高光谱成像鉴别玉米品种早期抗倒性

doi:10.3964/j.issn.1000-0593(2022)04-1229-06

摘要
参考文献
相关文章 (15)

全文: PDF (2788 KB)
输出: BibTeX | EndNote (RIS)

摘要：针对传统玉米品种抗倒性鉴别方法费时费力、时效滞后的问题，采用高光谱成像数据结合机器学习方法对9叶期的玉米品种抗倒性进行鉴别，并给出适于进行玉米品种抗倒性鉴别的种植密度和建模方法。试验设置了5 000，7 000和9 000株·亩^-13个种植密度和6个典型的抗倒/不抗倒玉米品种，采集9叶期玉米顶叶的高光谱图像，使用目标区域分割的方式自动进行光谱图像反射率校正和目标光谱曲线提取。对采集的样本数据使用Kennard Stone算法划分样本训练集和测试集，用主成分分析法(PCA）和连续投影算法（SPA）提取光谱特征，建立了基于高斯核函数的支持向量机（SVM）模型并进行参数训练和优化。通过对不同种植密度下各特征提取方法的效果和各模型训练效果及其预测结果的对比，找到进行玉米抗倒性鉴别的最佳种植密度和建模方法。试验结果表明：在各种植密度下PCA方法对光谱特征的降维效果最为显著，而SPA算法选择的特征波长分布比较均匀、抗倒性分类特征比较明显；种植密度的增加对于玉米品种抗倒性的鉴别是有益的，在种植密度为7 000株·亩^-1时，使用SPA-SVM方法建立的模型训练效果和预测结果最佳，此时模型对训练集数据的10折交叉验证正确率为97.40%，对测试集数据的预测正确率为98.33%。

关键词：玉米抗倒；高光谱成像；主成分分析；连续投影；支持向量机

Abstract：Given the time-consuming, labor-intensive and time-lagging problems of traditional methods for identifying lodging resistance of maize, this study uses hyperspectral imaging data combined with machine learning methods to identify lodging resistance of maize at the 9-leaf stage. It gives the recommended planting density and modeling methods. The experiment set up 3 planting densities of 5 000, 7 000 and 9 000 plants·mu^-1 and 6 typical lodging resistant/non-lodging resistant varieties. Hyperspectral images of corn top leaves at the 9-leaf stage were collected, reflectance correction and target spectral curve extraction were automatically performed by segmentation of the target area. For the collected sample data, the Kennard Stone algorithm is used to divide the sample training set and the test set, the principal component analysis (PCA) and the successive projections algorithm (SPA) are used to extract the spectral features. A support vector machines (SVM) model of Gaussian kernel function is established, with the performing of parameter training and optimization. By comparing the effect of each feature extraction method and the training effect of each model, and its prediction results under different planting densities, the planting density and modeling method recommended for the identification of maize lodging resistance were found. The test results show that the PCA method has the most significant dimensionality reduction effect on the spectral features at various planting densities. At the same time, the characteristic wavelength distribution selected by the SPA algorithm is relatively uniform, and the lodging resistance classification characteristics are obvious. The increase of planting density is beneficial to identifying the lodging resistance of maize. When the planting density is 7 000 plants·mu^-1, the training effect and prediction results of the model established by the SPA-SVM method are the best. The 10-fold cross-validation accuracy of the model on the training set data is 97.40%. The prediction accuracy rate of the set data is 98.33%.

Key words：Maize lodging resistance; Hyperspectral imaging; PCA; SPA; SVM

收稿日期: 2021-04-09 修订日期: 2021-06-24

中图分类号:

S126

基金资助: 国家现代玉米产业技术体系建设项目（CARS-02），国家重点研发计划项目（2016YFD0300302）资助

通讯作者: 杨丽 E-mail: yangli@cau.edu.cn

作者简介: 张天亮，1993年生，中国农业大学工学院博士研究生 e-mail: tianliangzn@163.com

引用本文:

张天亮，张东兴，崔涛，杨丽，解春季，杜兆辉，钟翔君. 高光谱成像鉴别玉米品种早期抗倒性[J]. 光谱学与光谱分析, 2022, 42(04): 1229-1234.
ZHANG Tian-liang, ZHANG Dong-xing, CUI Tao, YANG Li, XIE Chun-ji, DU Zhao-hui, ZHONG Xiang-jun. Identification of Early Lodging Resistance of Maize by Hyperspectral Imaging Technology. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2022, 42(04): 1229-1234.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2022)04-1229-06 或 https://www.gpxygpfx.com/CN/Y2022/V42/I04/1229

[1] WU Qiong, YANG Ke-jun, ZHANG Yi-fei, et al(吴琼，杨克军，张翼飞, 等). Journal of Maize Sciences（玉米科学）, 2017, 25(4)：40.
[2] Zhang Ying, Du Jianjun, Wang Jinglu, et al. Biological Research, 2018, 51(1): 40.
[3] TIAN Zai-min, HUANG Zhi-hong, CHEN Jian-xin, et al（田再民，黄智鸿，陈建新, 等）. Journal of Maize Sciences（玉米科学）, 2016, 24(5): 83.
[4] Qin Haiming, Wang Cheng, Zhao Kaiguang, et al. PLOS ONE, 2018, 13(5): e197510.
[5] Feng L, Zhu S, Zhang C, et al. Molecules, 2018, 23(12): 3078.
[6] LI Xiao-bin, WANG Yu-shun, FU Li-hong, et al(李晓斌，王玉顺，付丽红，等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016，36(12)：179.
[7] Chen Y, Zhao X, Jia X. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015, 8(6): 2381.
[8] ZHAO Fan，YAN Zhao-ru，XUE Jian-xin, et al(赵凡，闫昭如，薛建新, 等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2021, 41(1): 201.
[9] Chang C, Lin C. ACM Transactions on Intelligent Systems and Technology, 2011, 2(3): 1.
[10] XIE Wen-yong，CHAI Qin-qin，GAN Yong-hui, et al(谢文涌，柴琴琴，甘勇辉, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2020, 36(14): 203.
[11] Arthur D, Vassilvitskii S. K-Means++: The Advantages of Careful Seeding, Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, 2007. 1027.
[12] Munera S, Amigo J M, Aleixos N, et al. Food Control, 2018, 86: 1.
[13] Araújo M C U, Saldanha T C B, Galväo R K H, et al. Chemometrics and Intelligent Laboratory Systems, 2001, 57(2): 65.
[14] Wu Y, López S, Zhang B, et al. Journal of Real-Time Image Processing, 2019, 16(1): 99.
[15] Wang L, Chang C, Lee L, et al. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(9): 4887.