摘要
针对传统的种子鉴别方法主要依靠人工鉴别,主观性强,费时费力,效率低下。以鉴别玉米品种登海605为例,利用深度学习和卷积神经网络相结合的方法,构建胚面、胚乳、双面混合3类数据集。对VGG16模型用不同FT(Fine-tuning)策略进行微调,结果表明,在FT75%训练策略下模型的测试准确率最高,在3类数据集上均为100%,同时在FT75%-VGG16探讨了不同数量全连接层神经元数量对网络性能的影响,最终选定2048为最终神经元数量。
Traditional seed identification is mainly conducted by manual labor,which is subjective,time-consuming and inefficient.In this paper,taking the identification of maize variety Denghai 605 as an example,three kinds of data sets including embryo surface,endosperm and double-sided mixture were constructed by using the method of combining the deep learning with convolution neural network.The VGG16 model was fine-tuned with different FT(fine-tuning)strategies.The results showed that the model had the highest test accuracy under the FT75%training strategy,which was 100%on all three types of data sets.And the influence of different numbers of neurons in the full connection layer on the network performance was discussed in FT75%-VGG16,and 2048 was finally selected as the final number of neurons.
作者
王佳
马睿
赵威
郭宏杰
马德新
Wang Jia;Ma Rui;Zhao Wei;Guo Hongjie;Ma Dexin(Qingdao Agricultural University,Qingdao 266109)
出处
《中国粮油学报》
CSCD
北大核心
2023年第8期229-234,共6页
Journal of the Chinese Cereals and Oils Association
基金
山东省重点研发计划项目(2019GNC106001)
青岛市民生科技计划项目(18-6-1-112-nsh)
淄博市重点研发计划项目(2019gy010101)
山东省高等学校青创人才引育计划项目(202202027)。
