| 摘要: |
| 为实现水稻氮素营养的快速、准确识别，采用改进的VGG 16网络和迁移学习相结合的水稻氮素营养诊断识别方法，以杂交稻‘两优培九’为试验对象进行田间试验，设置4组不同的施氮水平(施氮量分别为0、210、300和390 kg/hm2)，在水稻幼穗分化期和齐穗期，扫描获取水稻叶片图像数据；通过图像预处理方法，对数据进行扩充；构建改进的VGG16和迁移学习相结合的网络模型对水稻叶片图像数据进行氮素营养诊断识别。结果表明:1)在幼穗分化期时，改进的VGG16网络的识别准确率为93.1%，模型大小约为迁移学习VGG16模型的1/6，训练时间约为1 261 s。2)在水稻幼穗分化期和齐穗期，该模型微调后的识别准确率均能达到95%以上。基于迁移学习和改进的VGG16网络所建立的水稻氮素营养诊断模型具有较好的泛化能力，可以预测水稻氮素营养状况，为水稻氮素营养诊断提供参考。 |
| 关键词: 水稻 氮素营养诊断 VGG16网络 迁移学习 微调 |
| DOI：10.11841/j.issn.1007-4333.2023.06.20 |
| 投稿时间：2022-07-31 |
| 基金项目:国家自然科学基金项目(62162030；61562039) |
|
| Nitrogen nutrition diagnosis in rice based on improved VGG16 network and transfer learning |
|
ZHANG Linpeng1,YANG Hongyun1*,QIAN Zheng1,GUO Ziwei2
|
| (1.School of Software, Jiangxi Agricultural University, Nanchang 330045, China;2.School of Computer and Information Engineering, Jiangxi Agricultural University, Nanchang 330045, China) |
| Abstract: |
| In order to realize the rapid and accurate identification of nitrogen nutrition in rice, an improved VGG 16 network combined with transfer learning was used to diagnose and identify rice nitrogen nutrition. Nitrogen a application rates were respectively 0, 210, 300 and 390 kg/hm2 at the young panicle differentiation stage and full heading stage of rice. The rice leaf image data were obtained scanning and the data were then processed and expanded by image preprocessing. A network model was constructed by combining the improved VGG16 and transfer learning for the nitrogen nutrition diagnosis and identification of rice leaf image data. The results show that: 1)At the differentiation stage of young ears, the recognition accuracy of the improved VGG16 network is 93. 1%, the model size is about 1/6 of that of the transfer learning VGG16 model, and the training time is about 1 261 s. 2)The recognition accuracy of the model after fine-tuning can reach more than 95% in the young panicle differentiation and full heading stages of rice. In conclusion, the rice nitrogen nutrition diagnosis model based on transfer learning and the improved VGG16 network has good generalization ability, which can predict the nitrogen nutrition situation of rice, and can provide a reference for the nitrogen nutrition diagnosis of rice. |
| Key words: rice nitrogen nutrition diagnosis VGG16 network transfer learning fine-tuning |
