基于稀疏表示算法的高通量玉米果穗粒型识别系统

发布时间：2018-04-18 18:26

本文选题：识别 + 机器视觉　；参考：《农业工程学报》2017年S1期

【摘要】：玉米籽粒粒型是评估玉米产量和品质的重要表型参数之一,为了提高籽粒粒型的识别率,同时满足高通量以及无损测量的要求,该文以果穗整体为研究对象,基于稀疏表示的方法构建了高通量玉米果穗籽粒粒型识别系统(果穗未脱粒)。以掉落抓拍法硬件采集平台采集3种不同粒型(硬粒型、马齿型、半马齿型)的玉米穗图像,首先使用帧差法获取果穗轮廓,再通过G通道分离、OTSU算法(最大类间方差法)得到籽粒轮廓信息,提取籽粒部分颜色、形状、纹理特征作为分类依据,每种粒型取200粒作为训练样本构成稀疏表示算法的判别字典,对每一个测试样本计算稀疏表示系数,根据最小重构误差判定籽粒粒型类别。结果表明,该方法不需要传统的果穗脱粒再进行籽粒类型统计,识别正确率达到94.8%,测量速度达到28穗/min,大大提高了玉米粒型统计的效率。
[Abstract]:Grain type is one of the important phenotypic parameters to evaluate the yield and quality of maize. In order to improve the recognition rate of grain type and meet the requirement of high throughput and nondestructive measurement, the whole ear of corn was studied in this paper.Based on sparse representation, a kernel type recognition system for high throughput corn ears was constructed.Three kinds of corn ear images of different grain types (hard grain type, horse tooth type, half horse tooth type) were collected on the hardware acquisition platform of drop capture method. Firstly, the contour of ear was obtained by frame difference method.Then the kernel contour information is obtained by G channel separation algorithm (maximum inter-class variance method), and some color, shape and texture features of grain are extracted as the basis of classification.Each grain type takes 200 grains as the training sample to form the discriminant dictionary of sparse representation algorithm. The sparse representation coefficient is calculated for each test sample and the grain type category is determined according to the minimum reconstruction error.The results showed that the method did not need the traditional kernel threshing and grain type statistics, and the recognition accuracy was 94.8 and the measuring speed was 28 panicles / min, which greatly improved the efficiency of corn kernel type statistics.
【作者单位】：中国农业大学信息与电气工程学院;农业部农业信息获取技术重点实验室;
【基金】：中国农业大学基本科研业务费专项,公益性行业(农业)科研专项“作物育种材料农艺性状信息高通量获取与辅助筛分技术”(No.201203026)
【分类号】：S513;TP391.41

【参考文献】