基于高光谱成像的小麦穗发芽特性快速检测与评价方法研究

The fast and non-destructive detection of the pre-harvest sprouting of wheat is very important for the screening of breeding materials. Currently, pre-harvest sprouting is measured by human inspection or α-amylase detection method after whole spike sprouting experiment or seed germination experiment. These methods were destructive, laborious and time-consuming. Therefore they were mainly used for high generation detection, and were difficult for early generation screening. More over, human inspection could not discover the early germination of wheat. In this project, the wheat seeds after threshing were considered as measuring object, hyperspectral imaging coupled with rotating reflection mirror was applied to acquire the hyperspectral images of wheat seeds. The hyperspectral image response during the dynamic process of imbibition, sprout and gemmate was studied. By using image processing and spectroscopy analysis method, a serial characteristic image parameters and spectral parameters were extracted, which could reflect the outside configuration and inside biochemical components of seeds. The image and spectra information were used to construct a sprouting extent prediction model by a data fusion method of fuzzy comprehensive evaluation method. Then a serial of wheat varieties that had known pre-harvest sprouting resistance were used as control materials. The pre-harvest sprouting resistance of samples to be measured was identified through comparison. Finally, a fast and non-destructive detection and eveluation method for the pre-harvest sprouting resistance of wheat was built, which could provide base for the large scale screening of early generation breeding materials.

小麦穗发芽特性的快速无损检测对于育种材料的筛选具有重要意义，目前主要通过整穗发芽、籽粒发芽实验后目测得出发芽率，或测定α-淀粉酶活性来鉴定，这些方法都耗时耗力、具有破坏性，主要用于高代品系检测，难以用于大批量早代材料筛选，且目测难以观察早期发芽较轻的种子。本项目以脱粒后的小麦籽粒为测量对象，采用反射镜旋转扫描高光谱成像技术获取种子的可见/近红外光谱图像，探明小麦种子吸胀、萌动、发芽等一系列动态变化过程中的高光谱图像响应规律，综合应用图像处理和光谱分析，筛选出反映种子外部形态、内部生化成分动态阶段性变化的特征图像指标和特征光谱指标，应用模糊综合评判等方法建立融合光谱、图像信息的穗发芽程度预测模型。然后引入穗发芽抗性明确的小麦品种作为对照材料，通过比较待测材料与对照材料的穗发芽程度，实现穗发芽抗性的快速检测。最终构建一套小麦穗发芽抗性快速无损检测与评价方法，为大规模育种早代材料的筛选奠定基础。

本项目围绕育种过程中小麦穗发芽特性快速检测、筛选的需求，采用高光谱成像技术，对小麦籽粒、麦穗进行检测研究，并与机器视觉成像技术进行对比分析。主要开展了6方面研究，取得进展及结果如下：（1）通过预实验表明基于高光谱成像特征波段提取的RGB合成彩色效果图要优于基于机器视觉的麦穗发芽成像图。发芽麦穗与未发芽麦穗的光谱反射率在675nm处存在明显差异。（2）通过对发芽0h、12h、24h、48h的小麦种子高光谱图像响应研究，发现不同品种的小麦种子，由于种子吸水及穗发芽抗性的差异，其光谱特征也存在差异。（3）在对比研究中，提出了一种基于机器视觉形态信息的小麦穗发芽程度检测方法，该方法通过获取小麦胚芽鞘的图像，结合相应的形态学分析，得到胚芽鞘的长度，结合种子发芽度分级标准，实现对小麦穗发芽程度的定量检测。（4）对小麦穗发芽特性中品种因素进行了识别研究，首先提取了小麦籽粒的12个形态特征，提取了胚、胚乳等不同区域的光谱特征，然后融合形态特征和光谱特征建立了分类识别模型，结果表明采用PLSDA和LSSVM建立的信息融合分类模型识别率较单一信息模型有所提升。（5）对小麦穗发芽特性中存储年份因素进行研究，通过获取2007-2012年六个年份单粒小麦种子的近红外高光谱图像，发现相邻年份之间的二分类识别率能达到97.44%，六个年份籽粒混合分类结果虽受过渡年份相似性影响但识别率也能达到82.50%，从一定程度上说明了高光谱数据能反映小麦籽粒储藏过程的差异性变化。（6）在前述特征提取、影响因素分析的基础上，本项目探索建立了小麦穗发芽程度预测模型，以发芽率作为测量指标。首先获取了360个单粒小麦种子的高光谱成像数据，利用LSSVM对单粒种子发芽与否进行定性预测，发芽与未发芽的平均识别率达到72.5%。然后获取了145份发芽率在0%-100%的群体种子光谱，采用PLS对发芽率进行定量预测，模型预测集相关系数为0.63。上述研究结果表明高光谱图像与小麦穗发芽之间存在一定相关性，为后续开展育种材料初筛提供了参考。项目执行过程中共发表论文7篇，其中SCI收录2篇，EI收录3篇；获得授权发明专利2项，超额完成了考核指标。