基于颜色和纹理特征的玉米干旱识别

doi:10.11924/j.issn.1000-6850.casb18030021

摘要/Abstract

摘要： 探究以颜色和纹理为特征，以神经网络建模为方法，识别玉米干旱的效果。利用可见光成像方式采集不同干旱胁迫下的玉米图像，通过编程从玉米图像中自动提取颜色和纹理特征变量，以多个BP神经网络集成学习的方法构建玉米不同生长发育阶段的干旱识别模型，识别不同干旱程度的玉米植株。结果表明：玉米出苗—拔节阶段的模型在训练和验证时的平均识别准确率和平均识别精度均在90%以上，识别误差均值小于0.1；拔节—抽雄阶段的模型在训练和验证时识别干旱的平均准确率和平均识别精度均在85%以上，识别误差均值为0.1和0.14；抽雄—成熟阶段的模型在训练和验证时识别干旱的平均准确率分别为85.36%和84.27%，平均精度在均在80%以上，识别误差均值为0.15和0.16。玉米出苗—拔节、拔节—抽雄、抽雄—开花3个阶段的干旱识别模型对田间玉米干旱的平均识别准确率在75%左右，平均识别精度在80%以上，平均识别误差依次为0.22、0.21、0.29。总之，出苗—拔节阶段的玉米干旱识别模型识别干旱的能力最强，拔节—抽雄阶段的模型次之，抽雄—成熟阶段的模型较差，同一模型对同一干旱程度的识别，模型训练时的识别效果最好，验证时的识别效果次之，测试效果较差，同一模型对不同干旱程度的识别，总体表现为对中旱水平识别效果最不理想，对适宜和特旱水平的识别效果最好。研究结果为玉米干旱特征模式识别和利用表型特征实现玉米旱情自动监测预警提供参考。

关键词: 切花红掌, 切花红掌, 天使, 火焰, 玛丽西亚, 光合作用, 日变化

Abstract: 【Objective】The purpose of this paper is to explore the effect of identifying maize drought, for which the color and texture features of maize and the neural network modeling method were used. 【Method】The maize images under different drought stress were collected by visible light imaging method, from which the color and texture features of maize were extracted automatically by programming. Then the drought identification models of different maize growth stages were built by the ensemble learning of multiple BP neural networks. 【Result】The results show that both the average identification accuracy and the average identification precision of the maize emergence-jointing stage model are more than 90% at training and validation, and the average value of training and validation errors are less than 0.1. The average identification accuracy and the average identification precision of the jointing-tasseling stage model are more than 85% at the training and validation, and the average identification errors in training and validation are 0.1 and 0.14. The average drought identification accuracies of the tasseling-maturing stage model at training and validation are 85.36% and 84.27%, and the average precision is more than 80%, and the average identification errors are 0.15 and 0.16. All the maize growth stage drought identification models can identify field maize drought with the about 75% average accuracy and the more than 80% average identification precision, and the average identification error values of three growth stage models are 0.22, 0.21, 0.29. 【Conclusion】In conclusion, the ability of identifying drought of maize emergence-jointing stage model is strongest, followed by the jointing-tasseling stage drought identification model, but the drought identification model for tasseling-maturing stage is relatively poor. For the same drought level identification using the same model, the identification effect at training is best, second at validation, but relatively poor at testing. For the different drought level identifications using the same model, the identification effect for medium drought is not ideal, but better for drought-free and extreme level. The results provide a reference for the maize drought pattern recognition and the automatic monitoring, and lay the foundation for the early warning of maize drought by using phenotypic characteristics.

中图分类号:

S166

岳焕然,李茂松,安江勇. 基于颜色和纹理特征的玉米干旱识别[J]. 中国农学通报, 2018, 34(24): 18-28.

参考文献

[1] Tardieu F, Cabrera-Bosquet L, Pridmore T, et al. Plant Phenomics, From Sensors to Knowledge[J]. Current Biology. 2017, 27(15): R770-R783.
[2] Singh A, Ganapathysubramanian B, Singh A K, et al. Machine Learning for High-Throughput Stress Phenotyping in Plants[J]. Trends in Plant Science. 2016, 12(2): 110-124.
[3] Perez-Sanz F, Navarro P J, Egea-Cortines M. Plant phenomics: an overview of image acquisition technologies and image data analysis algorithms[J]. GigaScience. 2017: x92.
[4] Witt S, Galicia L, Lisec J, et al. Metabolic and Phenotypic Responses of Greenhouse-Grown Maize Hybrids to Experimentally Controlled Drought Stress[J]. Molecular Plant. 2012, 5(2): 401-417.
[5] Araus J L, Serret M D, Edmeades G O. Phenotyping maize for adaptation to drought[J]. Frontiers in Physiology. 2012, 3: 1-20.
[6] Masuka B, Araus J L, Das B, et al. Phenotyping for Abiotic Stress Tolerance in MaizeF[J]. Journal of Integrative Plant Biology. 2012, 54(4): 238-249.
[7] 刘宪锋, 朱秀芳, 潘耀忠, 等. 农业干旱监测研究进展与展望[J]. 地理学报. 2015, 70(11): 1835-1848.
[8] 吴冬平. 我国农业旱情监测技术现状及发展趋势分析[J]. 自然灾害学报. 2015, 24(4):24-35.
[9] 苏志诚, 孙洪泉. 国内外干旱监测评估信息化建设现状分析及建议[J]. 中国防汛抗旱. 2017, 27(03):19-21.
[10] 袁杰, 杜建军, 郭新宇. 基于可见光图像的玉米植株表型检测研究进展[J]. 中国农业科技导报. 2016, 18(06): 90-96.
[11] 唐俊, 邓立苗, 陈辉, 等. 基于机器视觉的玉米叶片透射图像特征识别研究[J]. 中国农业科学. 2014, 47(03): 431-440.
[12] 张超, 乔敏, 刘哲, 等. 基于无人机和卫星遥感影像的制种玉米田识别纹理特征尺度优选[J]. 农业工程学报. 2017, 33(17): 98-104.
[13] Arnal Barbedo J G. Digital image processing techniques for detecting, quantifying and classifying plant diseases[J]. SpringerPlus. 2013, 2(1): 660.
[14] Peter A, Abdulkadir S, Abdulhamid U F. Detection of Optimum Maturity of Maize Using Image Processing and Artificial Neural Networks[J]. Science World Journal. 2017, 12(2): 24-27.
[15] Grzesiak M T, Filek W, Hura T, et al. Leaf optical properties during and after drought stress in triticale and maize genotypes differing in drought tolerance[J]. Acta Physiologiae Plantarum. 2010, 32(3): 433-442.
[16] Zhuang Shuo, Wang Ping, Jiang Boran, et al. Early detection of water stress in maize based on digital images[J]. Computers and Electronics in Agriculture. 2017, 140: 461-468.
[17] 路志英, 刘书辰, 宫志宏. 基于多特征融合的玉米前期图像的旱情识别[J]. 激光与光电子学进展. 2017, 54(8): 135-143.
[18] Ma C, Zhang H H, Wang X. Machine learning for Big Data analytics in plants[J]. Trends in Plant Science. 2014, 19(12): 798-808.
[19] Haindl M, Kittler J, Roli F. Multiple Classifier Systems[M]. Springer Berlin Heidelberg, 2007: 981-986.
[20] 堵纯信, 曹春景, 曹青, 等. 玉米杂交种郑单958的选育与应用[J]. 玉米科学. 2006, 14(06): 43-45.
[21] NY/T 2284-2012. 玉米灾害田间调查及分级技术规范[S]. 2012.
[22] Gonzalez R C, Woods R E, Eddins S L. Digital image processing[M]. Publishing House of Electronics Industry, 2010.
[23] Petrou M, Sevilla P G. Image processing: dealing with texture[M]. DBLP, 2006.
[24] Mishra S, Sarkar U, Taraphder S, et al. Multivariate statistical data analysis-principal component analysis[J]. International Journal of Livestock Research. 2017, 7(5): 60-78.
[25] 徐富强, 刘相国. 基于优化的RBF神经网络的变量筛选方法[J]. 计算机系统应用. 2012, 21(03): 206-208.
[26] Srivastava N, Hinton G, Krizhevsky A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research. 2014, 15(1): 1929-1958.
[27] 沈花玉, 王兆霞, 高成耀, 等. BP神经网络隐含层单元数的确定[J]. 天津理工大学学报. 2008, 24(05): 13-15.
[28] 严威凯. 双标图分析在农作物品种多点试验中的应用[J]. 作物学报. 2010, 36(11): 1805-1819.
[29] Baret F, Madec S, Irfan K, et al. Leaf rolling in maize crops: from leaf scoring to canopy level measurements for phenotyping[J]. bioRxiv. 2017.
[30] Sirault X R R, Condon A G, Wood J T, et al. phenotyping leaf rolling in cereals using computer vision and functional data analysis approaches[J]. Plant Methods. 2015, 11(1):52.
[31] 王传宇, 郭新宇, 杜建军, 等. 基于时间序列图像的玉米植株干旱胁迫表型检测方法[J]. 农业工程学报. 2016, 32(21): 189-195.
[32] Sruti Das Choudhury, Saptarsi Goswami, Srinidhi Bashyam, et al. Automated stem angle determination for temporal plant phenotyping analysis[J]. The IEEE International Conference on Computer Vision(ICCV). 2017: 2022-2029.