农作物种子纯度高光谱图像检测技术研究

项目名称： 农作物种子纯度高光谱图像检测技术研究

项目编号： No.61271384

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 黄敏

作者单位： 江南大学

项目金额： 70万元

中文摘要： 种子纯度是种子质量的重要参数，是评定种子等级的主要依据。研究快速、无损的种子纯度检测方法已成为一个亟待解决的课题。项目针对传统检测方法存在的分类特征信息不足，分类器精度难以保证问题，拟将高光谱图像技术引入到种子纯度检测领域，实现种子形态学、结构和化学特征信息的多波段提取；并构建基于相容粒度空间的种子纯度多波段融合识别模型，提高检测精度和识别模型的稳健性；在此基础上，形成基于高光谱图像技术的农作物种子纯度无损检测方法体系。研究内容包括：（1）基于区域主动轮廓模型的种子高光谱图像区域分割；（2）基于小波分析、灰度共生矩阵、信息熵等技术的种子原始特征参数生成；（3）基于Boostrap重采样和流形学习算法的特征选择和融合；（4）基于相容粒度空间模型和支持向量数据描述的种子纯度检测分类模型。本项目研究成果对提高我国种子质量自动化检测水平，促进农业增收和改善我国种子企业的竞争力具有重要意义。

中文关键词： 高光谱图像；图像处理；波段选择；模型更新；种子纯度检测模型

英文摘要： The seed purity is one of the most important quality parameters and the primary reference standard for seed grading and sorting. It is urgent to develop a rapid and nondestructive method for detecting seed purity. Conventional methods for seed purity can not provide comprehensive feature information, and have a low accuracy in classification. This study is focused on the detection of seed purity using hyperspectral imaging to achieve multi-wavelengths extraction including the seed morphological, structural and chemical information. A multi-wavelengths fusion model is also developed to improve the detection accuracy and robustness of seed purity classification models. The main work includes (1) A region active contour model is adopted for image segmentation；(2) Wavelet analysis, gray level concurrence matrix and information entropy techniques are used for the original features generation；(3) Boostrap resample and supervised manifold learning algorithms are developed for feature selection and nonlinear fusion；(4) Support vector data description and tolerance granular space model are investigated to develop the classification models for seed purity.This advanced technique would improve the accuarcy of automatic detection of seed quality in China, which can not only enhance the profitability of agriculture, but also

英文关键词： Hyperspectral imaging technology；image processing；band selection；model updating；seed purity detection model