Identification of Fusarium Head Blight in Winter Wheat Ears Using Continuous Wavelet Analysis
Identification of Fusarium Head Blight in Winter Wheat Ears Using Continuous Wavelet Analysis
复制标题
利用连续小波分析识别冬小麦穗赤霉病
DOI：
10.3390/s20010020
复制
发表时间：
2020-01-01
期刊：
影响因子：
3.9
通讯作者：
Jiang, Jing
中科院分区：
文献类型：
--
作者：
Ma, Huiqin;Huang, Wenjiang;Jiang, Jing
冬小麦赤霉病产生的高毒真菌毒素脱氧雪腐镰刀菌烯醇（DON）是严重影响人畜健康的问题。直接在穗上鉴定病害对于选择性收获是重要的。利用连续小波分析（CWA）技术对小麦穗的反射光谱（350 ~ 2500 nm）进行光谱识别。首先，对每个反射光谱进行连续小波变换，并生成作为波长位置和分解尺度的函数的小波功率尺度图。用线性回归法计算了小波功率与侵染率的决定系数R2。保留基于R2值以降序排列的前5%区域与统计学显著（t检验的p值< 0.001）小波区域的交叉点作为敏感小波特征区域。每个敏感区域的R2值最高的小波功率被保留作为初始小波特征。通过对初始小波特征之间的相关性分析，得到相关系数R，并根据相关系数R设定阈值选择最优小波特征。结果识别出六个小波特征，包括（471 nm，尺度4）、（696 nm，尺度1）、（841 nm，尺度4）、（963 nm，尺度3）、（1069 nm，尺度3）和（2272 nm，尺度4）。利用Fisher线性判别分析建立了基于6个小波特征的赤霉病鉴别模型。该模型表现良好，提供了88.7%的总体精度和0.775的kappa系数，这表明使用CWA获得的光谱特征可以潜在地反映冬小麦穗赤霉病的侵染。
Fusarium head blight in winter wheat ears produces the highly toxic mycotoxin deoxynivalenol (DON), which is a serious problem affecting human and animal health. Disease identification directly on ears is important for selective harvesting. This study aimed to investigate the spectroscopic identification of Fusarium head blight by applying continuous wavelet analysis (CWA) to the reflectance spectra (350 to 2500 nm) of wheat ears. First, continuous wavelet transform was used on each of the reflectance spectra and a wavelet power scalogram as a function of wavelength location and the scale of decomposition was generated. The coefficient of determination R2 between wavelet powers and the disease infestation ratio were calculated by using linear regression. The intersections of the top 5% regions ranking in descending order based on the R2 values and the statistically significant (p-value of t-test < 0.001) wavelet regions were retained as the sensitive wavelet feature regions. The wavelet powers with the highest R2 values of each sensitive region were retained as the initial wavelet features. A threshold was set for selecting the optimal wavelet features based on the coefficient of correlation R obtained via the correlation analysis among the initial wavelet features. The results identified six wavelet features which include (471 nm, scale 4), (696 nm, scale 1), (841 nm, scale 4), (963 nm, scale 3), (1069 nm, scale 3), and (2272 nm, scale 4). A model for identifying Fusarium head blight based on the six wavelet features was then established using Fisher linear discriminant analysis. The model performed well, providing an overall accuracy of 88.7% and a kappa coefficient of 0.775, suggesting that the spectral features obtained using CWA can potentially reflect the infestation of Fusarium head blight in winter wheat ears.