| 董晶晶,吴静珠,刘倩,刘翠玲,毛文华,张银桥.小麦不完善粒的高光谱图像检测方法研究[J].电子测量与仪器学报,2017,31(7):1074-1080 |
| 小麦不完善粒的高光谱图像检测方法研究 |
| Research on hyperspectral image detection method of wheat unsound kernel |
| |
| DOI：10.13382/j.jemi.2017.07.013 |
| 中文关键词: 高光谱成像 光谱特征 图像特征 不完善粒 支持向量机 |
| 英文关键词:hyperspectral image spectral feature image feature unsound kernel support vector machine (SVM) |
| 基金项目:国家国际科技合作专项（2014DFA31660）、土壤植物机器系统技术国家重点实验室开放课题（2014 SKL 05）、河北省科技计划（16272916）资助项目 |
|
|
| Author | Institution |
| Dong Jingjing | Beijing Key Laboratory of Big Data Technology for Food Safety, School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China |
| Wu Jingzhu | Beijing Key Laboratory of Big Data Technology for Food Safety, School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China |
| Liu Qian | Beijing Key Laboratory of Big Data Technology for Food Safety, School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China |
| Liu Cuiling | Beijing Key Laboratory of Big Data Technology for Food Safety, School of Computer and Information Engineering, Beijing Technology and Business University, Beijing 100048, China |
| Mao Wenhua | Chinese Academy of Agricultural Mechanization Sciences, Beijing 100083, China |
| Zhang Yinqiao | Chinese Academy of Agricultural Mechanization Sciences, Beijing 100083, China |
|
| 摘要点击次数: 3184 |
| 全文下载次数: 12093 |
| 中文摘要: |
| 为了实现高光谱图像处理技术对小麦不完善粒的快速准确鉴别，研究了一种基于小麦不完善粒高光谱图像的光谱和图像特征，结合多分类支持向量机的不完善粒的识别方法。实验采集小麦不完善粒的高光谱图像，对图像进行图像增强、阈值分割等处理后，提取7个纹理特征和5个形态特征作为分类器的输入，应用多分类支持向量机分别建立并比较基于光谱特征、基于图像特征以及基于光谱和图像特征组合的不完善粒识别模型的分类精度。基于光谱特征建立的4分类模型总识别率达9473%，黑胚粒与正常粒的识别率分别为100%、98.63%，效果较好，但虫蚀粒与破损粒的识别精度均低于90%；基于图像特征的不完善粒识别率相对较低；融合光谱与图像特征建立的4分类支持向量机模型总识别率达97.89%，其中虫蚀粒识别率从89.79%提高到95.91%，破损粒识别率从84%提高到94%，识别效果最佳。实验结果表明，高光谱成像技术可以快速、无损鉴别单籽粒小麦不完善粒，该技术在小麦种子质量快速、高通量、无损检测领域具有的应用潜力。 |
| 英文摘要: |
| In order to identify wheat unsound kernel by hyperspectral image processing technology quickly and accurately, a detection method was researched based on spectral and image feature of wheat unsound kernel combined with multi classification support vector machine. The hyperspectral images of wheat unsound kernels were collected and processed by image enhancement and threshold segmentation, which were used to extract 7 texture features and 5 morphological features as input to the classifier. The identification accuracy of model was established by multi class support vector machine and then compared with different feature combination (spectral features, image features, spectral and image feature). The total identification rate of the 4 classification models based on spectral features was 94.73%, and the identification rate of black germ kernel and sound kernel was 100% and 98.63% respectively, but the identification rate of insect damaged kernel and broken kernel were less than 90%. The recognition rate of unsound kernel based on image feature was relative lower. And when the spectral and image features were integrated, the recognition rate of the four class support vector machine model was 97.89%, the recognition rate of insect damaged kernel was increased from 89.79% to 95.91%, and the recognition rate of broken kernel was increased from 84% to 94%. The results show that the hyperspectral image can quickly and non destructively identify the unsound kernel of single grain wheat, which has potential application on rapid, high through and non destructive detection of wheat seed quality. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
