基于Micro-CT的单籽粒小麦内部害虫早期检测机理及方法研究

For the detection of young larvae and eggs inside single grain cereals and the accurate criterion of the internal pest infection, wheat kernels infested by Sitophilus zeamais (Motschulsky) and Rhyzopertha dominica (Fabricius) are used as the research object. The two species of insects do most harm to wheat seriously. During the period from eggs metamorphosis into adult gradually, and the relationship among the main component content such as protein and water，density of wheat kernels and the image features of micro-computed tomography are studied. The characterization intensity and the distribution characteristics on the micro-computed tomography images are analyzed for the healthy kernels and the infested kernels. The action mechanism of changes of main chemical composition content inside the infested kernels to micro-computed tomography imaging will be explored. Studying the position effect of wheat kernels imaging, and obtaining wheat kernels clear projection images of the optimal parameter combination, then it will be optimized and analyzed. The three-dimensional cone beam reconstruction model of wheat kernels will be constructed based on the filtered back projection algorithm. The microscopic effect of early infection on wheat kernels is analyzed based on three-dimensional visual structural model. The insect damage law of the different insect instars is studied. The variation law of the micro-computed tomography image features will be explored to the infested wheat kernels, and then the expression of the three-dimensional image feature are determined. For the three-dimensional gray scale images and binary images of grain kernels, the basic statistical features, histogram transform features and morphological features of wheat kernels image are extracted. The optimal image feature space will be set up for the classification of infested kernels and healthy kernels. The intelligent classification model is established for wheat kernels. The difficult problem that the young larvae and eggs inside single wheat kernels could not be detected non-destructively and accurately will be overcome.

针对单籽粒小麦内卵和低龄幼虫的检测以及内部害虫侵染的准确判别，以小麦中危害最为严重的玉米象、谷蠹侵染的粮粒为研究对象，研究害虫由卵逐步变态发育过程中，粮粒内蛋白质、水分等主要成分含量与密度、显微CT图像特征之间的关系，分析完善粒和含虫粮粒在显微CT图像上的表征强度与分布特性，探索侵染粒内主要化学成分含量变化对显微CT成像的作用机理；研究粮粒成像的位置效应，优化分析粮粒清晰投影图像获取的最佳参数组合，构建基于滤波反投影算法的粮粒三维锥束重建模型；建立粮粒立体结构模型，分析早期侵染对粮粒的微观效应，研究不同虫态的虫蚀规律，探索害虫侵染后显微图像特征的变化规律，确定其三维图像特征的表达；针对粮粒的三维灰度图像和二值图像，提取粮粒的基本统计特征、直方图变换特征、形态特征等；构建适于早期侵染粒和完善粒分类的优化特征空间，建立粮粒智能分类模型，以解决制约单籽粒小麦内卵和低龄幼虫不能准确无损检测的难题。

粮食安全关系到国计民生、经济繁荣和社会发展。全国每年因储粮害虫造成的粮食损失达数百万吨，但现有的储粮害虫检测技术无法对谷物的早期虫害侵染进行有效的检测。针对单籽粒小麦内卵和低龄幼虫的检测以及内部害虫侵染的准确判别，以小麦中危害最为严重的玉米象、谷蠹侵染的粮粒为研究对象，研究害虫由卵、低龄幼虫、高龄幼虫、蛹到成虫逐步变态发育过程中，粮粒内蛋白质、水分等主要成分含量与密度、显微CT图像特征之间的关系，分析了完善粒和含虫粮粒在显微CT图像上的表征强度与分布特性，探明了侵染粒内主要蛋白质、淀粉、水分等化学成分含量变化对显微CT成像的作用机理；深入分析粮粒长轴朝向和旋转台面之间的夹角与粮粒厚度范围、CT值区间、成像清晰度等之间的关系，确定了粮粒成像的位置效应；研究了阳极电压、输出功率、放大倍数、视角数等参数对粮粒显微投影图像的影响，确定了获取稳定清晰粮粒投影图像的最佳参数组合；构建了基于z-FDK的滤波反投影算法的粮粒三维锥束重建模型，提高了算法的适应能力、伪影校正效果和重建图像的质量；利用基于移动立方体法面绘制与光线投射法体绘制的方法实现了麦粒三维可视化，同时用Mimics对不同龄期的原始数据进行绘制，从重建的切片图像和三维可视化图像两方面分析早期侵染对粮粒的微观效应，探究了不同龄期米象及被侵染麦粒的变化规律；针对粮粒的三维灰度图像和二值图像，提取了不同龄期下虫体的8个二维特征、4个三维特征、7个不变矩特征以及7个纹理特征，构成26维原始特征空间。利用ABC算法优选出适应度为92.43的11个特征，利用KFDA压缩算法的前5个特征进行分类，构建了适于早期侵染粒和完善粒分类的优化特征空间，建立了基于CABC-SVM的粮粒智能分类模型，识别率达100%。解决了制约单籽粒小麦内卵和低龄幼虫不能准确无损检测的难题，为谷物的早期虫害侵染无损检测提供了全新、快速、有效的技术。