基于YOLOv8和伪彩色处理的玉米去雄机导航线提取算法  

Navigation Line Extraction Algorithm of Maize Detasseling Machine Based on YOLOv8 and Pseudo-color Processing

作　　者：刘雨杰 郭延超 杨宇昂 吕轩 王笑乐 杨洋[1,3] LIU Yujie;GUO Yanchao;YANG Yu’ang;LV Xuan;WANG Xiaole;YANG Yang(School of Engineering,Anhui Agricultural University,Hefei 230036,China;State Key Laboratory of Intelligent Agricultural Power Equipment,Luoyang 471039,China;Anhui Provincial Key Laboratory of Intelligent and Green Agricultural Equipment,Hefei 230036,China)

机构地区：[1]安徽农业大学工学院,安徽合肥230036 [2]智能农业动力装备全国重点实验室,河南洛阳471039 [3]智能绿色农业装备安徽省联合共建学科重点实验室,安徽合肥230036

出　　处：《拖拉机与农用运输车》2024年第5期47-51,68,共6页Tractor & Farm Transporter

基　　金：全国重点实验室开放课题,基于多源异构传感器的作物行感知及垄间导航路径动态规划研究(SKLIAPE2023012);安徽省重点研发计划项目:液态畜禽粪肥多路匀施精量供肥关键技术及装备研发(2022i01020011);国家级大学生创新训练项目:作物垄间自动对行驾驶系统(202410364035)。

摘　　要：抽雄期玉米的去雄打顶是提高玉米产量的重要任务,但该时期玉米的叶片、雄穗和土壤色彩难以分割,这给玉米去雄机的行间导航任务造成了困难。考虑到导航线提取的关键是在农业机械行驶区域内进行作物特征点的检测,故本文通过YOLOv8神经网络自主提取玉米雄穗感兴趣区域(ROI),并在ROI内采用Jet映射和Otsu算法分割雄穗、绿叶和土壤,然后采用FAST角点检测方法提取雄穗特征点并根据导航区域划分特征点集,最后采用最小二乘法拟合作物行检测线。实验结果表明,该算法实现了精确的、快速的玉米抽雄期导航线提取,处理单帧图像(600 pix×700 pix)平均耗时56.7 ms,平均偏差角为1.03°,能够满足玉米去雄机田间导航行驶的需求。Detasseling of maize at tasseling stage is an important task to increase maize yield,but the color of maize leaves,tassel and soil is difficult to separate,which makes it difficult for the Maize Detasseling Machine to navigate between crop rows.Considering that the key to navigation line extraction is the detection of crop feature points in the driving area of agricultural machinery,this paper independently extracts the ROI of maize male ear through YOLOv8 neural network,and uses Jet mapping and Otsu algorithm to segment maize tassel,green leaf and soil within the ROI.Then,the FAST corner detection method was used to extract the male tassel feature points and divide the feature points set according to the navigation region.Finally,the least square method was used to simulate the detection line of the cooperative object.The experimental results show that the proposed algorithm can achieve accurate and fast navigation line extraction during maize tasseling stage.The average time of processing a single frame image(600 pix×700 pix)is 56.7 ms,and the average deviation Angle is 1.03°,which can meet the requirements of Maize Detasseling Machine navigation in the field.

关 键 词：作物行检测 深度神经网络 伪彩色处理 ROI 农机导航 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]