| 基于快照集成卷积神经网络的苹果叶部病害程度识别 |
| |
| 引用本文： | 刘斌,徐皓玮,李承泽,宋鸿利,何东健,张海曦.基于快照集成卷积神经网络的苹果叶部病害程度识别[J].农业机械学报,2022,53(6):286-294. |
| |
| 作者姓名： | 刘斌 徐皓玮 李承泽 宋鸿利 何东健 张海曦 |
| |
| 作者单位： | 1. 西北农林科技大学信息工程学院;2. 农业农村部农业物联网重点实验室;3. 西北农林科技大学机械与电子工程学院;4. 陕西省农业信息感知与智能服务重点实验室 |
| |
| 基金项目： | 陕西省重点研发计划项目(2021NY-138)；;国家重点研发计划项目(2020YFD1100601-02-13)；;陕西省重点研发计划项目(2019ZDLNY07-06-01)；;国家级大学生创新创业训练计划项目(S202010712083)； |
| |
| 摘 要： | 针对苹果叶部病害程度识别准确率低的问题，构建了一种基于快照集成方法的苹果叶部病害程度识别模型。首先，通过多种数字图像处理技术对原始苹果叶部病害图像进行数据增强；然后，选取Inception-ResNet V2作为基模型，引入CBAM模块提升网络的特征提取能力，使用焦点损失函数缓解苹果叶部病害数据集类别不平衡问题；最后，通过快照集成方法进行模型集成，得到苹果叶部病害程度识别模型。利用苹果黑星病和锈病的早期和晩期病害数据集进行了模型验证，准确率高达90.82%,比单一Inception-ResNet V2模型的准确率提高了2.50个百分点。实验结果表明，基于快照集成的识别模型准确率较高，为苹果叶部病害程度识别研究提供了参考。
|
| 关 键 词： | 苹果叶部 病害识别 卷积神经网络 CBAM模块 焦点损失函数 快照集成 |
| 收稿时间： | 2021/7/12 0:00:00 |
Apple Leaf Disease Identification Method Based on Snapshot Ensemble CNN |
| |
| LIU Bin,XU Haowei,LI Chengze,SONG Hongli,HE Dongjian,ZHANG Haixi.Apple Leaf Disease Identification Method Based on Snapshot Ensemble CNN[J].Transactions of the Chinese Society of Agricultural Machinery,2022,53(6):286-294. |
| |
| Authors: | LIU Bin XU Haowei LI Chengze SONG Hongli HE Dongjian ZHANG Haixi |
| |
| Institution: | Northwest A&F University |
| |
| Abstract: | To address the problem of low recognition accuracy for identifying different apple leaf diseases, an apple leaf disease identification model was proposed based on snapshot ensemble. Firstly, the original dataset was augmented by various digital image processing methods. Then, an Inception-ResNet V2 was chosen as base model. The convolutional block attention module (CBAM) was introduced to enhance the feature extraction capability for apple leaf diseases. And focal loss was used to alleviate the imbalance of samples in each category. Finally, the model was integrated through snapshot ensemble to obtain the final identification model for different degrees of diseases on apple leaves. The image was input to the final model for identification. Compared with the original single Inception-ResNet V2, the recognition accuracy of the improved model was increased from 88.32% to 90.82%. Experimental results showed that the ensemble model had a high accuracy rate, which provided an idea and explored a approach for diseases of different degrees on apple leaves. |
| |
| Keywords: | apple leaf disease identification CNN CBAM module focal loss snapshot ensemble |
|
| 点击此处可从《农业机械学报》浏览原始摘要信息 |
| 点击此处可从《农业机械学报》下载免费的PDF全文 |
