近红外光谱与荧光光谱对比的黄瓜白粉病分割与检测

Segmentation and Detection of Cucumber Powdery Mildew by Comparison of Near-Infrared and Fluorescence Spectra

黄瓜白粉病是一种传播速度快、发生频率高的蔬菜病害,一旦爆发将对产量产生严重的打击,因此对黄瓜白粉病的识别与尽早防治具有重要的意义。采用便携式光谱仪采集了黄瓜叶片的近红外光谱反射率曲线与荧光光谱强度曲线,采用LI-6400光合作用测量仪测量叶片的光合速率,并采集了叶片的图像信息。首先,采用图像分割技术对白粉病进行等级划分;其次,对净光合速率与光谱之间进行相关性分析;最后,利用定性分析以及定量预测两种方法,结合黄瓜患白粉病叶片及健康叶片的光合速率指标建立白粉病检测模型。从分析结果可知,利用二值化将黄瓜叶片区域作为感兴趣区域(ROI)分割出,根据R-G-B与L^(*)a^(*)b^(*)色彩空间中颜色的差异可以有效提取白粉病斑面积;通过皮尔逊相关性分析光合速率与光谱之间的相关性强度,得到光合速率与光谱具有较强的负相关,并且随着反射率及光谱强度的增高,相关性减弱,表明采用光谱及相关性较大的波段对光合速率进行预测具有可行性;经过准确率比较,选择集成学习(ensemble learner)中的子空间判别(subspace discriminant)算法对定性模型进行最终分析,得到近红外光谱模型更加稳定,识别准确率更高;采用偏最小二乘回归模型(PLSR)进行定量预测,通过比较7种不同的预处理方法,验证得知MSC预处理可以有效去除光谱干扰信息,其中近红外光谱模型R^(2)更高,且RMSEP<RMSEC。预测值与实际测量值对比可知,近红外光谱模型得出的预测值与实际测试值更相近,且健康样本与患白粉病样本区分明显,表明该模型具有更高的鲁棒性。结果表明,利用近红外光谱与光合速率指标相结合建立的模型以及图像识别系统可以实现对黄瓜白粉病的快速识别与病情分级,为黄瓜病害诊断提供了方法和参考依据。

As a disease with fast transmission speed and high frequency,cucumber powdery mildew will deal a serious blow to cucumber yield once it breaks out;therefore,it is of great significance for the identification and early prevention of cucumber powdery mildew.This study,used a portable spectrometer to collect the reflectance curves of near-infrared(NIR)spectral and the intensity curves of fluorescence spectral of cucumber leaves.LI-6400 photosynthetic meter was used to measure the photosynthetic rate of cucumber leaves,and we also collected the image information of cucumber leaves.Firstly,powdery mildew was classified by image segmentation.Secondly,the Pearson Correlation between net photosynthetic rate and spectrum was analyzed.Finally,Finally,a powdery mildew detection model was established using qualitative analysis and quantitative prediction methods combined with photosynthetic rate indexes of cucumber leaves.The results showed that the cucumber leaf region was segmented by binarization as the region of interest(ROI),and the powdery mildew spot area could be extracted effectively according to the color difference between RGB and L^(*)a^(*)b^(*) color space.Pearson Correlation analyzed the correlation between the photosynthetic rate and the spectrum.Results showed that the photosynthetic rate and the spectrum were negatively correlated.The correlation weakened with the increasing reflectivity and spectral intensity,which indicated that it is feasible to predict the photosynthetic rate using bands with intense spectral correlations.After comparing the prediction accuracy,the qualitative model was finally analyzed by the Subspace Discriminant algorithm in Ensemble Learner,and the NIR spectrum model was stable,and the recognition accuracy was high.The PLSR model was used for quantitative analysis,and the MSC was used as a preprocessing method to effectively remove spectral interference information,of which the R^(2) obtained by the NIR spectrum model was high,and the RMSEP was smaller than the RMSEC.In addition,the predicted results of the NIR spectral model were more similar to the expected values,and the healthy samples were clearly distinguished from the ones with powdery mildew infection,indicating that the model is highly robust.The above results showed that the image recognition system and the photosynthetic rate detection model based on NIRspectroscopy could be used to identify cucumber powdery mildew and classification quickly and accurately,which provided a method and reference for the diagnosis of cucumber disease.