【作者】 刘飞；

【导师】 何勇；

【作者基本信息】 浙江大学 ， 生物系统工程， 2011， 博士

【摘要】 数字农业和农业物联网技术作为现代农业最前沿的发展领域之一,是当今世界发展农业信息化,实现农业可持续发展的关键和核心技术。数字农业和农业物联网要求快速、实时、准确和定位化的获取植物生长信息,而传统的实验室测量分析和信息获取方法已经不能满足现代农业的发展要求。因此,研究和开发植物生命信息快速无损检测技术和传感仪器等软硬件平台已经成为现代农业亟待解决的关键问题。本研究以种植广泛、经济价值高和作为可再生能源的油菜为研究对象,通过连续4年(2007-2010年)的试验研究,建立了油菜苗期、花期、角果期等全生长期的叶片和冠层养分信息快速检测模型,首次实现了非生物胁迫(除草剂丙酯草醚)下油菜叶片乙酰乳酸合成酶活力、蛋白质含量、氨基酸含量等生理信息的快速检测,并研发了油菜生物胁迫(菌核病)早期诊断识别方法和系统,对油菜种植的精细化管理和作业,提高油菜籽的产量和品质具有重要意义。主要创新性成果有：(1)提出了光谱数据预处理-特征波长提取-线性和非线性建模预测的光谱分析技术路径,建立了油菜苗期、花期、角果期,以及苗-花-角果生命期叶片和冠层养分的快速检测模型,实现了油菜养分信息的精确、高效无损检测。系统地研究了原始光谱(Raw)和7种光谱预处理方法,并应用连续投影算法提取特征波长,建立了线性(多元线性回归MLR和偏最小二乘法PLS)和非线性(最小二乘-支持向量机LS-SVM)预测模型。对油菜苗期、花期、角果期,以及苗-花-角果生命期SPAD值的预测相关系数分别为0.7149、0.9431、0.9215和0.8557。基于Cropscan多光谱辐射信息(15通道)的油菜冠层SPAD值检测最优模型LS-SVM的预测相关系数为0.7122。(2)应用可见/近红外光谱技术,建立了油菜叶片光谱信息与生理信息的定量关系模型,首次实现了非生物胁迫(除草剂丙酯草醚)下油菜叶片乙酰乳酸合成酶活力(ALS)的快速检测。在可见光谱(400-780 nm)、近红外光谱(781-2500 nm)和可见/近红外光谱(400-2500 nm)范围的最优预测相关系数分别为0.9026、0.9179和0.9379。在近红外光谱(1100-2500 nm)范围,应用回归系数(RC)提取的10个特征波长所建最优模型对ALS的预测相关系数为0.9395。(3)提出了连续投影算法(SPA)、回归系数法(RC)、载荷系数法(x-Loading Weights)和独立组分分析法(ICA)提取特征波长的准则,建立了除草剂丙酯草醚胁迫下油菜叶片蛋白质含量的快速检测方法和模型。在可见／近红外光谱(400-2500 nm)范围,油菜鲜叶可溶性蛋白含量、非可溶性蛋白含量和总蛋白含量最优预测相关系数分别为0.9351、0.9067和0.9338。在近红外光谱(1100-2500 nm)范围,油菜干叶可溶性蛋白含量最优模SPA-LS-SVM(Raw)的预测相关系数为0.9887。(4)提出了直接正交信号校正(DOSC)和连续投影算法(SPA)的优化组合次序为DOSC-SPA,建立了除草剂胁迫下油菜叶片17种氨基酸和总氨基酸含量的预测模型,阐明了除草剂丙酯草醚对总氨基酸含量的作用机理和响应规律。所建模型对氨基酸含量的预测相关系数均大于0.95,应用DOSC-SPA提取特征波长所建直接方程模型(幂函数方程)对总氨基酸含量的预测相关系数为0.9968。(5)应用光谱和多光谱成像技术,建立了油菜生物胁迫(菌核病)早期识别的光谱诊断模型、多光谱图像反射特性诊断模型和多光谱图像纹理特征诊断模型,并开发了油菜菌核病早期诊断识别系统。应用DOSC-SPA提取的组合模拟波段、基于多光谱图像提取的植被指数(归一化植被指数NDVI、绿波段归一化植被指数GNDVI和比值植被指数RVI)和纹理特征(惯性矩、同质性、二阶角矩、相关性和熵)所建菌核病诊断模型的识别率均达到了100%,对实际生产具有重要意义。上述研究成果实现了油菜生长全过程、全方位生命信息的精准、高效检测,为开发油菜养分信息、生理信息和生态信息的快速检测仪器和传感器奠定了理论基础,具有广阔的应用前景。更多还原

【Abstract】 Digital agriculture and internet of things (IoT) in agriculture are the most frontier technologies in modern agriculture, and they are also the key and kernel technologies for the development of modern agriculture and the realization of sustainable agriculture. Digital agriculture and IoT in agriculture require the fast, real time, accurate and positional plant growth information acquisition technologies. Obviously, the traditional lab measurements and information acquisition methods cannot meet the demands of modern agricultural development. Therefore, the study on nondestructive detection methods for plant growth information and development of information detection sensors or instruments are the key problems to be dealt of modern agriculture. This study is mainly focused on the Brassica napus L., which is a widely planted, high economic valued and alternative energy resource plant. Based on a four-year experimental study, the SPAD value detection model for oilseed rape leave and canopy were developed for different growing stage, including seedling, blooming, podding stage. The physiological information (acetolactate synthase (ALS) activity, protein content and amino acids) in oilseed rape leaves were firstly measured in a fast and nondestructive way under herbicide propyl 4-(2-(4,6-dimethoxypyrimidin-2-yloxy)benzylamino)benzoate (ZJ0273) stress. A new method and system were also developed for plant disease (Sclerotinia) early diagnosis of oilseed rape leaves. These results were helpful for the precision management and operation of oilseed rape planting and were also meaningful for the improvement of of rapeseed yield and quality. The main creative results were achieved as follows:(1) A spectral analysis technical method was proposed as spectral preprocessing-effective wavelength selection-linear and nonlinear calibration model. The fast, high precision and nondestructive models were developed for SPAD value detection in oilseed rape leave and canopy during different growing stage, including seedling, blooming, podding stage. A complete comparison was performed among raw spectra and different spectral preprocessing methods. Successive projections algorithm (SPA) was applied for effective wavelengths (EWs) selection. Linear multiple linear regression (MLR) and partial least squares (PLS), and nonlinear least squares-support vector machine (LS-SVM) models were developed for the detection of fresh leaf SPAD value. The regression coefficients (Rp) were 0.7149,0.9431,0.9215 and 0.8557 for seedling, blooming, podding and seedling-blooming-podding-stage, respectively. An exploration research was proceeded to utilize multi-channel (15 channels) spectroscopy for the SPAD detection of oilseed rape canopy, and the best LS-SVM model achieved Rp=0.7122.(2) Vis/NIR spectroscopy was applied for the quantitative relationship between spectral information and physiological information in oilseed rape leaves, and this study firstly realized the fast detection of ALS under herbicide (ZJ0273) stress. The correlation coefficients were 0.9026,0.9179 and 0.9379 for visible region (400-2500 nm), near infrared region (781-2500 nm) and Vis/NIR region (400-2500 nm) models, respectively. In near infrared region (1100-2500 nm),10 EWs selected by regression coefficient (RC) were applied for ALS detection, and the best prediction results was Rp=0.9395.(3) The principles for effective variable selection were proposed, inclueding successive projections algorithm (SPA), regression coefficients (RC), x-loading weights (x-LW) and independent component analysis (ICA). Fast detection methods and models were developed for protein content detection under herbicide (ZJ0273) stress. In visible/near infrared region (400-2500 nm), the optimal results Rp for soluble protein content, unsoluble protein content and total protein content were 0.9351,0.9067, and 0.9338, respectively. In near infrared region (1100-2500 nm), the best model for soluble protein content using dried leaf spectra was SPA-LS-SVM (Raw) with Rp=0.9887.(4) The optimal combination order for direct orghogonal signal correction (DOSC) and SPA were obtained with DOSC-SPA. Fast and nondestructive models were developed for 17 amino acids and total amino acid (TAA) in oilseed rape leaves. The functional mechanism and response rules were descript for TAA under herbicide stress. The result indicated that the Rp by optimal model were over 0.95 for all 17 amino acids. The selected effective wavelength by DOSC-SPA were applied for direct function development for TAA detection, and the results indicated that power function obtained the best prediction results with Rp=0.9968.(5) The early diagnosis models were developed for Sclerotinia in oilseed rape using spectral and multi-spectral imaging technology, including spectral diagnosis model, multi-spectral imaging reflectance diagnosis model and multi-spectral imaging texture feature diagnosis model. An early diagnosis system was also developed for Sclerotinia in oilseed rape. An early diagnosis model for Sclerotinia in oilseed rape leaves using combinational-simulated wavelengths by DOSC-SPA was developed with a good discrimination ratio of 100%. A multi-spectral imaging discrimination model with a discrimination ratio of 100% was developed based on vegetation index (normalized difference vegetation index, Green normalized difference vegetation index and ratio vegetation index) and texture features (moment of inertia, homogeneity, second derivative angular moment, correlation and entropy). The results were quite helpful for practical applications.The above results realized the fast and high precision detection of oilseed rape for whole growing stage and whole growth information. They also supplied theoretical basis of detection instruments and sensors for the determination of oilseed rape nutritional information, physiological information and ecological information, which had a promising application prospect.更多还原