融合超绿通道的低辨识度草坪杂草识别算法  

A low-resolution lawn weed recognition algorithm integrated ultragreen channels

作　　者：化春键 宋一鸣 蒋毅 俞建峰 陈莹[3] HUA Chunjian;SONG Yiming;JIANG Yi;YU Jianfeng;CHEN Ying(School of Mechanical Engineering,Jiangnan University,Wuxi Jiangsu 214122,China;Jiangsu Key Laboratory of Advanced Food Manufacturing Equipment&Technology,Wuxi Jiangsu 214122,China;School of Internet of Things Engineering,Jiangnan University,Wuxi Jiangsu 214122,China)

机构地区：[1]江南大学机械工程学院,江苏无锡214122 [2]江苏省食品先进制造装备技术重点实验室,江苏无锡214122 [3]江南大学物联网工程学院,江苏无锡214122

出　　处：《东北农业大学学报》2024年第7期83-92,共10页Journal of Northeast Agricultural University

基　　金：国家自然科学基金项目(62173160)。

摘　　要：为提高草坪维护自动化程度,实现除草机器人自动识别精准施药,针对自然环境中草坪杂草难以识别的问题,以YOLOv7为基础目标检测网络,提出一种增强图片颜色信息的草坪杂草检测模型。对RGB图片进行超绿处理,抑制背景干扰信息,突出杂草颜色信息。扩充网络输入通道为四通道,将超绿灰度图与RGB图片融合得到新的四通道图片,提高网络对草坪环境的抗干扰能力。引入三维注意力模块SimAM,提高网络针对低辨识度杂草识别精度,增强网络对草坪杂草目标的识别效果。改进后YOLOv7-4ch网络对草坪杂草识别效果较好,自然环境下识别草坪杂草的mAP@0.5为91.7%,帧速率100 f·s^(-1),模型大小为67.9 Mb,可满足除草机器人识别需求。改进网络与基础YOLOv7网络相比,mAP@0.5提高2.3%,与YOLO系列其他算法相比有较大提升。该算法可有效识别自然环境中的草坪杂草,为其精准化施药提供技术支持。To enhance the automation of lawn maintenance and achieve precise herbicide application by weeding robots,a lawn weed detection model was proposes based on the YOLOv7 object detection network to address the challenge of weed identification in natural environments.The model enhanced the color information of images by performing super-green processing on RGB images,suppressing background interference,and highlighting the color information of weeds.The network input channels were expanded to four channels by fusing the super-green grayscale image with the RGB image,improving the network's resistance to environmental interference.A three-dimensional attention module,SimAM,was introduced to improve the network's recognition accuracy of low-distinguishability weeds,enhancing the detection performance for lawn weeds.The improved YOLOv7-4ch network demonstrates excellent detection performance for lawn weeds,achieving an mAP@0.5 of 91.7%and a frame rate of 100 fps in natural environments,with a model size of 67.9 MB,meeting the recognition requirements of weeding robots.Compared to the base YOLOv7 network,the proposed network improved mAP@0.5 by 2.3% and also showed significant improvements over other YOLO series algorithms.This algorithm could effectively identify lawn weeds in natural environments and supported precise herbicide application for lawn weed control.

关 键 词：目标检测 YOLOv7 草坪杂草 精准施药 通道融合 

分 类 号：S126[农业科学—农业基础科学]