一种基于改进卷积神经网络的葡萄叶片病害集成识别方法
An Ensemble Recognition Method for Grape Leaf Diseases Based on an Improved Convolutional Neural Network
陈诗瑶 1孔淳 1冯峰 1王志军 1孙博2
作者信息
- 1. 山东农业大学信息科学与工程学院,山东 泰安 271018;山东省苹果技术创新中心,山东 泰安 271018
- 2. 山东农业大学信息科学与工程学院,山东 泰安 271018
- 折叠
摘要
为有效提升葡萄叶片病害识别的精度和效率,实现葡萄病害的及时防治进而提高产量和质量,本文提出一种基于改进卷积神经网络的葡萄叶片病害集成识别方法,对常见的三种葡萄叶片病害进行自动准确的识别.首先,利用Bagging集成学习算法生成多个有差异的训练子集;然后,将SE、CA注意力机制分别引入ResNet152、DenseNet121与MobileNetV3模型,得到改进后的三种神经网络基学习模型,并在生成的训练子集上进行训练;最后,利用加权平均的思想将这些模型进行集成.在葡萄叶片病害数据集上进行的实验表明,该集成模型的识别准确率达到了99.38%,因而是一种比较有效的葡萄叶片病害识别方法.
Abstract
To effectively improve the accuracy and efficiency of grape leaf diseases recognition,and to achieve timely prevention and control of grape diseases,thereby improving yield and quality,this paper proposes an ensemble recognition method for grape leaf diseases based on an improved convolutional neural network.Initially,the Bagging ensemble learning algorithm is used to generate multiple diverse training subsets;Subsequently,the SE(Squeeze-and-Excitation)and CA(Channel Attention)attention mechanisms are respectively integrated into the ResNet152,DenseNet121 and MobileNetV3 models,resulting in three improved neural network-based learning models,which are then trained on the generated training subsets.Finally,these models are integrated using the idea of weighted averaging.Experiments conducted on a grape leaf diseases dataset demonstrate that the recognition accuracy of this ensemble model reaches 99.38%,making it a relatively effective method for grape leaf diseases recognition.
关键词
葡萄叶片病害识别/卷积神经网络/集成学习/Bagging算法/图像识别Key words
Grape leaf diseases recognition/convolutional neural network/ensemble learning/bagging algorithm/image identification引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据