基于图像分析的小麦籽粒高通量表型系统研究

Study on High-Throughput Phenotyping System of Wheat Grains Based on Image Analysis

小麦籽粒表型信息的准确测量对于小麦基因选择和新品种选育至关重要。现阶段,很多小麦育种者使用手工方法提取小麦籽粒表型信息,效率低下,准确率不高。为解决以上问题,本研究开发了一款基于图像分析的小麦籽粒高通量表型系统。该系统采用Matlab GUI开发,具有籽粒形态分析和颜色分析功能,可以测量小麦籽粒长径、短径、面积、长宽比及提取籽粒的颜色信息;另外还内置卷积神经网络训练和判别功能,可以通过软件训练自己的数据集进行品种判别。通过对7个品种手工测量与系统测量结果的误差分析,本系统测量绝对误差平均值0.1 mm,相对误差平均值3.4%,与手工测量结果一致,并且本系统测量结果受小麦籽粒摆放角度和种脐摆放位置的影响较小,可以大幅提高检测效率。另外,本系统运行配置方便,整个系统软硬件价格低,为广大小麦科研人员提供了一种简单高效的小麦籽粒表型信息提取工具。

Accurate measurement of phenotyping information of wheat grains is essential for wheat gene selection and new variety breeding.Many wheat breeders extract phenotyping information of wheat grains manually at present stage,which is lower in efficiency and accuracy.A high-throughput phenotyping system for wheat grains was developed in this paper based on image analysis to solve these problems.This system developed by Matlab GUI with functions of grain morphology analysis and color analysis,which could measure wheat grain length,width,area,ratio of length to width and extract grain color information.In addition,convolution neural network training and discrimination functions were also built in,which could be used to train model neural network for variety identification.Error analysis results of manual measurement and system measurement of 7 varieties showed that the average absolute error for the system measurement was 0.1 mm,and the average relative error was 3.4%,which consistent with the manual measurement results.The results was little influenced by placement angle and hilum placement position of wheat grains by this system,so that the detection efficiency could be greatly improved.Moreover,this system configured easily,and the software and hardware of entire system cost less.It provided a simple and efficient tool to extract grain phenotyping information for broad wheat scholars.