化学成像与二维相关协同机制下种子冻害特征分析与识别

Seed is the most critical factor of agricultural reproduction, but its vigor is greatly reduced by freezing injury. This project takes main varieties of maize seed in production base of China as the research object. Microscopic spectrometers combined with digital image analysis method are used to observe various organs of maize seed affected by freezing injury. Meanwhile, freezing temperature and time are applied as outside disturbance to know the law of influence. So a convincing scientific explanation about the change of component and structure inside maize seeds can be obtained to clarify the microscopic mechanism of identifying the freezing injury of seeds based on spectrum and image. Moreover, combined with microcosmic, mesoscopic chemical imaging technology and two-dimensional correlation spectroscopy analysis, the issues relate to analyze the large amount of hyperspectral image data are addressed to make the synergistic mechanism clear, which can improve the efficiency and precision of spectral image analysis. Then the revealed theoretical basis on the scale of molecules is used to guide the rapid extraction and accurate recognition of the features. What’s more, the established effective feature extraction and quantitative methods are proposed to investigate the feasibility of nondestructive identification of seed freezing injury based on the optimal bands. And then the diagnosis technology and automatic recognition device of multiband imaging can be developed. So this work pays attention to innovation of basic theory and technology method simultaneously. The research is expected to provide not only a reliable basis for features analysis using synergistic mechanism, but also form a technical support to rapid and accurate inspection of seed freezing injury, which is a key to the improved delivery of high quality seed products.

种子是农业再生产最关键的因素，受冻害影响则其种用价值大打折扣。本项目以我国玉米制种基地主要品种为研究对象，从种子各器官受冻害影响的显微观测出发，施加温度和时间外扰认识变化规律，明确种子冻害特征谱图识别的微观机理，获得其组分结构变化令人信服的科学解释；结合微观、介观化学成像技术和二维相关光谱分析，突破高光谱图像数据量大分析困难的瓶颈，明确能提高谱图分析效率和精度的协同机制；基于分子尺度的理论基础实现种子冻害特征的快速提取和准确分辨，利用二维相关分析优选高光谱特征波段，提出种子冻害快速无损检测方法，进而建立诊断技术和多波段成像自动识别装置。本研究针对种子冻害快速鉴别和准确识别难点问题，阐明化学成像与二维相关光谱分析的协同机制，揭示种子冻害特征变化规律与谱图识别机理。研究成果将为快速准确检测种子冻害奠定理论基础，为建立快速无损检测技术及装置提供新的思路和方法，有望对种子质量精细管控形成重要支撑。

本项目围绕种子冻害快速鉴别和准确识别难点问题，以我国河西走廊玉米制种基地主要品种为研究对象，首先针对目前冻害玉米种子评价指标单一、缺少相对全面的冻害程度判定标准的情况，测试不同冻害玉米种子生理生化指标并研究其变化，综合各项指标的聚类结果明确冻害程度类别划分条件，为冻害种子的显微特征分析和检测识别及冻害样本的制备提供依据；接着对受不同程度冻害影响的种子主要部位（种皮和种胚细胞）的微观结构进行表征，明确玉米种子受冻害影响的微观变化规律；在此基础上基于近红外光谱技术和高光谱成像技术，研究群体和单粒玉米种子的冻害识别方法，利用二维相关分析提取光谱特征波长，并与其他特征提取算法获得的特征波长进行融合，为建立基于特征波长的冻害玉米种子快速无损多光谱检测系统提供依据；对比卷积神经网络方法与传统建模算法对冻害玉米种子的识别效果，利用深度学习在数据处理方面的优势，建立了端到端的冻害玉米种子无损识别模型，进而形成玉米种子冻伤诊断技术和多波段成像自动识别装置。通过研究明确了化学成像与二维相关光谱分析的协同机制，揭示了种子冻害特征变化规律与谱图识别机理。在本项目的支持下，课题组共发表了SCI、EI收录的第一标注论文13篇，获得美国PCT国际发明专利1项，授权中国国家发明专利2项，并取得软件著作权登记2项，培养多名研究生和青年科研骨干。项目研究成果为快速准确检测种子冻害奠定了理论基础，形成了建立玉米冻伤种子快速无损检测技术及装置进行种子质量精细管控的关键支撑。