【作者】 程洪；

【导师】 史智兴；

【作者基本信息】 河北农业大学 ， 计算机应用技术， 2007， 硕士

【摘要】 随着信息技术的发展，图像处理与识别技术在种子资源品质检测方面发挥着越来越重要的作用。本文以计算机图像处理技术为主要技术手段，综合运用数字图像、摄影、光学、植物学、色度学、模式识别等方面的知识，对玉米品种的自动识别问题进行了研究。一、具体研究内容1．玉米种子图像的采集为了提高自动识别的准确性同时尽量减少对图像采集的约束条件，对图像拍摄方法进行了研究。选取本地区常用的郑单958、辽单565、京科25作为被识别对象，对不同条件下拍摄的图像进行了分析对比。2．玉米种子图像预处理背景分割：用几种不同的分割方法进行试验，通过对比验证，采用迭代法进行背景分割。去除噪声：背景分割后图像中存在孤立的噪声点，通过查阅资料，采用自适应中值滤波法。颜色校正：在研究中找到了一种基于标准白板的颜色校正方法。此法根据标准白板的颜色来校正种子图像的颜色，大大提高了图像中种子颜色的准确性。种子分离：主要采用区域标记法来实现种子图像中的标准白板的提取与单粒种子分离。3．玉米种子特征提取除了利用玉米种子的颜色、形状、大小等常见特征外，本文发现玉米种子白色部分(胚部)与黄色部分(冠部)的面积比例以及形状对玉米种子的品种识别具有一定作用，并对黄白面积比例进行了深入研究，提取了白色部分的面积、黄色部分的面积、黄色部分的颜色等新特征。颜色特征的提取采用RGB模型与HSV模型相结合的方法。4．特征集优化与新特征的价值验证本文采用基于遗传算法与支持向量机特征优化方法，把新特征与常规特征有区别的进行优化。从两组最优特征子集的识别率与算法运行过程中的特征选择率两方面证实了部分新特征是很有价值。对特征数据进行分析，进一步证实白色部分与黄色部分的面积比例特征的可用性；通过试验验证，黄色部分的蓝色分量与饱和度对识别的价值要优于整个玉米种子的相应特征。5．玉米品种识别采用支持向量机方法对玉米品种进行识别。三个玉米品种的识别率分别为90.1％、93.5％、96％。二、本文的独特之处1．提出了一种简单实用的颜色校正方法，使得种子图像的拍摄条件无需固定照明，把玉米品种识别摆脱试验室限制的进程向前推进了一步。2．提出了玉米种子白色部分与黄色部分的面积比例等特征作为识别玉米品种的新特征。更多还原

【Abstract】 Along with the development of information technique, digital image processing and identifying technology becomes a vital one in inspecting the quality of seed by computer. The main purpose of this paper is to deeply research the corn species identification mainly by computer image processing technology. As well this paper involves in other knowledges such as photograph, optics, plant physiology, chroma, pattern recognition etc.First, the main contents of the paper are as follows:1. The image of corn seed collectionThe method of shooting image is researched on to advance the identification accuracy by computer and reduce restrictive conditions when collect image. Zhengdan958, Liaodan565, Jingke25 which are in common use in this region are selected as the researched objects. This paper analyzes their images which are obtained under different conditions.2. The image of corn seed pre-processParttioning Background: This paper decides to adopt iterative method through starting experiment by a few partition methods and comparing the experiment results.Smoothing image: After parttioning background, there are same isolated noise points in the image. By consulting datum, self-adjusing median filter is adopted to smoothing image.Correcting color: A method is found out in the research that correct image color according to the color of the standard whiteboard. Using the method to correct color improves the accuracy of corn seed color in the image.Seed separation: Standard whiteboard and separate seeds are distilled Mainly by labeling area method.3. The characteristic of corn seed pick-upIn addition to a few general characteristics about color, shape and size. This paper finds out the ratio of white part(embryo) and yellow part(crown) of corn seed and their shapes are useful. Then, by deeply researching in this aspect, new characteristics such as the size of the two parts and the color of the yellow part are picked up. RGB model and HSV model are combined together during picking up the color characteristics.4. The character of corn seed optimization and the value of new characteristics validationThis paper uses support vector machine (SVM) combined with genetic algorithm(GA) to discriminatively optimize new characteristics and general characteristics to validate the value of the new characteristic. The recognition rates of the two best character and the select rates of characteristics approve that some new characteristics are very useful. Through analyzing data of the ratio of white part and yellow part, the value of the characteristic are further confirmed. By experimentalizing, the color characteristics of the yellow area are better than the whole corn seed’s to the species identification.5. The corn species identificationUsing Support Vector Machine to identify the breed of corn seed, the separate rates are 90.1%, 93.5%, 96%.Second, the innovative aspects of this paperImages of corn seeds are obtained out of fixed illuminate room and a simple and useful method of color correction is put forward which accelerates the identification process to get away from laboratory. And that some characteristics which are found out by partitioning the corn seed to white part and yellow part are valuable to the identification.更多还原