【作者】 李翠英；

【导师】 贺立源；

【作者基本信息】 华中农业大学 ， 资源环境信息工程， 2008， 硕士

【摘要】 目前,烤烟烟叶的分组定级都是通过培训专业人员,手工地凭主观经验进行的。针对传统分级方法的缺点,本文探索了数字图像处理与模式识别相结合进行烤烟烟叶自动分组的方法,提出基于轮廓特征的自动分组技术。此方法在对烤烟烟叶进行边缘提取与拟合后,以拟合多项式的系数及烤烟烟叶长宽比作为分类特征,并采用贝叶斯判别分析法构建烤烟烟叶部位分组模型,从而实现部位组的自动分类。本文主要介绍了研究中图像采集系统的架构及图像读取、显示技术;分析了烤烟烟叶的图像平滑、图像分割、二值化以及图像旋转等预处理算法,采用改进的快速中值滤波法进行图像的平滑去噪,较好地保留了图像的边缘特性,并结合蓝色相对变异系数与小面积去除法更加有效地实现了图像的背景分割;对比了几种边缘提取方法以及曲线拟合技术的优缺点,最终选用了二元二次正交多项式作为边缘拟合的主要方法,并以二元二次正交多项式系数与烤烟烟叶的长宽比建立了标准烟叶的特征值数据库;尝试使用误判率最小的贝叶斯判别分析法构建烤烟烟叶部位组的判别模型。试验结果表明,利用贝叶斯判别分析法建立的烤烟烟叶自动分组模型分类效果较好,并有望推广于烤烟烟叶的品种鉴别应用中。更多还原

【Abstract】 The grading of flue-cured tobacco leaf is still accomplished by manpower at present. Experts are trained to classify the flue-cured tobacco leaf by hand according to their sense and experience.To overcome the flaw of the conventional method of classifying the flue-cured tobacco lear,the study of image processing and pattern identification in the flue-cured tobacco leaf automatic classification is carded out and a contour-based automatic grading technique is introduced.A grading model is established by Bayes analysis through the coefficients of fitting polynomial and length-width ratio after the edge extracting and fitting,then flue-cured tobacco leave’s growing position can be automatically classified at last.This paper introduced the framework of image collecting system and techniques in image reading and displaying.On the basis of analyzing many pretreatment algorithm for flue-cured tobacco leaf,such as image smoothing,segmentation,binarization and rotation, the improved fast median filter was applied to denoising,while the relative variation coefficient method combined with the small area removing method was used in image segmentation.After a comparison among many edge detecting and fitting methods,the two element orthogonal polynomials showed advantages,so it was adopted as the main edge fitting method.And a character database was structured for the coefficients of the two element orthogonal polynomials and also for the length-width ratio of flue-cured tobacco leaves.Lastly,the classification model was established for flue-cured tobacco growing position by using the Bayes discriminant analysis,a least wrong classifying rate method.Experimental results showed that it worked efficiently with the tobacco automatic grading system by using the Bayes discriminant analysis.The practical results also showed that it could be extended into the application of variety identification for flue-cured tobacco leaf.更多还原