| 摘 要: 为促进玉米产业发展和种质资源保护,提出了基于改进MobileViT的MR-MobileViT模型。首先,采用MobileViT作为基础网络模型,并将其激活函数从SiLU替换为 Meta-ACON-C;其次,在特征融合阶段加入残差连接,增强特征表达的丰富性。实验结果表明,相较于MobileViT基础网络,MR-MobileViT在准确率、精确率、召回率和F1值上分别提升了2.60%、2.40%、2.72%和2.69%。与常用模型(如 RegNet、ResNet、GoogLeNet)相比,准确率分别提高了0.77%、0.94%和4.73%。该研究表明,MR-MobileViT在玉米拔节期品种识别中表现优异,为玉米品种的准确识别提供了新的解决方案。 |
| 关键词: 玉米拔节期品种识别 MobileViT Meta-ACON-C 残差连接 |
|
中图分类号:
文献标识码: A
|
|
| Research on Maize Variety Identificationat Jointing Stage Based on Improved MobileViT |
|
YI Qijia1, LIU Chengzhong1, HAN Junying1, ZHOU Yuqian2, LI Yongsheng2
|
(1.College of Information Science and Technology, Gansu Agricultural University, Lanzhou 730070, China;
2.Crops Research Institute, Gansu Academy of Agricultural Sciences, Lanzhou 730070, China)
1339586348@qq.com; liucz@gsau.edu.cn; hanjy@gsau.edu.cn; yuqianzhou2008@163.com; lys087@163.com
|
| Abstract: To promote the development of the maize industry and the conservation of germplasm resources, this study proposes an MR-MobileViT model based on an improved MobileViT. Firstly,MobileViT is adopted as the base network model, with its activation function replaced from SiLU to Meta-ACON-C. Secondly, residual connections are incorporated during the feature fusion stage to enhance the richness of feature expression. Experimental results show that, compared to the base MobileViT network, MR-MobileViT achieves improvements of 2.60% , 2.40% , 2.72% , and
2.69% in accuracy, precision, recall, and F1-score, respectively. When evaluated against widely used models (e.g.,RegNet, ResNet, GoogLeNet), its accuracy increases by 0. 77% , 0. 94% , and 4. 73% , respectively. This study demonstrates that MR-MobileViT excels in maize variety identification at the jointing stage, providing a novel solution for accurate maize variety recognition. |
| Keywords: maize jointing stage variety identification MobileViT Meta-ACON-C residual connections |
