基于Yolov4模型的玉米幼苗与杂草识别检测  

Identification and Detection of Maize Seedlings and Weeds Based on Yolov4 Network

作　　者：李雪峰 施晨辉 宋名果 甘仲辉 乔芷柔 孟庆宽[1] LI Xuefeng;SHI Chenhui;SONG Mingguo;GAN Zhonghui;QIAO Zhirou;MENG Qingkuan(College of Automation and Electrical Engineering,Tianjin Vocational and Technical Normal University,Tianjin 300222;Tianjin Boyi Pneumatics Co.,LTD.,Tianjin 300457)

机构地区：[1]天津职业技术师范大学自动化与电气工程学院,天津300222 [2]天津博益气动股份有限公司,天津300457

出　　处：《热带农业工程》2023年第1期1-6,共6页Tropical Agricultural Engineering

基　　金：国家级大学生创新创业训练计划项目(No.202110066016)。

摘　　要：作物生长过程中杂草与作物争夺水分、养分和光照,阻碍作物正常生长。除草是农业生产中的一个重要环节,对提高作物产量和质量起着决定性作用。本文以玉米幼苗及其常见的6种伴生杂草为研究对象,将Yolov4模型引入到作物与杂草识别检测中。检测模型先采用CSPDarknet53前置基础网络进行图像特征提取,然后在颈部网络中利用空间金字塔模块和FPN+PAN结构实现多尺度层级特征融合,最后通过检测头网络输出预测目标类别和位置信息。结果表明,本文模型对作物与杂草平均识别精度达到94.59%,检测一幅图像的平均时间为30.42 ms,相比于Faster-RCNN和SSD模型,具有识别速度快与检测精度高等优点,可以为自动化除草所涉及的苗草识别问题提供有效技术参考。Weeds compete with crops for water,nutrients and light,which prevent the normal growth of crops during the crop growth period.Weeding is an important part of agricultural production and plays a decisive role in improving crop yield and quality.In this study,maize seedlings and six common associated weeds are taken as the research object,Yolov4 model is introduced into crop and weed identification and detection.The detection model uses CSPDarknet53 as a pre-basic network to extract image features,and then the spatial pyramid module and FPN+PAN structure are used in the neck network to realize multi-scale hierarchical feature fusion.Finally,the predicted target category and location information are output by the detection head network.Experimental results show that the average recognition rate of crops and weeds for the proposed model reaches 94.59%,and the average detection time of an image is 30.42 ms.Compared with Faster-RCNN and SSD models,Yolov4 has the advantages of fast recognition speed and high accuracy,and can provide an effective technical reference for identifying problems of seedlings and grasses involved in automatic weeding.

关 键 词：深度学习 自动除草 苗草检测 Yolov4模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]