基于低空遥感成像技术的油菜菌核病检测研究

发布时间：2018-06-17 10:37

本文选题：油菜菌核病 + 光谱技术　；参考：《浙江大学》2017年硕士论文

【摘要】：油菜是我国最为主要的油料作物之一,但其产量和品质往往因病害而出现大幅下降。因此,实现快速有效地病害检测并制定合理的防治措施,对保障油菜的产量和品质具有重要意义。传统的病害检测方法通常局限于实验室环境下的微观尺度检测,不仅流程繁琐、滞后性较强,同时会令研究样本遭到破坏,很难满足现代农业对精准生产的要求。为克服传统方法所存在缺陷,本研究以油菜作为研究对象,利用搭载高光谱成像仪和热红外成像仪的无人机模拟平台,分别从冠层尺度和叶片尺度对健康及染病的油菜样本实现了判别分析。主要研究内容如下:(1)从冠层尺度获取高光谱图像数据,对健康和染病油菜样本进行了检测研究。分别采用移动平均法(MAS)、多项式卷积平滑法(SG)、标准正态变量变换(SNV)、多元散射校正(MSC)及去趋势化法(De-trending)对获取到的冠层高光谱数据进行预处理,得到图像中整株样本的光谱反射值。其后,分别基于全波段和特征波长建立偏最小二乘法(PLS-DA)、支持向量机(SVM)、极限学习机(ELM)、K-近邻分类算法(KNN)模型,进行健康和染病样本的判别分析。在基于全波段信息的分析中,不同预处理方法和建模方法组合性能差异明显,其中采用MSC预处理结合ELM建模方法得到的分类效果最优,建模集和预测集的结果达到100%。然后,分别采用连续投影算法(SPA)、二阶导数(2ndDer)及遗传算法-偏最小二乘法(GA-PLS)进行特征波长提取,并利用提取的特征波长建立PLS-DA、SVM、ELM及KNN模型进行分析。结果显示基于SPA所选取特征波长结合ELM模型得到的分类效果最优,建模集和预测集的结果都达到100%。(2)基于光谱数据计算油菜植被指数,通过相关性分析和单因素方差分析获取优选植被指数,并分别基于单一及组合优选后的植被指数对健康和染病样本进行建模判别分析。结果显示,优选得到的DVI、TVI、RVSI、RDVI、CARI及OSAVI等指数与油菜健康与染病状态相关性较好,可对其进一步建模分析。针对单一和组合优选植被指数的判别结果显示,组合建模条件下的分类效果优于单一植被指数。(3)从叶片尺度获取高光谱图像数据,对健康和染病油菜样本进行检测研究。分别采用MAS、SG、MSC、Detrending及SNV对获取到的叶片高光谱数据进行预处理,得到叶片病斑区域的光谱反射值。其后,分别基于全波段和特征波长信息建立PLS-DA、SVM、ELM、KNN模型,进行健康和染病样本的判别分析。在基于全波段信息的分析中,其中采用MSC预处理结合ELM建模方法得到的分类效果最优,建模集和预测集的结果达到100%。进一步地,在基于特征波长的分析中,分别采用SPA、2ndDer及GA-PLS进行特征波长提取,并利用提取的特征波长建立PLS-DA、SVM、ELM及KNN模型进行分析。结果显示基于SPA方法所选取特征波长结合ELM模型得到的分类效果最优,建模集和预测集的结果达到100%。(4)从冠层尺度获取热红外图像数据,对健康和染病油菜样本进行早期的识别诊断。提取样本冠层尺度的温度值,并对其进行生理指数的监测。然后利用平均温度和最大温差区分健康和染病油菜,并进行单因素方差分析。结果表明,健康和染病油菜的最大温差差异明显,且随着天数的变化该差值基本保持不变;健康和染病植株的平均温度差值起初无明显变化,但随着天数的变化差值逐渐增大。其单因素方差分析表明,最大温差在油菜染病后第1天即存在显著性差异(P0.01)。进一步地,分析油菜生理指数(气孔导度、光合速率、二氧化碳浓度及蒸腾速率)随染病程度的加重发生的变化,该变化可以直观检测出菌核病对油菜的染病程度,且发现健康油菜的生理指数高于比染病油菜,并对生理指数与温度进行相关性分析。结果显示,光合速率、二氧化碳浓度与蒸腾速率与温度之间存在显著相关性。(5)针对叶片尺度的热红外数据对健康和染病油菜样本进行早期的识别诊断。获取样本中染病叶片健康区域和病斑区域的温度信息。其中热红外图像可以直观的识别出病害侵染过程,并利用像素点的值来判别健康和染病区域的温度差异。采用温度信息中的最大温度、最小温度、平均温度以及最大温差对健康和染病油菜植株进行识别,根据以上温度信息区分染病叶片的健康与染病区域,并对其进行单因素方差分析。结果表明,健康和染病区域的最大温度、最小温度、最大温差以及平均温度都存在较明显的差异,且病斑区域温度高于健康区域。其单因素方差分析表明,最大温差在第1天即存在显著性差异(P0.01),可以实现对油菜菌核病的早期诊断识别。
[Abstract]:Rapeseed is one of the most important oil crops in our country, but its yield and quality tend to decrease greatly because of disease. Therefore, it is of great significance to realize rapid and effective disease detection and make reasonable control measures to ensure the yield and quality of rapeseed. The traditional detection method of disease is usually limited to the microcosmic environment. In order to overcome the defects of the traditional methods, this study takes the rapeseed as the research object and uses the UAV simulation platform carrying hyperspectral imager and thermal infrared imager, respectively, from the canopy layer. The scale and leaf scale were used to discriminate the healthy and infected rapeseed samples. The main contents are as follows: (1) obtaining hyperspectral image data from the canopy scale, detecting healthy and infected rapeseed samples, using mobile averaging (MAS), polynomial convolution smoothing (SG), standard normal variable transformation (SNV), and multivariate analysis. The scattering correction (MSC) and detrending method (De-trending) preprocess the obtained canopy hyperspectral data and get the spectral reflectance of the whole plant samples. Subsequently, partial least squares (PLS-DA), support vector machine (SVM), ELM, and K- nearest neighbor classification algorithm (KNN) model are based on the full band and characteristic wavelengths respectively. Discriminant analysis of healthy and infected samples. In the analysis based on all band information, different preprocessing methods and modeling methods have obvious differences in performance, in which MSC preprocessing combined with ELM modeling method has the best classification effect, and the results of modeling set and prediction set reach 100%., and the continuous projection algorithm (SPA) is used respectively, two The order derivative (2ndDer) and genetic algorithm partial least squares (GA-PLS) are used to extract characteristic wavelengths, and the extracted characteristic wavelengths are used to establish PLS-DA, SVM, ELM and KNN models. The results show that the classification results based on the characteristic wavelengths selected based on SPA and ELM model are the best, and the results of the modeling set and the prediction set are all based on the 100%. (2) based on the model wavelengths. The optimum vegetation index was obtained by correlation analysis and single factor analysis of variance. The results showed that the selected indexes of DVI, TVI, RVSI, RDVI, CARI and OSAVI, and the health and dyeing of rape were obtained. The correlation of the disease status was better, and it could be further modeled and analyzed. According to the discriminant results of the single and combination optimization of the vegetation index, the classification results under the combined modeling condition were better than the single vegetation index. (3) the data of hyperspectral images were obtained from the blade scale, and the samples of healthy and infected oil and vegetables were detected. MAS, SG, MSC, Det were used respectively. Rending and SNV preprocessed the acquired hyperspectral data, and obtained spectral reflectance of the leaf spot area. Then, based on the whole band and characteristic wavelength information, PLS-DA, SVM, ELM, KNN models were established for discriminant analysis of healthy and infected samples. In the analysis based on all band information, MSC preprocessing combined with E. The LM modeling method has the best classification effect, and the results of modeling set and prediction set reach 100%. further. In the analysis of characteristic wavelengths, feature wavelengths are extracted with SPA, 2ndDer and GA-PLS respectively, and the extracted characteristic wavelengths are used to establish PLS-DA, SVM, ELM and KNN models. The results show that the SPA method is selected. The feature wavelength combined with the ELM model is the best. The results of the modeling set and the prediction set reach 100%. (4) to obtain the thermal infrared image data from the canopy scale, the early identification and diagnosis of the healthy and infected rape samples. The temperature values of the sample canopy scale are extracted and the physiological index is monitored. Then the average temperature is used. The difference of the maximum temperature difference between healthy and infected rapeseed was obvious, and the difference of the average temperature between healthy and infected plants remained unchanged, but the difference of average temperature between healthy and infected plants had no obvious change at first, but the difference increased with the change of days. The single factor analysis of variance showed that the maximum temperature difference was significantly different in first days after rape (P0.01). Further, the changes of the physiological index of rape (stomatal conductance, photosynthetic rate, carbon dioxide concentration and transpiration rate) were changed with the aggravation of the disease degree, which can directly detect the degree of disease of sclerotia to rape. The physiological index of healthy rapeseed was higher than that of infected rapeseed, and the correlation between physiological index and temperature was analyzed. The results showed that there was a significant correlation between photosynthesis rate, carbon dioxide concentration and transpiration rate and temperature. (5) early identification and diagnosis of healthy and infected rapeseed samples by thermal infrared data of leaf scale. The temperature information of the healthy area and the spot area of the infected leaves is obtained. The thermal infrared image can identify the infection process of the disease intuitively, and use the value of the pixels to distinguish the temperature difference between the healthy and the infected areas. The results showed that the maximum temperature, the minimum temperature, the maximum temperature difference and the average temperature were significantly different between the healthy and infected areas, and the temperature of the spot area was higher than that of the healthy area. The single factor ANOVA analysis showed that the maximum temperature difference had significant difference on the first day (P0.01), which could realize the early diagnosis and identification of Sclerotinia sclerotiorum.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S435.654;S127

【相似文献】