摘要
油菜是中国最重要的油料作物之一,及时、准确、快速地对油菜长势进行监测,对其生长管理以及产量预估等具有重要意义。基于无人机影像并结合深度学习算法,实现了油菜作物长势的快速监测。首先,利用无人机影像建立了幼苗期油菜的可见光图像数据集,并由农学专家将作物长势标注为优秀、一般、较差3种情况。然后选择5种神经网络模型:EfficientNet、ShuffleNetv2、ResNet、DenseNet、ResNeXt分别进行优化和实验。实验结果显示,DenseNet的识别精度最高,达到89.12%,但是从综合精度和推理时间来看,ResNet的表现更佳。该结果表明,利用无人机和深度学习技术可以帮助农业管理人员快速、有效地完成大规模油菜作物的长势监测任务。
Rapeseed is one of the most important oil-bearing crops in China.Timely,accurate and rapid detection of rapeseed growth is of great significance for its growth management and yield estimation.Based on unmanned aerial vehicle(UAV)images and deep learning algorithms,we realize rapid monitoring of rapeseed crop growth.Firstly,the visible light image data set of rapeseed at seedling stage is established by UAV images.And the crop conditions are labeled as excellent,normal,and poor by agronomists.Then 5 neural networks(EfficientNet,ShuffleNetv2,ResNet,DenseNet,and ResNeXt)are selected for optimization and experiment.The results show that the classification accuracy of DenseNet is the highest,reaching 89.12%.However,considering comprehensive accuracy and inferring time,the ResNet performs better.The use of UAV and deep learning technology can help agricultural managers complete the task of monitoring rapeseed quickly and effectively.
作者
张瑞杰
李俐俐
李礼
姚剑
乔江伟
ZHANG Ruijie;LI Lili;LI Li;YAO Jian;QIAO Jiangwei(School of Remote Sensing and Information Engineering,Wuhan University,Wuhan 430079,China;School of Artificial Intelligence,The Open University of Guangdong,Guangzhou 510091,China;Oil Crops Research Institute of the Chinese Academy of Agricultural Sciences,Wuhan 430062,China)
出处
《测绘地理信息》
CSCD
2021年第S01期227-231,共5页
Journal of Geomatics
关键词
深度学习
无人机
油菜
长势监测
deep learning
unmanned aerial vehicle(UAV)
rapeseed
growth monitoring
