摘要
马铃薯在国内的农业资源中占据着重要地位,拥有丰富多样的品种。准确识别马铃薯品种对于推动马铃薯育种发展至关重要。因此,本研究基于深度学习网络提出了一种多品类马铃薯植株识别模型。该模型采用了基于Swin Transformer的架构,并通过对注意力机制的巧妙改进,有效提升了模型的特征提取能力。与此同时,通过减少模型参数量,该模型的准确率得到了显著提升。原始的Swin Transformer在对30个马铃薯品种进行植株识别时的准确率为95.0%,而改进后的Swin Transformer达到了97.1%,提升了2.1个百分点。研究结果明确显示,改进后的Swin Transformer模型在对马铃薯植株进行识别分类方面优于原始Swin Transformer模型。深度学习网络模型在马铃薯植株种类识别方面展现出可行性,为其在实际生产中的推广应用提供了有力支持。
Potatoes play a vital role in China's agricultural landscape,boasting a diverse array of varieties.The accurate identification of these varieties is crucial for advancing potato breeding.This study proposes a multi-variety potato plant identification model based on the deep learning network.The model leverages the architecture of the Swin Transformer,enhancing feature extraction through clever improvements in the attention mechanism.Simultaneously,a reduction in the number of model parameters results in a significant boost in accuracy.The original Swin Transformer achieves 95.0%accuracy in identifying 30 potato varieties.In contrast,the enhanced Swin Transformer achieves an impressive 97.1%,marking a substantial 2.1 percentage point improvement.These results unequivocally demonstrate the superiority of the enhanced Swin Transformer model in the identification and classification of potato plants over its original counterpart.The deep learning network model has shown feasibility in potato plant species identification,which provides strong support for its popularization and application in actual production.
作者
马柏雄
刘成忠
韩俊英
曲亚英
邢雪
MA Baixiong;LIU Chengzhong;HAN Junying;QU Yaying;XING Xue(College of Information Science and Technology,Gansu Agricultural University,Lanzhou 730070,China;Potato Research Institute of Gansu Academy of Agricultural Sciences,Lanzhou 730070,China)
出处
《智能计算机与应用》
2025年第9期12-18,共7页
Intelligent Computer and Applications
基金
国家自然科学基金(32360437)
甘肃省高等学校产业支撑计划项目(2021CYZC-57)
甘肃省高等学校创新基金项目(2021A-056)。
