the convergence and stability of the model.Experimental results show that compared with the original model,the accuracy rate of the improved model increases by 2.1%,the recall rate increases by 2.8%,and the mAP value increases by 0.2%.Meanwhile,the size and computation amount of the model are reduced to 68.2%and 62.6%of the original model,respectively,reflecting the effective‐ness of the improved algorithm in this paper.

关 键 词：杂草识别 PP-LCNet Effcient-RepGFPN RFAConv MPDIoU 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]