摘要
水稻病害是影响水稻产量的重要因素之一,使用传统机器学习方法识别农作物病虫害效果并不理想,因此该研究使用深度学习技术结合迁移学习方法识别常见水稻病害。使用当前深度学习领域经典网络模型VGG、ResNet、DenseNet、InceptionResNet、Xception模型作为预训练模型,通过比较不同模型在新任务上的表现,选取性能最好且最稳定的Xception模型作为最终模型。试验结果显示,DenseNet、InceptionResNet、Xception的识别准确率可以达到97%,尤其是Xception模型不仅可以达到98.50%的最高识别准确率而且是最稳定的。该研究通过试验探讨了适用于常见水稻病害智能识别的最佳模型,同时表明了使用迁移学习方法解决新任务的有效性。
Rice disease is one of the most important factors affecting rice yield. The effect of traditional machine learning method to identify crop diseases and insect pests is not idea, so that we used deep learning technology combined with transfer learning method to identify common rice diseases.The classical models were adopted, such as VGG, ResNet, DenseNet, InceptionResNet and Xception to compare the performances on new task and select the Xception model as the final model based on the accuracy and stable performance. The results showed that the recognition accuracy of DenseNet, InceptionResNet and Xception model could reach more than 97%. In particular, Xception model could not only achieve the highest recognition accuracy of 98.50%, but also was the most stable of these models. This study explored the best model for rice diseases intelligent recognition task and showed the effectiveness of using transfer learning method to solve new task.
作者
王忠培
张萌
董伟
朱静波
孔娟娟
钱蓉
WANG Zhong-pei;ZHANG Meng;DONG Wei(Institute of Agricultural Economics and Information,Anhui Academy of Agricultural Sciences,Hefei,Anhui 230031)
出处
《安徽农业科学》
CAS
2021年第20期236-242,共7页
Journal of Anhui Agricultural Sciences
基金
安徽省农业科学院农业大数据研究与应用团队(2021YL051)。
关键词
水稻病害
迁移学习
深度学习
智能识别
Rice disease
Transfer learning
Deep learning
Intelligent recognition
作者简介
王忠培(1981-),男,安徽金寨人,助理研究员,博士,从事智能农业研究;通信作者:董伟,副研究员,硕士,从事农业信息化研究。
