小麦分蘖形态学特征X射线-CT无损检测

Non-destructive detection of wheat tiller morphological traits based on X-ray CT technology

针对传统小麦分蘖形态学特征需采用人工接触或有损方法获取,不仅过程繁琐、主观性强而且会影响小麦后续生长。为实现小麦分蘖性状快速、准确、无损测量,该文提出一种基于X射线断层成像的小麦分蘖形态学特征提取方法。首先构建了小麦分蘖X-ray CT断层扫描成像系统,采用滤波反投影(filtered back projection,FBP)断层重建算法和图形处理单元(graphics processing unit,GPU)加速算法快速获取小麦分蘖茎秆断层图像,设计了专门的分蘖图像分析算法实现对小麦分蘖形态学特征参数(分蘖数、分蘖角度、分蘖茎粗和分蘖壁厚)的无损检测。该研究对107株小麦植株测量结果表明:分蘖数测量准确率可达100%,分蘖角度、茎粗和壁厚的平均测量误差分别为3.65%,4.84%和7.86%。该技术相较于人工测量方法和石蜡切片方法,能够对小麦分蘖形态学特征进行较为精准无损检测,实现单株测量效率约200 s,对于小麦功能基因组和抗倒伏能力品种的筛选具有重要的研究意义。

Wheat tillers play an important role for nutrition transport to support the wheat growth. The wheat stem diameter and thickness are closely related to lodging resistance. Meanwhile tiller number and tiller angle directly determine the plant type of wheat. Therefore, the morphological trait extraction of wheat tillers is very important to the study of wheat genetics research, breeding improvement and functional genes location. With the development of the wheat cultivation and genetic breeding, the fast and accurate measurement of morphological traits for wheat tillers is imperative. However, the traditional method for tiller trait measurement is still manual, which is destructive and time consuming. Although a lot of efforts had been made to extract the tillers traits generally based on visible light, it is not able to acquire the inner information of wheat tiller and is affected seriously by tillers overlap. To solve the problem, a nondestructive technology for wheat tillers measurement was proposed and equipped with X-ray CT imaging device. In this study, the X-ray CT imaging system was constructed with the Micro-focus X-ray source and flat detector, which was used to obtain the sinogram images of wheat tiller with the spatial resolution 61 μm by 61 μm, and totally 360 images were collected for every one degree rotation for each plant. Then the FBP and GPU algorithms were adopted to reconstruct the tomography image of wheat tillers based on the sinogram images, and the inner information of wheat tiller was visible in the image. Moreover, the specialized image analysis algorithms were designed to analyze the wheat tomography image, in which the algorithms of background subtraction, OTSU segmentation, removing small region, and connected region identification were applied to extract the tiller regions. After that, the wheat tiller morphological traits were extracted by the following methods, the tiller numbers were counted based on the number of connected areas, the stem diameter was computed by the information of area external rectangle, the tiller wall thickness was extracted with the information of area external rectangle and cavity rectangle, and tiller angle was obtained by the triangle relation of tomography images at different heights. Finally this method was evaluated by 107 wheat plants, which belonged to five different wheat varieties. After the wheat plants were measured by the system automatically, the plants were measured by manual method for comparison to evaluate the system measurement accuracy. The experimental results showed that the system measurement accuracy of the tiller number was 100%, the mean absolute percentage error of tiller angle, the stem diameter and the stem wall thickness were 3.65%, 4.84% and 7.86%, respectively and the RMSE for above traits were 2.96, 0.17 mm, 0.12 mm, respectively. The R2 value of tiller angle, the stem diameter and the stem wall thickness were 0.77, 0.91 and 0.87, respectively. The results demonstrated that this method had a good consistency with manual method, and performed a high accuracy for wheat tiller morphological trait measurements. In this study, the image acquisition efficiency was about 200 s per plant and the time used for image analysis was about 120 s per plant. Considering the parallel implement of image acquisition and analysis, the system efficiency was about 200 s per plant and was able to measure approximate 432 wheat plants in one day. Compared with manual method, this technology was able to detect the internal information of wheat tiller with high-accuracy and nondestructive. Moreover, it was able to extract novel phenotypic traits, which may contribute to the functional genomics and lodging resistance research of wheat plants. In future, more detailed information of wheat tiller such as vascular bundle, leaf sheath could be analyzed based on the higher resolution X-ray imaging device and more intelligent algorithms.