摘要
针对目前当归产业病虫害识别方法缺失、人工提取特征存在主观因素及卷积神经网络训练需要大量数据等不足,提出1种基于多卷积神经网络融合的当归病虫害识别方法。构建当归常见病虫害数据集;选择在当归病虫害数据集中表现性能最好的ResNet50、InceptionNetV3、VGG19、DenseNet2014个网络作为模型融合的基学习器;使用XGBoost(极度梯度提升)算法作为元学习器,得到基于多卷积神经网络融合的当归病虫害识别模型。结果表明,该融合模型比单个卷积神经网络模型具有更高的识别准确率,并优于其他融合方法融合的模型,对当归病虫害识别的查准率、查全率、F 1值分别达到98.33%、97.14%、97.68%。本研究提出的基于XGBoost融合方法融合的模型实现了当归常见病虫害的精确分类,对常见病害的识别准确率达到98.33%,为当归产业提供了一种有效的病虫害识别方法。
Aiming at the lack of identification methods for diseases and insect pests in Angelica sinensis industry,the subjective factors in the process of artificial feature extraction and the large amount of data required for training convolutional neural networks,a method for identification of diseases and insect pests of Angelica sinensis based on the fusion of multiple convolutional neural networks was proposed.The dataset of common diseases and insect pests of Angelica sinensis was constructed.Four networks,ResNet50,InceptionNetV3,VGG19 and DenseNet201,with the best performance on the dataset of Angelica sinensis were selected as the base learner for model fusion.The XGBoost(extreme gradient boosting)algorithm was used as a meta-learner to obtain the Angelica sinensis diseases and insect pests recognition model based on the fusion of multiple convolutional neural networks.The results showed that the fusion model had higher recognition accuracy than a single convolutional neural network model,and was superior to other fusion models.The precision rate,recall rate and F 1 value of the identification of Angelica sinensis pests and diseases reached 98.33%,97.14%and 97.68%,respectively.The model based on XGBoost fusion method proposed in this study realized the accurate classification of common diseases and insect pests of Angelica sinensis,and the identification accuracy rate of common diseases reached 98.33%,which provided an effective method for identification of diseases and insect pests in the industry of Angelica sinensis.
作者
郭标琦
王联国
GUO Biao-qi;WANG Lian-guo(College of Information Science and Technology,Gansu Agricultural University,Lanzhou 730000,China)
出处
《江苏农业学报》
CSCD
北大核心
2024年第1期121-129,共9页
Jiangsu Journal of Agricultural Sciences
基金
甘肃省重点研发计划项目(21YF5GA088)。
关键词
当归病虫害分类
卷积神经网络
极度梯度提升(XGBoost)融合方法
classification of Angelica sinensis diseases and insect pests
convolutional neural network
extreme gradient boosting(XGBoost)fusion method
