English

中国农机化学报

中国农机化学报 ›› 2022, Vol. 43 ›› Issue (7): 116-123.DOI: 10.13733/j.jcam.issn.20955553.2022.07.017

• 农业信息化工程 • 上一篇    下一篇

基于特征选择的小麦籽粒品种识别研究

冯继克1,郑颖1,李平1,李艳翠2,张自阳3,郭晓娟1   

  1. 1. 河南科技学院信息工程学院,河南新乡,453003; 
    2. 河南师范大学计算机与信息工程学院,河南新乡,453000;

    3. 河南科技学院生命科技学院,河南新乡,453003
  • 出版日期:2022-07-15 发布日期:2022-06-27
  • 基金资助:
    河南省科技攻关项目(212102110298、182102210048、222102210171、212102210431);河南省高等学校重点科研项目计划(20A520013)

Research on the identification of wheat grain varieties based on feature selection

Feng Jike, Zheng Ying, Li Ping, Li Yancui, Zhang Ziyang, Guo Xiaojuan.   

  • Online:2022-07-15 Published:2022-06-27

摘要: 为有效地对小麦籽粒品种进行分类,判别影响小麦籽粒品种识别的特征,进行基于特征选择的小麦籽粒品种识别研究。首先采集农大3416-18、内乐288、衡水6632、百农419、洛麦28和新麦26六个品种的小麦籽粒图像18 000张,对采集的图像进行预处理,提取小麦籽粒的颜色特征、形态特征和纹理特征三大类共28个特征值,并对特征进行相关性分析。然后分别构建不同特征融合模型以及数据降维和数据增强模型。最后进行试验分析,基于纹理+形态+颜色三个特征融合模型平均识别准确率为91.02%,其中基于纹理+形态+颜色特征模型的洛麦28识别率最高,达97.0%;经过线性判别分析,降维处理的小麦特征数据识别准确率达86.19%,模型训练时间仅0.87 s;基于数据增强后的平均识别准确率达94.26%。试验表明基于特征选择的小麦籽粒识别是可行的,有助于育种工作者对小麦籽粒识别做出更准确判断,具有一定的实际意义。


关键词: 小麦籽粒, 特征选择, 品种识别, 特征融合, 数据降维

Abstract: In order to effectively classify wheat grain varieties and identify the characteristics that affect the identification of wheat grain varieties,  this paper research the identification of wheat grain varieties based on feature selection. First,  18 000 images of wheat grains from six varieties of Nongda 3416-18,  Neile 288,  Hengshui 6632,  Bainong 419,  Luomai 28,  and Xinmai 26 were collected. After preprocessing the collected images,  a total of 28 feature values were extracted in three categories of color,  morphology,  and texture features of wheat grains,  and the correlation analysis of features was carried out. Then,  different feature fusion models,  feature data dimensionality reduction models,  and data enhancement models were constructed. Finally,  the experimental analysis showed that the average accuracy of the threefeature fusion model based on texture + shape + color was 91.02%. Luomai 28,  based on the texture + shape + color feature model,  was the highest,  reaching 97.0%. After dimension reduction by Linear Discriminant Analysis (LDA),  the recognition accuracy of wheat feature data recognition was 86.19%,  and the training time of the model was only 0.87 s. After data enhancement,  the average recognition accuracy reached 94.26%. The experiments showed that the wheat grain recognition based on feature selection was feasible,  which was helpful and practical for breeders to make more accurate judgments on the identification of wheat grains.

Key words: wheat grain, feature selection;variety recognition;feature fusion, data dimensionality reduction

中图分类号: