基于InceptionV3的烟草病害识别

  背景和目的  烟草病害严重制约影响烟草的产量及质量。为解决烟草病害诊断方式精准性差、效率低等问题。  方法  以5种较为常见的烟草病害（烟草花叶病、黄瓜花叶病毒病、烟草赤星病、烟草野火病、烟草气候性斑点病）为研究对象，基于InceptionV3网络使用迁移学习方法构建烟草病害识别模型，对比测试原始数据集、数据增强后数据集、MSRCR数据集和图像融合数据集。  结果  图像融合数据集的识别准确率为90.80%，平均识别时间为1.33 s，比原始数据集的识别准确率（70.00%）提高了29.71%。  结论  该方法能快速准确识别烟草病害，可为烟草病害的防治提供理论基础。 

Tobacco disease identification based on InceptionV3

Tobacco is an important economic crop in China. Tobacco leaf diseases are various and complicated. At present, the diagnostic methods of tobacco diseases still have some problems such as poor precision and low efficiency. In order to solve the above problems, this article takes the five common tobacco diseases (tobacco common mosaic disease, tobacco cucumber mosaic virus disease, tobacco brown spot disease, tobacco wildfire, tobacco climate spot disease) as the research objects, and proposes an InceptionV3 network based tobacco disease identification model. Using the transfer learning, the proposed model was trained on four different datasets, including the original dataset, the data enhanced dataset, the MSRCR dataset, respectively. The experiments results show that the proposed tobacco disease identification model achieved a disease identification accuracy rate of 90.80% on the image fusion data set, which is 29.71% higher than the recognition accuracy rate (70.00%) on the original dataset, and an average identification time of 1.33s, . The proposed method realizes the rapid and accurate identification of tobacco diseases. This paper provides a theoretical basis for the prevention and control of tobacco diseases.