基于EfficientDet网络的超低空无人机大豆幼苗识别  

Recognition of Soybean Seedlings Using UAV-basedImagery and EfficientDet

作　　者：康烨 李金阳 邱金凯 张伟[1] 许秀英 Kang Ye;Li Jinyang;Qiu Jinkai;Zhang Wei;Xu Xiuying(College of Engineering,Heilongjiang Bayi Agricultural University,Daqing 163319,China)

机构地区：[1]黑龙江八一农垦大学工程学院,黑龙江大庆163319

出　　处：《农机化研究》2023年第9期7-13,共7页Journal of Agricultural Mechanization Research

基　　金：国家现代农业产业技术体系项目(CARS-04-PS32);黑龙江八一农垦大学校内培育课题(XZR2017-10);黑龙江省大学生创新创业训练计划项目(202010223007)。

摘　　要：为提高田间大规模高通量情况下大豆幼苗真叶期(V1)、复叶期(V2)形态的识别率,从而为判断出苗率、苗期整齐度及茎叶喷雾除草最佳时期确定提供依据,提出了一种基于超低空无人机可见光图像的大豆幼苗识别方法。该方法以四旋翼无人机采集到的RGB图像为基础,将其转换至HSV颜色空间从而提取出图像中的大豆幼苗植株,并利用K-means方法找到最佳分割阈值,其真叶期阈值点坐标为(83,183,201)、(31,25,74),复叶期阈值点坐标为(79,215,225)、(29,72,61)。基于提取出的大豆幼苗图像训练深度学习网络,得到大豆幼苗识别模型。测试结果表明:模型可有效识别真叶期与复叶期两个幼苗关键时期的大豆幼苗,真叶期幼苗识别准确率达到92.68%,复叶期幼苗识别准确率达到90.09%,可有效完成田间大豆幼苗苗期判断任务,及时指导大豆种植管理决策。In order to improve the identification rate of soybean seedling morphology at true leaf stage(V1)and compound leaf stage(V2)under large-scale and high-throughput conditions,and to lay a foundation for judging seedling emergence rate and uniformity of seedling stage,a soybean seedling identification method based on ultra-low altitude UAV-based imagery was proposed.The method was based on the RGB image of the quadrotor UAV acquired,which was converted into HSV color space to extract the soybean seedling in the image,and used K-means find the best threshold,V1 point coordinates of(83,183,201),(31,25,74),and V2 point coordinates of(79,215,225),(29,72,61).The soybean seedling recognition model was obtained based on the extracted soybean seedling image training deep learning network.Tests results showed that the model could effectively identify soybean seedlings in the two key stages of V1 stage and V2 stage,and the identification accuracy reached 92.68%and 90.09%.The results showed that the soybean seedling stage judgment task based on ultra-low altitude UAV image could be effectively completed in the field and timely guide the soybean planting management decision.

关 键 词：无人机图像 HSV颜色空间 深度学习 大豆 

分 类 号：S252[农业科学—农业机械化工程]